GTech

What Is Next Best Action? The AI Deciding the Best Next Move in Banking

OZGUR SARIGUL — Tue, 11 Aug 2026 02:01:55 +0000

What Is Next Best Action? The AI Deciding the Best Next Move in Banking

Next Best Action (NBA) is a decisioning approach that picks, in real time, the single most valuable move to make for each customer right now. Present a loan offer? Send an alert? Do nothing at all? NBA answers that with learning models, not static rule tables. And it’s no accident it became one of banking’s most-discussed analytics topics in 2026. Here’s the telling part: in one industry study, NBA came out as the single most-wanted AI capability among CDP buyers, ahead of content generation and even segmentation.

This piece covers what NBA is, where it parts ways with classic campaign management, and the decisioning architecture underneath it.

What Is Next Best Action?

NBA is a strategy that fuses customer data, business rules and AI to determine the most fitting action for each individual. The “action” doesn’t have to be a sale. Reminding someone about an investment opportunity, suggesting they review a card limit, or making no offer at all are all valid actions.

That’s exactly where it splits from traditional campaign management. The old way: define a segment, blast everyone in it the same offer, measure weeks later. NBA cares about the individual instead of the segment, and the moment instead of the calendar. The second a customer opens the app, it fires whatever action matters most in that context.

An example makes it concrete. Recommending a card with foreign-spend perks to a frequent traveler makes sense. But if that same customer has called support three times in three months, the right next action isn’t an offer, it’s routing them to a fix. NBA creates value to the degree it can tell those two apart. If it can’t, it’s just a faster spam engine.

How Does NBA Differ From a Classic Recommender?

A recommender usually chases one question: “Which product will this customer buy?” NBA asks a wider one: “What’s the most valuable thing we can do for this customer, on this channel, at this moment?”

The difference comes from folding in channel and timing too. The same offer that flops over email can convert as an in-app nudge. NBA solves “what to recommend” and “where and when to send it” at once.

There’s another layer. A well-built NBA weighs the long-term effect of an action. An aggressive offer that earns a click today can annoy the customer and erode their value tomorrow. Getting that balance right is what separates NBA from a plain scoring engine.

The Decisioning Architecture Behind NBA

A mature NBA program rests on three layers that feed one another, not a single model. They run in sequence, each looking at a different question.

Layer 1: Propensity and uplift models

The first layer does the baseline scoring. Propensity models say “how inclined is this customer toward this action?” Uplift models go further and measure the actual effect: was the customer converting anyway, or did we genuinely make a difference?

This distinction is critical. Sending an offer to someone who would have converted regardless is burning budget. The real gain sits with the group that only acts because of the intervention.

Layer 2: Contextual bandits for real-time decisions

The second layer is a contextual bandit rooted in reinforcement learning. The system sees the customer’s context (behavior, device, timing), picks an action, and soon gets a reward signal: a click, a conversion, an app open.

The bandit’s power is in managing the explore-exploit tradeoff. It keeps tuning between repeating the known high-reward action and trying new options. So the system never locks onto one “winning” offer; it updates live as fresh signals arrive. These are common in tech companies’ recommendation systems today, and increasingly in enterprise marketing.

Layer 3: Agentic AI and compositional reasoning

The third layer is the newest. Foundation-model agents move past selecting a single action to chaining several steps. In a wealth scenario, an AI copilot handles the portfolio review and drafts the next-best-action suggestion, freeing the advisor to focus on the client’s real goals.

You have to think of the three together. A system on propensity scores alone stays static; one leaning only on the bandit drifts from long-term strategy. The value is in the layers feeding each other.

How Does NBA Create Value in Banking?

In Türkiye most banks already use AI across fraud detection, virtual assistants and more. Personalization sits at the center of it. AI-driven systems can offer budgeting advice from spending habits and sense a credit need weeks before the customer notices.

The gains organizations see from NBA don’t come from a single campaign; they come from a continuously learning decisioning layer that gets a little sharper with every new reward signal.

But success hinges largely on data. Real-time decisions need customer signals streaming instantly, models that are observable and retrainable, and a cleanly captured action-reward loop. The real bottleneck in most NBA projects isn’t model accuracy; it’s whether the data feeding the decision shows up on time and reliably.

Where Should You Start?

The move to NBA doesn’t happen in one step. The healthiest path starts narrow: one channel, one measurable action, a clear reward definition. Once the propensity and uplift layer is solid, you add the real-time contextual bandit layer; agentic structures come online as the infrastructure matures.

At GTech, our Symphony Sense and Analytics solutions help banks build this decisioning infrastructure end to end. The right next action only takes on meaning once it sits on the right data foundation.

Sources

CDP.com: Next Best Action
BCG: The Science Behind Next-Best Action Programs
Braze: Contextual Bandits
Backbase: AI in banking 2026

Prepared by: GTech Data Science Team

The post What Is Next Best Action? The AI Deciding the Best Next Move in Banking appeared first on GTech.

Customer Lifetime Value (CLV): Formula & Predictive Models

OZGUR SARIGUL — Mon, 10 Aug 2026 17:54:51 +0000

Customer lifetime value (CLV) is the total profit a business can expect from one customer across the entire relationship, factoring in how much they spend, how often they buy, and how long they stay. It tells you the ceiling on what you can spend to acquire a customer and still profit.

That’s the textbook version, and it’s also where most articles on this topic stop. The formula is simple enough to teach in five minutes. Applying it to a real customer base, especially in banking or fintech, is where it gets interesting.

What Is Customer Lifetime Value?

CLV answers a question that quarterly revenue can’t: is this customer actually worth what it cost to win them? A customer who buys once for $200 looks identical to a customer who’ll spend $50 a month for four years, right up until you calculate CLV. Then the second one is worth roughly twelve times more.

This is why CLV sits next to customer acquisition cost (CAC) in almost every growth conversation. CAC tells you what you spent. CLV tells you what you got back. Neither number means much without the other.

The Basic CLV Formula

The standard version of the formula has three inputs:

CLV = Average Purchase Value × Purchase Frequency × Customer Lifespan

Here’s what that looks like with real numbers:

Input	Value
Average purchase value	$80
Purchase frequency (per year)	4
Customer lifespan (years)	5
Customer lifetime value	$1,600

Multiply average purchase value by frequency and you get annual customer value, $320 in this case. Multiply that by lifespan and you get CLV. It’s the same logic e-commerce, SaaS, and retail businesses have used for years, and it’s a fine starting point.

Where the Basic Formula Falls Apart

The formula assumes every customer behaves like an average customer. Real customers don’t. Someone who buys once a year for a decade and someone who buys weekly for six months can produce the same average purchase frequency while representing completely different value and risk.

It also treats customer lifespan as a known number, when it’s actually the hardest part to estimate. And it ignores the time value of money entirely. A dollar of profit five years from now isn’t worth the same as a dollar today, but the basic formula doesn’t discount future cash flows, so it tends to overstate CLV for long relationships.

For a bank or fintech, there’s a bigger problem. A checking account customer doesn’t have one purchase frequency. They have deposits, withdrawals, a mortgage inquiry, a credit card application, and a support call, all in different rhythms, sometimes years apart. The basic formula was built for single-product, roughly regular purchase patterns. Banking relationships are neither.

Predictive CLV Models: A More Accurate Approach

This is the part that separates a marketing 101 explanation from something a data team can actually put into production.

BG/NBD and Gamma-Gamma

The Beta-Geometric/Negative Binomial Distribution model, usually shortened to BG/NBD, was introduced by Peter Fader, Bruce Hardie, and Ka Lok Lee in 2005 as an easier-to-implement alternative to an older model called Pareto/NBD. Instead of assuming a fixed purchase frequency, it treats each customer’s buying pattern and dropout probability as randomly distributed across the customer base, then estimates the odds that a given customer is still “alive” and how many more transactions to expect from them.

BG/NBD only predicts transaction counts. A companion model called Gamma-Gamma estimates how much each of those future transactions is likely to be worth. Put the two together and you get a CLV estimate built on actual purchase history instead of a single company-wide average.

Both models were designed for non-contractual relationships, where there’s no subscription telling you when a customer has left. That happens to describe most retail banking relationships fairly well: a customer can simply stop transacting without ever formally closing anything.

Where Machine Learning Fits In

BG/NBD and Gamma-Gamma work well with limited data and are easy to explain to a non-technical stakeholder, which matters more than people admit. Machine learning models, gradient boosting and neural network approaches in particular, can outperform them when you have enough historical data and want to fold in variables the probabilistic models can’t handle directly: channel, product mix, credit score, branch location, campaign response history.

The tradeoff is the usual one. More variables and more flexibility, less interpretability. A churn model that flags a customer as high-risk is only useful if someone can explain why, especially in a regulated industry. This is part of a broader shift toward predictive analytics for financial services, where machine learning increasingly complements or replaces static, rule-based scoring.

CLV in Banking and Fintech Looks Different

Most CLV content is written for e-commerce and SaaS, where one customer generally has one product and one purchase pattern. Financial services customers rarely work that way. A single retail banking customer might hold a savings account, a mortgage, and a credit card, each with its own margin, risk profile, and churn behavior.

That’s part of why banks tend to build CLV into broader customer analytics rather than treating it as a standalone metric. GTech’s Symphony Analytics platform, for example, scores customers for attrition risk and clusters them by behavior rather than balance alone, which gets closer to a true lifetime value picture than a single average-purchase formula ever could. Bank of Kigali’s recent data backbone rollout is a useful real-world example: unifying credit, customer, and deposit data first, then building segmentation and personalization on top of it.

None of this requires abandoning CLV as a concept. It just means the “customer” in the formula is really a bundle of products, and the model needs to reflect that.

CLV vs. CAC: The Ratio That Actually Matters

A commonly cited rule of thumb is a 3:1 ratio between CLV and CAC, meaning a customer should be worth at least three times what it cost to acquire them. Below that, growth becomes expensive to sustain. Well above it, you’re probably underspending on acquisition and leaving growth on the table.

The ratio is more useful than either number alone. A CLV of $10,000 sounds great until you learn the CAC was $8,000.

How to Increase Customer Lifetime Value

Retention is the highest-leverage lever here, and the economics back that up more specifically than the industry-wide stat that gets thrown around usually suggests. The original research behind it, Reichheld and Sasser’s 1990 Harvard Business Review study, found that reducing the customer defection rate by just 5% raised profits by 85% in one bank’s branch network specifically, not as a universal law across every industry (Reichheld & Sasser, HBR 1990). For a bank thinking about CLV, that’s a more relevant number than the inflated “25-95% for everyone” version that circulates elsewhere.

Beyond retention, three levers move CLV in practice:

Cross-sell and next-best-action. Recommending the right product to the right customer at the right time, based on actual behavior rather than a generic segment, raises purchase frequency without raising acquisition cost.
Early churn detection. Catching an at-risk customer three months before they leave gives retention teams time to act. Catching them the week they close the account doesn’t.
Segment-specific personalization. Treating a high-frequency, low-margin customer the same as a low-frequency, high-margin one wastes marketing spend on both.

Frequently Asked Questions

What is a good customer lifetime value?

There’s no universal number. What matters is the CLV to CAC ratio, and a ratio below 3:1 usually signals a growth model that’s more expensive than it looks.

How is CLV different from LTV?

They’re the same metric. LTV is just the more common shorthand in SaaS and subscription businesses, while CLV shows up more in retail and financial services writing.

Can CLV predict churn?

Not directly, but the two are closely linked. Predictive CLV models like BG/NBD produce a “probability of being active” for each customer as a byproduct, which is effectively a churn signal.

How often should CLV be recalculated?

For businesses with fast-changing purchase behavior, quarterly is reasonable. For relationships that move slowly, like most banking products, annually is usually enough, with churn scoring running more frequently in between.

Prepared by: GTech Data Science Team

The post Customer Lifetime Value (CLV): Formula & Predictive Models appeared first on GTech.

Silent Attrition: The Hidden Cost of Banking Churn

OZGUR SARIGUL — Mon, 10 Aug 2026 15:15:02 +0000

A customer can walk away from a bank almost entirely without ever closing an account. The paycheck moves to a different bank, savings drift into a fintech app, card spending shifts to a competitor’s card, and the old account stays open with a few dollars sitting in it. In banking, this is called silent attrition, and most of the churn numbers banks report never see it.

Because it doesn’t match the textbook definition of “the customer closed their account,” silent attrition doesn’t show up as a loss in standard reporting. But the effect on revenue is real: balances shrink, transaction volume drops, and cross-sell opportunities disappear. This piece looks at what silent attrition actually is, why banks miss it, which signals expose it, and how those signals can be turned into a working scoring practice.

What Is Silent Attrition?

Explicit churn is a clean event: the customer closes the account, cancels the card, signs the form. It’s easy to measure because it’s a single, dated transaction that shows up in the system immediately and rolls straight into the monthly churn number.

Silent attrition works differently. The customer never formally ends the relationship; they just shrink it. The paycheck moves to another bank, but the old account stays open to hold a small balance. Investment products drift to a fintech app. Card spending shifts to a competitor’s card. What’s left is an account that looks active in the system but is economically hollowed out. Picture a customer who once held a paycheck account, a credit card, and a small investment account, and who over a couple of years ends up with nothing but a low-balance checking account left behind; this is an illustrative pattern, not a specific case, but it’s one that shows up repeatedly across industry sources.

The industry sometimes calls this “silent switching,” or simply silent churn; the term used most consistently in English-language sources is silent attrition. The common thread is always the same: the customer still counts as “active” in the system, while the economic relationship with the bank has already thinned out considerably. The account is open. The relationship, in practical terms, is not.

Explicit churn is a single event; silent attrition unfolds over months.

Why Do Banks Miss It?

The root of the problem is measurement. Most banks track churn with a single blended rate: accounts closed during a period divided by total accounts. The standard formula is: (customers lost during the period / total customers at the start of the period) x 100. That calculation only counts customers who actually closed an account; a customer whose balance has quietly drained away while the account stays open never enters the denominator.

That blended view also masks cohort-level differences. Younger customer segments evaluate their primary banking relationship far more actively than older cohorts, who tend to hold the aggregate number steady. The result: the dashboard says “everything is fine” while the bank’s most valuable growth segment is quietly slipping away.

The second issue is where the data sits. As long as an account stays open, most CRM and reporting systems keep flagging that customer as “active.” Signals like a declining balance, reduced transaction frequency, or recurring outbound transfers sit in separate tables; unless someone combines and interprets them, no alarm ever fires. The credit team looks at one dashboard, the digital channel team looks at another, branch operations looks at a third; none of them sees the full picture on its own.

The third issue is organizational. By definition, silent attrition belongs to no single department. It never reaches the retention team’s radar because the account never closes, and it’s rarely a sales priority because the balance shrinking isn’t dramatic enough to trigger urgency. That gap between teams is exactly what lets the problem grow unnoticed for years.

What Do the Numbers Say?

Industry data suggests that annual customer attrition in banking runs somewhere between 15% and 25%, with rates climbing to three times that for accounts open less than six months. This figure comes from the US market and only counts accounts that actually closed; if dormant accounts were factored in, the picture would likely look considerably worse.

The impact is easier to grasp with a concrete example: at a US bank with $50 billion in assets, every one-point increase in the attrition rate can translate into roughly $12 million in lost annual net interest income. Scaled down to a mid-sized bank in Turkey, the exact figures would differ, but the mechanism is identical: shrinking balances, falling interest income, lost cross-sell potential.

A single case described by a banking consultant makes the scale of the problem more concrete. At a $10 billion US institution, analysis found that 64% of customers held an account elsewhere in a product category the bank itself offered; more than 30% of that group had three or more deposit relationships at other institutions. Looking at transaction data, 37% of customers were making a recurring transfer (not a payment, an actual transfer) to a competing institution every month. That institution’s reported churn rate sat under 3%; its actual silent attrition was an order of magnitude larger. This is one institution’s own analysis, not a generalizable industry average, but the method it used is repeatable: scan existing product categories for external relationships and recurring outbound transfers.

Which Signals Actually Matter?

Catching silent attrition doesn’t require exotic technology; it requires asking the right questions of transaction data banks already collect. In practice, a handful of signals do most of the work.

Average balance trend. Not a single low month, but a sustained decline across several months, especially when it coincides with a paycheck or recurring deposit disappearing.

Recurring outbound transfers. Not bill payments or purchases, but transfers to the same external account at regular intervals. This behavior is one of the strongest indicators that money is now being held somewhere else.

Narrowing product usage. A customer who once used a credit card, an investment account, and a deposit account, now left with nothing but a dormant checking account.

Declining digital engagement. A drop in mobile app login frequency, or falling below a defined threshold (say, 60-90 days), becomes meaningful especially when it shows up alongside other signals.

Unresolved complaints. A drop in transaction volume shortly after a complaint can be an indirect sign that the complaint was never actually resolved.

Cohort-level breakdown. Younger, digitally-oriented segments showing switching-evaluation behavior at rates well above the blended average.

None of these signals is conclusive on its own. A balance can dip for a month, a customer can make one large one-off purchase, app usage can drop during a vacation. But a customer showing three or four of these signals at once has entered a risk profile that warrants active outreach, even if they haven’t technically met the formal definition of churn yet.

None of the six signals is conclusive alone; they need to be read together.

Churn Risk and Churn Rate Are Not the Same Thing

It’s worth clearing up a terminology mix-up here. Churn rate is a backward-looking metric: it reports the percentage of customers lost during a defined period. Churn risk is a forward-looking prediction: a probability score for whether a given customer is likely to churn in the future. This is exactly where the silent attrition discussion matters most: a bank that only tracks the backward-looking rate will never see a customer who hasn’t closed their account yet but is already at high risk. Moving risk earlier means turning the behavioral signals above into a score.

Is Graph Analytics Necessary?

Some industry commentary argues that catching silent attrition requires moving to graph analytics: modeling customer, product, channel, and interaction data as a relationship network to surface connections standard tabular analysis can’t see. For complex, multi-institution relationship patterns, that approach adds real value, particularly in commercial banking, where a single corporate client may hold complicated relationships across several banks at once.

But for most of retail banking, it isn’t a prerequisite. Most of the signals listed above can already be detected once existing transaction data is properly modeled in a data warehouse with the right behavioral indicators added, without standing up a separate graph database layer. The critical point is that the data stops sitting in scattered tables and gets consolidated into one consistent layer that’s scored regularly. Graph analytics can be a useful next step for very complex corporate customer networks, but it isn’t the starting point. For most banks, the priority should be consolidating the scattered signals already sitting inside the existing DWH/BI stack.

Turning It Into a Scoring Practice: Four Steps

Detecting signals and turning them into an operational process are two different things. In practice, a four-step approach tends to work.

The first step is data consolidation. Balance, transaction, digital channel, and complaint data need to be visible in one layer, under one consistent customer identity. This is usually the most time-consuming step, but also the most critical one, since nothing downstream works well on scattered data.

The second step is defining behavioral indicators. The six signals above are a starting point; every bank needs to calibrate its own thresholds (say, a 20% balance decline, or 90 days without a login) against its own customer base.

The third step is scoring and prioritization. Signals shouldn’t be evaluated one at a time; customers showing multiple signals simultaneously should move to the top of the priority list.

The fourth step is triggering action. Once a score crosses a defined threshold, that information needs to flow automatically into the CRM or campaign system; otherwise detection stays a report that never turns into an intervention.

Four steps that close the gap between detection and action.

A Short Scenario: What Scoring Looks Like in Practice

To see how the four steps above play out, walk through a fictional example: this is an illustrative scenario, not real customer data.

Take a customer we’ll call John. Three years ago he had a paycheck account, a credit card, and a small investment account. Over the last eight months, three things changed: his paycheck kept landing on schedule, but a fixed amount started moving out to another bank at the start of every month; his investment account balance dropped to zero; and his mobile app logins fell from about ten a month to two.

Taken individually, each of these changes has an innocent explanation: maybe he moved his investments elsewhere, maybe he’s doing more in-branch and less on the app. But taken together (the recurring outbound transfer, the narrowing product mix, the declining digital engagement), the picture points to John gradually shifting his economic relationship to another institution. His account is still open, his card still active; this would never trip the formal definition of churn. But once a scoring model combines these three signals and crosses a threshold, the retention team gets a chance to reach out with a retention offer while there’s still a relationship left to save; once the account actually closes, that chance is gone.

This example shows why signals need to be read together rather than one at a time. It also shows why automation isn’t optional: no analyst can manually scan thousands of customers’ transaction histories looking for this kind of pattern. That’s what a systematic scoring layer is for.

What Does This Mean for Turkish and Finnish Markets?

The rapid growth of digital banking and fintech alternatives in Turkey is raising the risk that deposit and spending relationships fragment. It’s now technically far easier for a customer to keep a paycheck account at their primary bank while shifting savings or investing into a separate app, which raises the odds that silent attrition becomes a more common pattern going forward.

In a market like Finland, where digital banking is far more mature, the picture reads almost in reverse: customers tend to stay with the same bank for decades, but maintaining relationships with multiple digital providers in parallel has also become normal. In both markets the common denominator is the same: an open account is no evidence of a healthy relationship. Unless the measurement methodology is built to catch that, both markets end up with the same blind spot.

What Should Banks Actually Do?

Detection alone isn’t enough; the signal has to turn into action. What seems to work in practice falls into four categories.

First, design loyalty and reward programs around engagement, not just balance. Rewarding a customer purely for “high balance” delays noticing a customer whose balance is already draining away.

Second, treat complaint and feedback channels as an early-warning system for unresolved issues. When a customer complains at a branch or through the call center, that’s usually the first visible moment of a silent attrition process already underway; if the issue isn’t resolved right there, the customer quietly drifts away instead.

Third, move reporting from a blended rate to cohort-level dashboards. Tracking younger, digitally-oriented, and newly-opened-account segments separately surfaces early warnings the blended average hides.

Fourth, shorten the gap between detection and action. Once a customer’s risk score rises, the relevant team needs that information within days, ideally in near-real time or on a weekly cycle; the longer that gap stretches, the smaller the window to intervene.

Closing Thoughts

Silent attrition is an invisible but very real source of revenue loss in banking. The scale of the problem may be larger than most institutions assume, because standard churn metrics are structurally blind to it. The fix isn’t a new category of technology; it’s a data layer that reads the transaction data banks already have with the right signals. GTech Symphony Analytics’ Predictive Churn Management module scores this kind of behavioral risk before an account closes, giving banks time to intervene proactively.

For a bank looking to manage silent attrition, the first move isn’t buying a new tool; it’s mapping where the existing data already sits. Balance, transaction, digital channel, and complaint data are already being collected; what’s usually missing is a shared layer where they talk to each other. Once that layer exists, the signals described above start flowing automatically, and retention teams can finally step in before an account closes rather than after.

Related GTech resource: Symphony Analytics: Predictive Churn Management

Sources

Parloa, 8 bank churn reduction strategies that work in 2026

Bank Director, Patching Your Leaky Bucket: How Banks Can Address Silent Churn (March 19, 2024)

The Financial Brand, How to Uncover Hidden Pain Points that Cripple Customer Retention (November 28, 2025)

Banking Transformed podcast (Evergreen Podcasts), The Silent Attrition Crisis in Banking (May 8, 2026)

PriceWeber, Silent Attrition: 5 Ways Primary Banks Can Combat the Threat (October 21, 2025)

The Financial Brand, Acting on ‘Silent Attrition’ Is Key to Customer Loyalty in Banking (April 24, 2023)

Prepared by: GTech Data Science Team

The post Silent Attrition: The Hidden Cost of Banking Churn appeared first on GTech.

Bank of Kigali Selects GTech’s Symphony Analytics to Build Its Data Backbone for the GenAI Era

Big Data — Fri, 30 Jan 2026 12:41:52 +0000

Building the Data Backbone for GenAI with

Bank of Kigali

GTech is pleased to announce that Bank of Kigali, Rwanda’s largest bank with public shareholding, has chosen GTech technology to accelerate its digital transformation journey.

As financial institutions globally race to adopt Artificial Intelligence, the difference between success and failure often lies in data quality. Recognizing that reliable Generative AI (GenAI) starts with high-quality, unified data, Bank of Kigali initiated this project to establish a “Single Source of Truth.” This initiative builds the essential data backbone that will power the bank’s future GenAI applications and drive sustainable growth.

Empowered by Symphony Analytics

The bank’s transformation is powered by GTech’s Symphony Analytics, a robust platform that creates the modern, enterprise-grade foundation necessary for advanced AI adoption.

Unlike traditional setups that require years of development, Symphony Analytics has accelerated the bank’s roadmap by pairing a pre-built financial model with ELT jobs, ready-made dashboards, and reports. The solution equips Bank of Kigali with Self-Service BI, 200+ KPIs, and 15+ dashboards, delivering consistent, decision-ready insights at scale.

Delivering Strategic Value

The implementation of Symphony Analytics delivers immediate and long-term value across the bank’s operations:

Unified Visibility: Data from multiple systems is now correlated and semantically standardized, providing the bank with clear visibility across credit, customer, and deposit portfolios.
Customer-Centric Innovation: A tailored customer segmentation program analyzes historical behaviors and transactions. This enables actionable segments for CRM and personal lending, allowing the bank to offer highly personalized financial products.
Regulatory Speed: The platform accelerates regulatory-aligned and auditable reporting, significantly reducing the reliance on IT teams for day-to-day data needs.

Secure, Scalable, and Always-On Architecture

In the banking sector, trust is non-negotiable. The architecture provided by GTech is platform-independent, easy to operate, and extensible. To safeguard sensitive information, the system features robust encryption, access controls, data masking, and continuous monitoring.

Services remain always-on with high availability, synchronized standby systems, and targeted RPO/RTOs, protecting the bank’s business continuity and performance.

A Success Story for the Region

To ensure sustained impact, GTech provided comprehensive training, knowledge transfer, and on-the-job enablement, underpinned by defined processes such as SLAs and clear escalation paths.

This successful implementation serves as a scalable reference for adopting advanced banking technology in Africa. By strengthening human capital and advancing data technologies, Bank of Kigali and GTech are setting a new benchmark for data-driven banking in the region.

The post Bank of Kigali Selects GTech’s Symphony Analytics to Build Its Data Backbone for the GenAI Era appeared first on GTech.

GenAI and Sustainable Transformation in the Finance Sector

Big Data — Thu, 22 Jan 2026 08:17:28 +0000

GTech Vision: GenAI and Symphony Transformation in Finance

As we stand on the threshold of 2026, the finance sector is passing through a pivotal turning point where business models are changing radically, going far beyond a mere technological evolution. Artificial intelligence is no longer just a trend; it is now the key to operational efficiency and competitive advantage. İbrahim Öztürk, GTech’s Deputy General Manager of Global Expansion and Marketing, penned a dossier for BThaber, detailing GTech’s GenAI approach, the new capabilities of the Symphony product family, and the “Sustainable Transformation” vision that is shaping the reality of finance today.

Here is the roadmap for transforming data into value in the financial world, as described by İbrahim Öztürk:

A New Era in Financial Transformation: Creating Value from Data

The dizzying speed of the technology world makes the transformation in the finance sector much more distinct now that we are in 2026. Artificial intelligence, data analytics, automation, and cloud technologies are now integral parts of financial operations. Today, the critical question for institutions has evolved from “Should we use AI?” to “How do we transform our current data into value in the fastest and most accurate way with GenAI?”.

Due to its data-intensive structure and high regulatory burden, the finance sector is at the very center of this transformation. GenAI minimizes routine analysis processes, allowing employees to focus their time on strategy development and creative processes, serving as a structural lever that expands workforce capacity.

Sustainable and Measurable Benefit with Symphony

The sustainability of value creation requires establishing a robust data architecture. As GTech, we are leading this transformation with our 25 years of experience. By equipping our flagship Symphony product family with GenAI capabilities, we have prepared a product portfolio that provides measurable benefits for financial institutions.

Our approach is based on taking GenAI beyond a theoretical concept and transforming it into a structure that creates concrete performance improvements in the daily operations of banks. Thanks to the GenAI features integrated into Symphony, banks can detect bottlenecks in processes early and perform risk assessments more consistently.

AI Support in Decision-Making Processes for Executives

The Personal Data Assistant, integrated into GTech’s Symphony Analytics platform, democratizes access to data. Utilizing the capabilities of Large Language Models (LLM), this system prevents executives from getting lost among complex reports.

Executives can receive instant responses by asking questions via a chatbot, just like messaging a teammate; for example, “What is the rate of increase in our customers coming from outside Ankara compared to last month?”. This structure, which transforms into a digital assistant (Symphony Assist) with Teams integrations, lightens the workload of data scientists and accelerates in-house AI projects by up to 10 times.

Fast Integration Requiring No Coding Knowledge: Prompt-to-Service

Symphony Labs goes beyond standard tools with its GenAI & LLM Integration Core. Our “Prompt-to-Service” technology allows business analysts to design workflows using natural language without needing deep coding knowledge. This technology shortens the integration time of development teams by 80%, providing tremendous speed in “Prompt-to-Market” time.

GenAI as Today’s Competitive Advantage

The pilot projects we carried out over the past year have proven the accuracy of our 2026 vision. Operational cost savings and increased data processing capacity show us that GenAI integration has surpassed the vision of being the technology of the future and has turned into today’s most critical competitive advantage.

As GTech, our goal is clear: To ensure that GenAI becomes a sustainable value generation mechanism in the financial world throughout 2026 and to continue guiding the sector across a wide geography.

Catch the Future Today with GTech

As the rules of competition are being rewritten in the finance sector, the GTech Symphony product family takes your institution one step ahead. Contact us to strengthen your data-driven decision-making processes with artificial intelligence and to meet our “Prompt-to-Service” technology.

Note: This article implies insights originally published in BThaber’s December 2025 issue.

The post GenAI and Sustainable Transformation in the Finance Sector appeared first on GTech.

Time to Stop Talking Innovation and Start Living It

Big Data — Mon, 08 Dec 2025 06:48:35 +0000

Many Talk About Innovation, Few Live It

The article “Many Talk About Innovation, Few Truly Live It” by GTech CEO Mine Taşkaya has been published in Platin Magazine.

Innovation is one of today’s most popular concepts. It features heavily in the business world, in presentations, vision documents, investor reports, and nearly every executive speech, and is generally defined as “creating something new.” However, the harsh reality is that very few institutions truly embrace innovation, accept it as a lifestyle, and genuinely implement it. Many companies live within a comfort zone, believing they are innovative simply by talking about it.

Therefore, we must now ask this question: Do institutions truly understand what innovation means and genuinely embrace it, or are they merely comforting themselves by talking about it?

In the corporate world, innovation is actually a process. It encompasses the entire cycle of developing new ideas, products, and services to strengthen the institution and ensure sustainable growth, as well as executing and implementing them. To achieve this, institutions must first question their habits, step out of their comfort zones, and view failure as a learning tool.

Real Innovation is “Structural,” Not Cosmetic

Unfortunately, it is not entirely clear how correctly the concept of innovation is understood within institutions. Sometimes, what companies believe to be innovation is merely cosmetic.

The most distinct feature of real innovation is that it is not superficial; it creates a radical transformation at its core in other words, it is “disruptive.” It requires looking from an unexplored angle, combining the entire system sometimes with a new technology and sometimes with a new business model, and tearing down one’s own established truths to move forward. It is at this point that we see real innovation and its massive impact.

Cosmetic innovation, on the other hand, usually consists of superficial, short-term, and symbolic changes made to a product. While these may appear innovative on the surface, it is impossible to speak of true innovation unless they touch the essence of the system.

Looking at the world, we can see that significant innovations have been realized especially in the field of financial technologies in the last 20 years. While some of these are due to technological advancements, we can clearly see that others were made possible by structuring the business model differently.

For example, the M-Pesa revolution, which redefined the way finance is conducted in Africa, came to life as a structure implemented by telecom operators, enabling access to basic financial services, because banks were late in taking steps regarding payments and digitalization.

One of the sharpest examples of disruptive innovation in Europe was the BNPL (Buy Now, Pay Later) trend. These models embedded payments into the shopping experience, transforming banking into an invisible background function. The consumer now makes the credit decision not at the bank, but on the platform where they shop. This paradigm shift is the real turning point that transformed players like Klarna from European startups into global financial giants.

Cleo AI stands as a disruptive innovation example in the fintech ecosystem as an autonomous AI assistant that completely redefines personal finance management. Going beyond the classic budgeting tools offered by banking apps, it interacts with the user through conversational language; it analyzes spending habits and generates real-time financial advice. Cleo’s most striking aspect is its ability to create behavioral change through gamified tasks that encourage financial discipline and humor-based feedback mechanisms.

In today’s business world, innovation is no longer in the “nice to have” category. Since innovation is directly related to an institution’s competitive power, customer experience, capacity to attract talent, and survival reflex, the chances of survival for institutions without innovation are now being seriously questioned.

Why Do We Talk About It So Much But Cannot Implement It?

It is possible to say that there is a wide gap between institutions’ innovation discourse and their practices. The most fundamental reason for this is that innovation is adopted as a superficial concept, perceived merely as a “trend.” When companies treat this concept not as a strategic transformation tool but as an element of marketing or image management, the discourse may be strong, but the action remains weak.

Another reason is corporate culture being closed to novelty. Real innovation requires encouraging risk-taking, learning from mistakes, and fostering different thoughts; however, this is rarely possible in hierarchical cultures.

It is also indisputable that short-term performance pressure is a significant obstacle to the implementation of innovation within the organization. Top management’s focus on short-term financial results naturally complicates the realization of innovation, the returns of which will emerge in the long term.

The Formula for Building an Innovation Culture

For an innovation culture to be permanent in the system, the management team needs to be able to make bold decisions.

The first of these is stepping out of the comfort zone and changing leadership behavior. The biggest enemy of innovation is the inability to give up on success formulas that were created years ago but have lost their effect over time, and continuing to play the new game with old conditions. Leaders play a crucial role here; leaders should not be the owners of innovation, but its carriers. The leader’s behavior shapes the culture. Therefore, a leadership approach that encourages novelty, rewards risk-taking, and establishes flexible decision-making mechanisms is essential.

The second is recognizing the right to make mistakes. In institutions where making mistakes is punished, no one wants to try anything new; everyone prefers to stay safe. This brings about corporate inertia; failure is not a side effect of innovation, it is its lifeblood.

Adopting an inspiring and encouraging leadership style instead of a controlling one, reducing hierarchical barriers to ensure ideas flow freely, and institutionalizing a rapid experimentation culture are actions that will help the institution become open to innovation.

Innovation does not work in cumbersome structures. Cross-functional teams, simplified processes, decision mechanisms that reduce delays, and a culture of rapid prototyping are the foundations of innovation. Old technologies are as much a hindrance to the development of innovation as clumsiness. Innovation capacity expands significantly with teams that possess strong data literacy and design thinking skills.

Sustainable Innovation: Starting Is Not Enough, Persistence Is Required

Innovation culture is not a campaign; it is a lifestyle. It starts with a few projects, but it does not grow with a single project. For innovation to be sustainable, three muscles must be continuously exercised:

The Strategy Muscle: The direction must be clear, yet remain flexible.
The Technology Muscle: Infrastructure must be continuously updated.
The Culture Muscle: Learning, curiosity, and courage must be supported.

When these muscles are exercised regularly, innovation becomes ingrained in the institution’s DNA. Innovation shifts from being “something done” to “something that is.”

Final Word: It’s Time to Stop Talking About Innovation and Start Living It

In today’s world, companies that distance themselves from innovation have no chance of long-term competition. Institutions that fear change, do not take risks, do not experiment, and do not learn will inevitably fall behind. Therefore, every institution must ask itself this question:

“Am I one of those who talk about innovation, or one of those who truly live it?”

Whatever the answer, the future points to a very clear truth: Those who adopt the innovation culture early will not only survive but also shape the future.

– This article was first published in Platin Magazine on December 1, 2025.

The post Time to Stop Talking Innovation and Start Living It appeared first on GTech.

Digital Revolution in the Insurance Industry: An AI and Big Data Driven Future

Big Data — Wed, 30 Apr 2025 06:57:32 +0000

Digital Revolution in the Insurance Industry: An AI and Big Data-Driven Future

GTech Partner and CTO Özgür Sarıgül evaluated the impact of rapidly evolving technology and the artificial intelligence (AI) revolution on the insurance sector, highlighting the role of AI, machine learning, big data analytics, and automation technologies in driving transformation.

Sarıgül stated:

“The insurance sector is one of the industries that feels the effects of digital transformation the fastest and deepest. AI, machine learning, big data analytics, and automation technologies are fundamentally reshaping core business processes. From risk assessment to customer experience, from policy pricing to claims management, digital solutions are now at the heart of operations.

AI-powered risk analysis and dynamic pricing models make it possible to process customer data in real time and offer personalized, competitive premiums. Unlike traditional methods that rely on fixed criteria, smart algorithms that assess customer behavior, past claims records, and external factors together enable companies to make more accurate and profitable decisions.

Moreover, big data analytics allows for much faster detection of suspicious activities and fraud attempts, resulting in both cost advantages and reduced reputational risks. These technologies not only enhance operational efficiency but also necessitate innovation and dynamism at a strategic scale. Thanks to machine learning and advanced analytics solutions, insurance companies can combine historical data, customer behavior, and environmental factors to predict claims and risks more accurately, paving the way for faster and more effective decisions that increase customer satisfaction.”

How does technology impact B2B communication between insurance companies and their assistance partners?

“Insurance companies must closely follow technological innovations when collaborating with their assistance sector partners. Assistance services, which intervene at critical moments when the insured needs urgent support, represent one of the most important touchpoints between the brand and the customer. Therefore, the quality and efficiency of B2B integrations are critically important.

Automation and AI-supported technologies speed up data flow between assistance companies and insurance firms, minimizing errors. For example, when a customer reports an accident, AI algorithms can instantly integrate with assistance companies’ systems and organize the right service within seconds, without the need for manual processes or phone traffic. As a result, services such as towing, ambulance dispatch, or other emergency assistance can be provided immediately.

Big data analytics further improves these processes by offering deep insights into the source, time, and type of service requests, enabling continuous optimization of operations for both insurance and assistance companies. Thanks to API-based integration solutions, data flow is accelerated, and operational efficiency significantly improves.”

GTech produces important technological solutions for the insurance ecosystem. What are your innovative technological services and solutions for the insurance and service provider ecosystem?

“At GTech, we combine our 25+ years of experience with an innovative vision to deliver data-driven, advanced analytics, and AI-based solutions to the insurance and assistance ecosystem. With our industry-specific products, we not only shorten what traditionally seemed like long and costly data warehouse projects but also strengthen companies’ strategic decision-making processes for the future.

Today, going beyond retrospective reporting and generating concrete future insights has become crucial. At GTech, we develop AI and advanced analytics models that move beyond backward-looking analyses to clarify future roadmaps. This enables our clients to not only examine past data in detail but also anticipate future risks and opportunities, allowing them to take proactive steps.

With our modern infrastructure and expertise, we support a wide range of needs, from real-time tracking of critical metrics like outstanding claims and earned premiums to developing more complex predictive models by integrating various data sources. Our scenario analyses, built using machine learning techniques, allow companies to test different future possibilities and base strategic decisions on solid foundations.

In our projects, we establish an analytical infrastructure that overcomes performance issues and generates value-added insights while embedding a data-driven management culture into companies. This enables executives not only to interpret past performance correctly but also to quickly plan future actions. As a result, a business model that increases competitiveness, anticipates risks, and captures opportunities early is implemented.

Ultimately, GTech’s innovative perspective and deep industry expertise provide companies with the ability to not only analyze the past but also predict the future and gain strategic advantages. By combining our data-driven value creation approach with advanced analytics and AI-based insights, we continue to play an active role in transforming the insurance and assistance sectors.”

– Originally published in Sigorta Partner on March 21, 2025.

The post Digital Revolution in the Insurance Industry: An AI and Big Data Driven Future appeared first on GTech.

Next-Generation Banking: Flexibility, Innovation and Digital Transformation

Big Data — Tue, 11 Mar 2025 06:33:56 +0000

Next-Generation Banking: Flexibility, Innovation, and Digital Transformation

GTech Partner for Global Expansion, Gürhan Taşkaya shared his insights on the transformation of the banking sector and innovative approaches in Business Diplomacy magazine.

As technology advances, consumers’ expectations for contemporary banking experiences are being reshaped day by day

Growing demands for simple, swift, and smooth financing solutions, heightened awareness and the desire to make more informed financial decisions through the analysis of financial data, and the eagerness to transition swiftly and effortlessly among various financial institutions are necessitating a daily redefinition of what banking means. To fulfill these demands, banks and financial institutions need comprehensive banking technologies that go beyond standalone technology investments. These systems must be able to communicate with each other, integrate seamlessly, and support growth through modular solutions.

At GTech, leveraging more than 24 years of expertise, we create technologies that facilitate a modern banking experience. Our solutions help dozens of clients worldwide stay competitive and innovative, empowering them to approach the future with greater confidence and stability. Designed exclusively for the banking industry, our Symphony product family offers essential banking solutions that financial institutions might need, including core banking, consumer finance, digital banking, open banking, regulatory reporting, and banking data warehousing. Also, each product and technology within the family integrates innovative features that provide swift responses to contemporary demands.

Being our flagship core banking solution, Symphony Banking provides a flexible and modular design tailored to address the demands of modern banking. It allows financial institutions to oversee their digital transformation initiatives with greater efficiency and speed. The service-oriented architecture and flexible design streamline banks’ swift product development and integration efforts, offering a cost-effective and adaptable infrastructure. By offering multilingual support, cloud infrastructure compatibility, and swift development capabilities, it enables institutions across various regions to construct a competitive and innovative future while fully adhering to local regulations. In a project with one of our bank clients, we effectively demonstrated the significance of our core banking system’s features by implementing the system within an impressively short timeframe of just six months. Throughout the project, the flexibility, modular design, and extensive integration capabilities of Symphony Banking allowed us to finalize the initiative swiftly and seamlessly.

In a separate project, one of our clients, one of Türkiye’s largest e-commerce companies, implemented remote customer acquisition and lending processes in the first quarter of 2024 by providing their customers with innovative solutions such as “Buy Now, Pay Later” (BNPL). Through automated processes, users can swiftly
determine their limits and easily access payment plans tailored to the product categories in their carts. This smooth experience has substantially boosted customer satisfaction and transaction volumes. The developed ecosystem made a substantial impact on the organization’s profitability and revenue objectives. Supporting this platform are GTech’s cutting-edge technologies, including our consumer finance solution Symphony Lending, our API integration tool Symphony Labs, and our regulatory reporting product Magic Reports. Symphony Labs features a hybrid API architecture (REST and SOAP) that provides robust transaction security, allowing the system to be developed and customized quickly. Meanwhile, Symphony Lending offers flexible support for the entire process from customer acquisition to collections, handling essential functions like automated collections, bank integrations, and dynamic product design. Our regulatory reporting solution, Magic Reports, utilizes an archived data model to ensure historical data is reported accurately and consistently, facilitating retrospective analysis. This holistic framework provides customers with an impeccable financial experience while boosting the competitive edge of financial institutions, thereby reinforcing Symphony Banking’s flexible, secure, and consistently innovation-driven architecture.

The data sharing and innovative partnerships provided by the open banking ecosystem are enabling the delivery of fast, personalized, and transparent financial services in the banking sector. Meanwhile, Symphony Labs offers a holistic platform that ensures this transformation happens securely and flexibly. Designed as an innovative platform to address the open banking and digital transformation requirements of banks and financial institutions, Symphony Labs provides extensive solutions including API management, secure data sharing, flexible authorization, and compliance support. One of the leading innovators in private sector venture banking, our client improves user experience by streamlining online banking, mobile banking, IVR, and call center integrations using Symphony Labs’ systematic and manageable infrastructure.

As data becomes increasingly crucial in the business landscape, its significance is expanding at an accelerated rate within the financial sector. Symphony Analytics, a key component of the Symphony product suite that seamlessly integrates with any banking system, enables financial institutions to aggregate complex financial data on a platform designed for more effective analysis. This facilitates the acquisition of the necessary insights in a significantly more accurate and swift manner. With its four-tier architecture (ODS, Data Warehouse, Datamart, and Reporting Layer), the platform ensures efficient and user-friendly data management. It allows for the thorough examination of essential banking data, including deposits, loans, credit limits, collateral, foreign exchange transactions, and commission revenues, utilizing comprehensive metrics and visual analysis tools. Symphony Analytics supports deep-dive analyses in crucial banking performance areas such as customer orientation, risk management, operational efficiency, and profitability through its Management Panel, Customer Panel, and detailed reporting features. This enhances the speed of decision-making processes and promotes a more competitive and innovative strategy in the development of new products and services.

At GTech, we adopt a comprehensive approach to the swiftly evolving landscape of modern banking, establishing ourselves as a reliable business partner in the digital transformation journeys of financial institutions. We improve organizations’ operational efficiency and innovation capabilities while allowing them to quickly access strategic insights that boost customer satisfaction. Utilizing flexible, secure, and scalable technologies, we enable the banking experiences of tomorrow to become a reality today.

This article was originally published in Business Diplomacy on January 22, 2025.

The post Next-Generation Banking: Flexibility, Innovation and Digital Transformation appeared first on GTech.

The Rise of a Cashless Society: What It Means for the Future of Finance

Big Data — Thu, 20 Feb 2025 13:13:11 +0000

The Rise of the Cashless Society and the Transformation of the Financial Ecosystem

GTech Deputy General Manager İbrahim Öztürk shared his insights on the rise of a cashless society and the transformation of the financial ecosystem in Para magazine.

The goal of a cashless society is frequently discussed. What are the key benefits and potential risks of a cashless society? Who benefits more from cashlessness, governments or businesses?

The vision of a cashless society is becoming increasingly important as digital transformation accelerates. Reducing cash usage offers distinct yet complementary advantages for both governments and businesses. Governments can leverage the transparency of cashless payments to curb the informal economy and track tax revenues more effectively.

The reduction of operational costs, such as printing, distributing, and managing cash, also leads to significant public resource savings. For businesses, cashless payment methods lower transaction costs and enhance data analytics capabilities, enabling them to provide more personalized and faster services to customers. Of course, this transformation comes with risks such as cybersecurity threats, data privacy concerns, and digital accessibility challenges. However, when supported by strong regulations, inclusive education policies, and advanced security technologies, the benefits of a cashless society can far outweigh the risks.

Why should we choose a fintech provider or use fintech platforms instead of a conventional bank?

Traditional banks have long been symbols of stability and trust. However, rapid technological advancements and evolving customer expectations have made flexibility and speed essential, necessitating new approaches. As a result, the rigid structure of traditional banking is increasingly being replaced by fintech-driven solutions.

Fintech companies, free from complex institutional layers, can implement innovations much faster. Customers now expect 24/7 access to financial services via the internet or mobile devices, low fees and commissions, and the ability to apply for and receive loan approvals within minutes.

Fintech platforms rapidly adopt cutting-edge technologies such as artificial intelligence, machine learning, and API management to continuously enhance and personalize the user experience. This results in lower costs, faster product development, and significantly improved customer satisfaction. In a recent project with one of our country’s leading e-commerce platforms, we launched a “Buy Now, Pay Later” model within just a few months of preparation. This solution allowed users to instantly check their credit limits while significantly increasing transaction volumes and customer satisfaction.

The core strength of the fintech world lies in its ability to quickly implement customer-centric and personalized experiences. Moreover, compliance with regulations and data security standards is now firmly established, making fintech solutions more appealing with each passing day.

What advantages do you offer?

At GTech, with over 24 years of experience, we develop technologies that enable a modern banking experience. By providing competitive and innovative solutions to our customers across different regions of the world, we help them build a strong foundation for their future. Our Symphony Product Family, specifically designed for the banking sector, offers a wide range of solutions, from core banking to consumer finance, digital banking to open banking, and regulatory reporting. Each technology within this suite is designed to adapt to the rapidly evolving needs of banks while ensuring full compliance with local regulations.

Our core banking system, Symphony Banking, with its service-oriented architecture and parametric structure, accelerates banks’ product development and integration processes. With multi-language support and cloud infrastructure compatibility, it provides a competitive future for institutions across different geographies. For example, in a project with one of our clients, we successfully implemented the entire core banking system within just six months, demonstrating the efficiency and agility of our solution.

At GTech, we focus on building a comprehensive ecosystem rather than just a single technology. From Symphony Lending, which manages consumer finance, to Symphony Labs, which supports digital banking and open banking processes, and our regulatory reporting and data analytics solutions (Magic Reports, Symphony Analytics), all our products are designed to meet the end-to-end needs of financial institutions. Last year, in a project with one of the leading private sector venture banks, we enabled seamless management of call center, IVR, and mobile banking integrations from a single point, showcasing the efficiency of our approach. By developing fast, transparent, and secure solutions that contribute to the “Cashless Society” vision, we continue to support our clients in achieving sustainable success.

This interview was originally published in the February 16-22 2025 issue of Para magazine.

The post The Rise of a Cashless Society: What It Means for the Future of Finance appeared first on GTech.

Building a Retrieval-Augmented Generation (RAG) System with Oracle 23AI, Cohere, and Flask

Big Data — Thu, 13 Feb 2025 06:45:58 +0000

What is RAG?

Retrieval-Augmented Generation (RAG) is a technique that combines traditional information retrieval with generative models. The idea is to enhance the relevance and accuracy of answers by retrieving context from external sources, such as a database or documents, and then using that information to generate responses.

GTech’s RAG solution retrieves relevant documents from the Oracle 23AI database based on the user’s query. It then uses Cohere’s LLMs to generate detailed answers. The workflow involves embedding both documents and queries into vectors, comparing them, and reranking the results to ensure that the most relevant information is used in the final response.

Key Components

Vector Embeddings with Cohere

At the core of this system are Cohere’s embeddings. These are dense vector representations of text that capture its semantic meaning. Cohere provides an embedding model (like cohere.embed-multilingual-v3.0, which we use in our project) that transforms both documents and queries into vectors. These embeddings are used to measure the similarity between documents and queries.

When processing a PDF document, the text is split into manageable chunks, cleaned, and then converted into embeddings. These embeddings are stored in an Oracle 23AI database using the vector datatype, which is specifically designed to handle such data efficiently. The use of Oracle’s 23AI DB vector datatype ensures high-performance storage, indexing, and querying, making it ideal for similarity searches in large datasets.

Initially, we tried using the Frankfurt region on Oracle Cloud Infrastructure (OCI) to generate embeddings. However, we encountered a limitation: this region does not support the cohere.embed-multilingual-v3.0 model. To resolve this, we switched to the Chicago region, which fully supports the model, allowing us to proceed smoothly with our implementation.

Knowledge Retrieval from Oracle 23AI Database

The next step is to retrieve relevant information from the Oracle 23AI database. To do this, the query vector is compared to stored document vectors using a Dot Product similarity search. This method identifies the top 10 most relevant documents based on vector distance.

The documents are stored in the Oracle database along with both their text and embeddings, enabling efficient querying for related content. This allows the system to pull contextually relevant documents in real-time. Oracle’s vector datatype ensures efficient storage and fast retrieval, even with large sets of document embeddings.

Reranking with Cohere’s Reranking Model

After retrieving a set of relevant documents, we need to rank them based on their relevance to the query. This is done using Cohere’s reranking model, which evaluates how well each document matches the user’s query.

The reranking model sorts the documents by relevance, ensuring that the most relevant ones are selected for generating the final response. The reranking model (rerank-multilingual-v3.0) is crucial for refining the results and improving the accuracy of the generated response.

Answer Generation with Cohere LLM

Once the relevant documents have been retrieved and reranked, the next step is generating the final answer. We use Cohere’s LLMs like command-r-plus for this task. The LLM takes the query and the ranked documents as input and generates a response based on the provided context.

A key feature of this system is that it combines the generative power of the LLM with specific, context-rich information from the retrieved documents. This ensures that the generated answer is not only relevant to the query but also grounded in the relevant documents.

User Interface with Flask

To make the system interactive, we built a simple web interface using Flask. The user submits a question through a web form, and the Flask server processes the query by calling backend functions. These functions retrieve relevant documents, rerank them, and generate an answer.

Flask handles the routing and communication between the frontend (HTML) and the backend, ensuring smooth interaction for the user. When a question is asked, the backend processes the query and returns the answer along with the relevant documents.

How It Works

User submits a question via the web interface.

The system converts the query into an embedding using Cohere.

The query is matched against stored document embeddings in an Oracle 23AI database, which uses the vector datatype.

The most relevant documents are retrieved and reranked using Cohere’s reranking model.

The system generates a response based on the retrieved documents using Cohere’s LLM.

The answer, along with the relevant documents, is returned to the user.

User Query—>Query Embedding (Cohere Model)—>Knowledge Retrieval (Oracle 23AI Database) —> Top 10 Relevant Documents (Dot Product Similarity)—> Reranking Documents (Cohere’s Reranking Model) —> Answer generation (staying in the context by the help of additional info/docs) – LLM—>Return Answer

Demo:

First image is proof that it can give casual responses like a normal chatbot.

Second image shows a response generated using RAG with data feeding from the document we uploaded.

Conclusion

This RAG-based system is a powerful combination of knowledge retrieval and generative AI. By integrating Cohere’s embeddings and reranking models with Oracle 23AI’s vector datatype and Flask, the system can provide contextually accurate answers to user queries, backed by external knowledge sources.

By combining retrieval-based approaches with generative models like Cohere’s LLM, we can handle a wide variety of questions and ensure users receive the most relevant and accurate answers based on the information available in the database.

Prepared by: GTech System and Database Management Team

The post Building a Retrieval-Augmented Generation (RAG) System with Oracle 23AI, Cohere, and Flask appeared first on GTech.