Ask a liabilities team why current account acquisition is below target and the answer is usually about data — the lists are stale, the numbers do not connect, the vendor over-promised. Sometimes that is true. More often the data was adequate and the programme lost the business somewhere between a lead arriving and a branch acting on it.
That gap is an operating model problem, not a procurement problem. This article is about the model: the five stages a business banking lead passes through, what fails at each, and which metric tells you where your own losses actually are. Each stage has a guide of its own on this site; this is the map that connects them.
What this page covers, and what it does not
This is the programme-level view. If your question is specifically where the leads come from — which registers cover which businesses, and what the coverage gaps are — that is a different question, answered in where new business current account leads come from. Start there if sourcing is your constraint; start here if you already have data and it is not converting.
Why bank acquisition is not general B2B lead generation
Three structural differences change the model, and importing a generic playbook fails on all three.
The product has a compliance gate. A software sale ends at a signature. A current account ends at KYC, and the documents required differ by constitution of business — a proprietorship, a partnership and a private limited company present entirely different files. A lead that cannot clear the gate was never a lead, and qualifying for it late is the most expensive way to find out.
Delivery is geographic. Most business banking relationships still anchor to a branch. That makes a lead's value dependent on which branch owns it, which is a routing problem that pure-digital funnels do not have.
The timing is unusually tight. A newly registered business needs an account almost immediately, and generally opens one. The bank that arrives first usually keeps the relationship, so the window is short and the cost of arriving second is close to total.
The five stages
| Stage | What it does | The usual failure |
|---|---|---|
| 1. Source | Identify businesses that exist and are in scope | Register coverage gaps — MCA misses most new businesses |
| 2. Qualify | Remove what cannot convert before anyone calls | Deferred to the RM, so qualification happens on the phone |
| 3. Route | Assign each lead to one owning branch | Two branches call the same business, or nobody does |
| 4. Contact | Reach the right person, compliantly, in the window | Data lag consumes the window before the first dial |
| 5. Convert and extend | Open, fund, and sequence what follows | The account opens and nothing follows |
Programmes are usually instrumented at stage 1 and stage 5 — leads bought, accounts opened — with nothing in between. The middle three are where the losses concentrate, and they are invisible without deliberate measurement.
Stage 1: Source
The sourcing question for Indian business banking is mostly a question of which register you are working from. MCA incorporation data covers companies and LLPs in depth and misses proprietorships and partnerships entirely, which is the majority of new business formation. The GST register is the widest business register available and reaches the segment MCA cannot see.
Neither is complete. The coverage boundary, and what each register can and cannot prove about a business, is worked through in where new business current account leads come from and, on the register mechanics, in GST data vs MCA company data.
The practical guidance at programme level is narrower: decide which segment you are actually resourced to serve before buying coverage of all of them. A branch network that cannot service micro-proprietorships profitably does not benefit from a feed that is mostly micro-proprietorships.
Stage 2: Qualify
Qualification that happens on a call is qualification you paid an RM's hour for. Most of it can be done on the file, before distribution, using fields that are already present:
- Constitution of business — determines the KYC file, the decision speed and the likely balance. It is the single most predictive field in the record and it is available before anyone dials. The document requirements by entity type are in current account KYC by entity type.
- Taxpayer type — TDS deductors, input service distributors, casual and non-resident registrations are registrations, not prospects. Removing them costs nothing and removes rows that were never going to convert.
- Status — cancelled and suspended registrations accumulate continuously. A status check dated four months ago is not a status check.
- Nature of business activity — decides whether your proposition is relevant at all, and it is more reliable than the industry and turnover fields vendors advertise around it, for reasons set out in filtering GST leads by industry.
- Registration age — a business at day 5 and one at day 150 are different conversations. Treating them as one list makes both worse.
A qualification pass on these five fields typically removes a meaningful share of a raw file, and every row it removes is an RM hour returned to the rows that can convert.
Stage 3: Route
This is the stage most programmes have never explicitly designed, and it is where duplicated effort lives. Without a routing rule, leads in overlapping catchments get called twice and leads in the gaps get called by nobody.
The complication specific to this data is that the address in a registration record is the declared principal place of business, which is frequently a residence, a chartered accountant's office or a virtual office provider. Naive pincode routing inherits all of that. The address-cluster problem and a workable allocation sequence are in branch catchment lead routing; the parsing issue underneath it — there is no city field in the register, only an address string — is in city-wise GST leads.
Two rules do most of the work: one pincode belongs to exactly one branch, and any address carrying an unusual number of unrelated registrations is flagged before distribution rather than discovered by an RM in a lift.
Stage 4: Contact
The window is real but it is not the number vendors quote. What consumes it is usually not the sales process — it is the lag between the registration becoming effective and the record reaching the CRM. A programme working from a feed with a three-week median lag is not competing for the window; it is arriving after it has closed, and no amount of dialler discipline recovers that.
Measure your own lag before optimising anything downstream of it. The measurement is two numbers computed from any delivered file, described in what "daily" and "fresh" actually mean, and the window itself is examined in the timing window in current account acquisition.
Two further contact-stage realities are worth designing around rather than discovering:
The person who answers is often the accountant. In a meaningful share of registrations, the contact details on file belong to the chartered accountant who filed the application, because they receive the OTPs. This is not a data error; it is what the underlying record is. A script that assumes the owner is answering misfires often, and a variant for the consultant case is cheap to write. Where these numbers come from is covered in new GST mobile number data.
The call is regulated separately from the data. Holding a lawfully obtained number does not make a marketing call lawful. TRAI's framework governs sender registration, numbering series and DND scrubbing; the DPDP Act governs the data. Both obligations sit with the bank, not the vendor. The checklists are in TRAI DND and cold calling rules and the DPDP Act B2B data checklist. Neither is legal advice.
Stage 5: Convert and extend
An opened account is not the outcome; a funded, transacting account with a widening product relationship is. The current account is an anchor product, and what follows it — and in what order — is the difference between a cost of acquisition that pays back and one that does not. The sequencing is in what follows the current account.
The failure to design for here is bundling everything at onboarding. A business that has existed for three weeks cannot use most of what a bank sells, and pitching all of it at account opening tends to lose the products that would have landed later.
What to measure
Instrument the transitions, not just the endpoints. A minimum set that locates a leak within one cycle:
- Median and 90th-percentile lag from registration date to CRM arrival. If this is large, fix it before anything else — every later metric is downstream of it.
- Qualification survival rate — share of sourced records that pass the file-level filters. Falling survival means the sourcing spec has drifted.
- Routing coverage — share of qualified leads assigned to exactly one branch. Anything not equal to one is duplication or abandonment.
- Contact rate and who answered — owner, accountant, office, dead. This distribution is more actionable than any vendor accuracy figure.
- First-contact-to-application and application-to-funded, held separately. They fail for entirely different reasons.
- Cost per funded account, not cost per lead. Lead-level pricing comparisons across vendors are not comparable anyway — see GST leads pricing in India.
If you can only add one, add the first. Data lag is the metric that most often explains a result that everyone else is attributing to sales performance.
What to look for in a data partner
Judged at programme level rather than on a feature list, four things matter and none is a percentage:
- A dated record. The registration date on every row is what makes lag measurable and window logic possible. Without it, stage 4 cannot be instrumented at all.
- Field-level provenance. Which register each column came from, stated plainly. Derived columns are legitimate; undisclosed derived columns are not.
- Fill rate and correctness quoted separately. A single blended accuracy figure conceals which of the two is weak — the anatomy of these claims is in reading a GST data accuracy claim.
- A stated deduplication key. GSTIN-level and PAN-level counts differ substantially, and a multi-state business is one prospect regardless of how many rows it occupies.
Providers approach this differently. FinScreener — built by the team that publishes this site, see our disclosure — works from new GST registrations with geographic and date filtering. Probe42 and Tofler work from MCA filings, which is deeper on companies and silent on the unincorporated segment. Evaluate them against the segment your branches can actually serve, using the sample method in the B2B data quality checklist.
Common questions
Is buying leads or generating them inbound better for business banking? They solve different problems. Inbound captures businesses already looking for a bank, which is a smaller and later-stage population. Registration-based outbound reaches businesses in the window before they have chosen, which is where the timing advantage exists. Most programmes that work run both and do not compare them on the same metric.
How many leads does a branch actually need? Fewer than most programmes distribute. A qualified list an RM can work in the window beats a larger list they cannot. Distribution volume that exceeds capacity produces stale leads and the appearance of poor data quality.
Can this be automated end to end? Stages 1 to 3 can be automated almost entirely — sourcing, file-level qualification and routing are rules over fields. Stage 4 is partly automatable and fully regulated. Stage 5 is a relationship. Automating the first three is where the return is, and it is the part most often left manual.
Why does the same data work for one bank and not another? Usually segment fit rather than data quality. A feed weighted toward micro-proprietorships performs for a bank resourced to serve them and fails for one whose account proposition assumes a larger balance. Check the fit before changing vendors.