Customer Segmentation for Banks and Fintechs: Why Relevance Still Wins

By Yamm Labs · Updated October 2026 · 13 min read

The brief from the cards team names one audience: salaried customers, 25 to 45, metro cities. The CRM holds dozens of fields on each of those customers, from the date salary lands to the last time they opened the app, and the brief uses three of them. The same email will reach a first-jobber in Pune who has never held a credit card and a 44-year-old in Gurugram who carries a balance on two. One needs a starter card with a low limit. The other should not be sold more credit at all. Customer segmentation is how a bank or fintech tells those two people apart before it pays to talk to them.

What is customer segmentation for a bank or fintech?

Customer segmentation for a bank or fintech is the practice of splitting the customer base into groups that share a need, a behaviour or a value to the firm, and then treating each group differently in product, message, channel and service. It is the step between holding data and using it. A segment earns its place when someone can say what that group should receive next, and why.

Financial firms segment on four kinds of signal. Product holding tells you what a customer already has: a salary account, a home loan, a term plan, two SIPs. Behaviour tells you what they do, such as how often they log in, whether salary arrives on the first or the seventh, and whether they pay the full card bill. Life stage covers age, income band, family and job changes. Needs are the hardest to capture. They describe what the customer is trying to do, like saving for a child’s college fees or sending money home from Dubai.

The firm type decides which signals carry weight. A large private bank in India has years of transaction history and often a branch relationship. A two-year-old payments app has rich behavioural data and almost no product holding. An insurer sees a customer at purchase and at renewal, and rarely in between. A Gulf bank with a mostly expatriate customer base may find nationality and salary-transfer status more useful than age. Start from the signals your firm already holds and trusts.

Why customer segmentation in banking matters more for a regulated brand

In a regulated firm, segmentation decides who is shown an offer, and regulators care about who is shown what. A lender that sends a pre-approved personal loan message to every customer with a salary credit has made a credit marketing decision about people it never assessed. An insurer that promotes a unit-linked plan to a segment built on income alone has made a suitability call in a spreadsheet. A segment rule is a business decision, and your compliance team will want to read it.

The data behind a segment is regulated too. India’s Digital Personal Data Protection Act, 2023 says consent must be free, specific, informed, unconditional and unambiguous, and limits processing to the data needed for the stated purpose. The same Act requires a firm to erase personal data when consent is withdrawn or the purpose is no longer served, unless another law requires it to be kept. The DPDP Rules, notified in November 2025, phase in over eighteen months and require a separate, plain consent notice that states the specific purpose. A customer who agreed to receive account alerts has not, by that act, agreed to be profiled for cross-sell. Your segment model needs to know the difference.

Lenders carry an extra layer. RBI’s digital lending rules expect data collected through lending apps to be need-based and backed by the borrower’s prior, explicit consent. In the UAE, the federal Personal Data Protection Law, in force since 2 January 2022, prohibits processing personal data without the owner’s consent, with some exceptions, and the DIFC has its own data protection law (UAE government portal). Data collected for one purpose and reused for another is the first thing a data protection review will find.

The commercial case is simpler. When every customer gets every campaign, unsubscribe rates climb, customers learn to ignore the SMS sender ID, and the paid media team keeps retargeting people who already hold the product. A tight segment spends less to reach people more likely to say yes. In a fintech with thin margins, that difference shows up in the cost per funded account within a quarter.

SIGNALNewsletter by Yamm Labs

The next piece, straight to your inbox.

New writing on BFSI brands, B2B and influencer marketing, and AI search.

Email *

Subscribe

Subscription form is not available at the moment

One click to unsubscribe. Privacy policy

How customer segmentation works, in five parts

Segmentation that holds up inside a bank or fintech has five parts. Each one has an owner and an artefact you can show an auditor.

1. A consented data layer. One record per customer, built from core banking, card, loan, policy and app data, with a consent flag for each purpose. Marketing reads from this layer and does not copy raw customer data into ad platforms. Our marketing analytics and martech work usually starts here, because every later part depends on it.
2. Segment definitions in plain words. Each segment has a name, an owner and a rule that a non-analyst can read, for example: salary credited in each of the last three months, no investment product held, app login in the last 30 days.
3. Eligibility and exclusion rules. The list of customers a segment may never reach: accounts in collections, people who opted out, minors, customers flagged by risk or fraud teams. Compliance signs this list, and it runs before any campaign logic.
4. A message and channel map. For each segment, the next product, the one message, the channel and a frequency cap. This is where brand voice and disclaimers meet the data. A segment offered a mutual fund SIP gets SEBI’s standard risk warning in the same block as the offer.
5. A test-and-retire loop. Every segment campaign keeps a random holdout. Segments that show no lift after a few cycles are merged or retired.

AI changes the second part more than the others. Clustering models can sort millions of customers by behaviour in hours and surface groups nobody would have written down, such as card customers whose spending shifts to travel every December. The weak point is explanation. If a model places a customer in a segment and that segment receives a loan offer, your compliance reviewer will ask why, and “the model grouped them” is not an answer anyone can file. Use models to propose segments. Then rewrite each one as a readable rule before it drives an offer.

The table compares six common approaches against what a regulated marketer has to defend: the data each needs, the consent load, whether compliance can follow the logic, and how quickly it produces a first campaign. The ratings are our working judgement, not a regulator’s view.

Most firms run two approaches at once: product holding for the core journeys and behavioural signals for timing. Needs-based research is worth doing once, to name the segments in language the brand team can use, even if daily assignment runs on simpler rules.

How to start customer segmentation without a data project that never ends

Start with one product line, one decision and the data you already hold with consent. A first project that tries to segment the whole bank tends to stall inside the data team for a quarter.

1. Pick the decision. “Who gets the credit card upgrade offer this quarter” is a decision. “Understand our customers better” is not.
2. List the fields the decision needs, and check the consent purpose behind each one. Drop any field you cannot tie to a stated purpose.
3. Write between three and six segments as plain rules. Fewer is fine.
4. Write the exclusion list with compliance before anyone designs a campaign.
5. Set a random holdout and agree the success metric with finance before the first send.
6. Run one cycle, read the result and rewrite the rules.

The data decisions come before the segments. We write these into the brief, because compliance and IT ask about them before they approve a customer data platform, a tag manager change or a new analytics tool.

• What is tracked only after consent. Analytics and advertising cookies on the website and in the app fire only after the visitor accepts them. Google’s consent mode passes that choice to Google tags. In the basic setup, tags are blocked until consent is granted. In the advanced setup, tags load first and send cookieless pings when consent is declined, which Google uses for modelling. Pick one setup on purpose and record why.
• How personal data stays out of analytics. Google’s Analytics policy prohibits sending data that Google could recognise as personally identifiable information, such as email addresses and personal mobile numbers. Mobile numbers and customer IDs leak through page URLs, form fields, site search terms and custom events, so audit all four. Join analytics data to customer records inside your own systems, using an internal or hashed ID.
• Retention settings. In Google Analytics 4, user-level and event data can be kept for 2 or 14 months on a standard property, with longer event-data options only on Analytics 360, and the setting affects explorations and funnel reports, not standard aggregated reports (Google Analytics Help). Choose the period the purpose needs and write down the reason. Apply the same logic to the CRM, since the DPDP Act expects erasure once the purpose is served.
• Where the data sits. Name every system that holds segment data, who can export from it, and whether any vendor processes it outside India. Lenders should check storage expectations under RBI’s rules with their compliance team.

Before approving a segmentation tool, compliance and IT reviewers tend to ask five things. What personal data does it receive, and under which consent purpose? Where is it stored, and for how long? Who at the vendor can see it? How does a customer’s withdrawal or erasure request reach it? And what happens to the data when the contract ends? Answer them on one page before the vendor demo and approval moves faster. This is general guidance, not legal advice; your compliance team decides.

A segment you cannot explain to your compliance team is a segment you cannot use.

How to measure whether customer segmentation is working

Measure a segment by the lift it produces over the same kind of customers left alone. Open rates and click rates tell you a message was seen. They do not tell you the segment changed anything.

Every segment campaign needs a holdout: a random slice of the segment that receives nothing, or the default message. The gap between the two groups is the effect of the segment. Without it, you are partly measuring customers who would have bought anyway, and every campaign looks like a success.

Incremental conversion = conversion rate of the treated group minus conversion rate of the holdout group

• Incremental conversion, from the formula above. It shows whether the segment changed behaviour.
• Cost per incremental customer: campaign cost divided by the extra customers over the holdout. This is the number a CFO will accept.
• Approval or activation rate inside the segment. For lenders and card issuers, a lead that fails underwriting is a cost, not a result.
• Unsubscribe and complaint rate by segment. A rise is the first sign a segment is receiving messages it does not want.
• Segment stability: the share of customers who stay in the same segment from one month to the next. Large swings mean the rule is reacting to noise.

Take a hypothetical bank that sends an SIP-start offer to 1,00,000 salaried customers with idle balances and holds back 10,000 of them at random. If 1.5% of the 90,000 treated customers start an SIP and 1.0% of the holdout do, the lift is 0.5 percentage points, or 450 extra SIPs. If the campaign cost ₹4.5 lakh, the cost per incremental SIP is ₹1,000. The lesson sits in the 1.0%. Two-thirds of the 1,350 SIPs would have started without the campaign, and a report with no holdout would have credited the campaign with all of them.

Holdouts are harder in paid media than in owned channels, because platforms do not let you hold back individual customers in the same way. Use geography or time-based tests there, and treat platform-reported conversions as a ceiling. Our guide to marketing measurement for financial brands covers those methods in more detail.

What goes wrong with segmentation, and the early signs

Most segmentation failures show up weeks before anyone calls them failures. The signal is usually in a meeting or a report that nobody connects to the segment model.

The last row needs the fastest response. If your analytics property already holds phone numbers from a form that passed them in the URL, fix the form first, then use the tool’s deletion features, and tell your data protection officer what happened and when.

When customer segmentation is the wrong choice

Segmentation is the wrong tool when the customer base is too small or the data too thin to support different treatment. A fintech with a few thousand users and one product gets more from fixing onboarding than from splitting a small base into smaller groups.

Some messages should never be segmented. Regulatory notices, changes to interest rates or charges, policy terms, renewal reminders and service alerts go to everyone they apply to, in the same words. RBI requires lenders to give borrowers a Key Facts Statement in a standard format, including the annual percentage rate, before a retail or MSME loan is signed. For commercial banks this rule now sits in RBI’s November 2025 Responsible Business Conduct Directions; other lenders should check the Directions that apply to them. That obligation applies to every borrower, whatever segment they sit in.

Be careful with segment-based pricing and offers that could look like unfair treatment. Showing a better fixed deposit rate to one segment and hiding it from another is a decision for product, legal and compliance together, not for a campaign tool. Insurers face the same question with riders and premium offers, and many of the insurance marketing questions we answer start there.

Brand campaigns aimed at people who are not yet customers also need a different method. CRM segments describe customers you already have. For acquisition, use audience research and platform targeting, and keep the CRM segments for what happens after the account opens.

What to tell your compliance team about a segmentation project

You can paste this into an email and adjust it.

“We plan to group existing customers into a small number of segments using product holding and app behaviour. Each segment is a written rule with a named owner. Every field we use is tied to a consent purpose, and fields without one are excluded. The exclusion list, covering collections, opt-outs, minors and risk flags, runs before any campaign and needs your sign-off.

“Analytics tools receive no names, phone numbers or email addresses, and retention is set to the shortest period the purpose needs. Each campaign keeps a random holdout so we can show the effect. We are asking you to approve the rules and exclusions once, and to review only new message copy in each campaign.” This is general guidance, not legal advice; your compliance team decides.

A fintech that sends this before the first meeting usually gets a list of edits back. A fintech that sends a deck about personalisation usually gets questions.

People also ask

What are the main types of customer segmentation in banking?

Banks usually segment by demographics, product holding, behaviour, needs and customer value. Most run product holding and behaviour together, because those two signals are reliable, already in the bank’s systems and easy for compliance to read as rules.

How is AI used in customer segmentation for banks and fintechs?

AI clustering sorts large customer bases by behaviour and finds groups that manual rules miss. Before a cluster drives an offer, rewrite it as a readable rule, so your compliance team can see why a customer is in it and check that every input has a consent purpose.

Do Indian banks need consent to use customer data for segmentation?

The DPDP Act ties consent-based processing to a specified purpose and limits it to the data that purpose needs. A customer who agreed to service messages has not necessarily agreed to marketing profiling. The DPDP Rules phase in over eighteen months from November 2025, so check with your compliance team how they apply to your segments.

How many customer segments should a bank start with?

Three to six segments for one product decision is a workable start. Add segments only when each new one has a distinct next action and enough customers to keep a holdout. Twenty segments that all receive similar messages add cost without adding relevance.