Three GST data providers, three headline claims: 99.9% accuracy, 97%+ accuracy, 72–78% accuracy. Read as quality signals, they suggest the first is dramatically better than the third. Read as measurements, they suggest something else — that the third vendor is the only one publishing a number low enough to be plausible for the field they were probably measuring.
Accuracy is not a property of a dataset. It is a property of a specific field, measured against a specific reference, at a specific time. A single percentage on a sales page has silently made all three choices for you, and different vendors make them differently. The spread between 72% and 99.9% is mostly definitional, not qualitative.
Four things "accuracy" can mean
| Definition | What is being measured | Typical achievable range |
|---|---|---|
| Format validity | The GSTIN passes its checksum, the PIN has six digits | Very high — near 100%, and nearly meaningless |
| Status correctness | The GSTIN was active as of a stated date | High, and genuinely useful if the date is recent |
| Field correctness | The phone number reaches the named business, the address is current | Much lower, and the number buyers actually care about |
| Deliverability | Emails do not bounce, calls connect | Lower still, and the only one that predicts pipeline |
A vendor measuring format validity can claim 99.9% honestly. A vendor measuring connect rate on mobile numbers for newly registered businesses cannot, and a vendor claiming it is describing a measurement they did not make. The claim in the 70s is the one most likely to have been measured on something that matters.
The disambiguating question
"Which field is that percentage measured on, against what reference, on what sample, and when?" Four clauses. A provider running real quality measurement answers without hesitation because they run the report internally. A provider that cannot answer has quoted a number from a marketing page, and you now know what it is worth.
Why the denominator is the whole game
Two files, same underlying registrations. File A has a phone number on 100% of rows, and 75% of those numbers connect. File B has a phone number on 35% of rows, and 96% of those connect.
File B's vendor advertises 96% accuracy. File A's advertises 75%. File A delivers more than twice as many usable contacts.
Fill rate and correctness must be quoted separately or the composite figure is uninterpretable. This is the most common way an accuracy claim misleads without containing a false statement — and it is why the B2B data quality checklist treats coverage and correctness as separate audits rather than a single score.
Accuracy decays, so an undated figure is not a figure
Even a correctly measured percentage describes a moment. GST registration data ages in specific, predictable ways:
- Cancellations and suspensions accumulate continuously. A status check from four months ago is not a current status.
- Addresses go stale silently. A business that moves does not necessarily file an amendment, and nothing in the record flags the drift.
- Phone numbers churn, and numbers belonging to consultants change client rosters rather than staying attached to the business.
- The industry declaration was made once, at application, and describes intent rather than current activity — see filtering GST leads by industry.
A vendor that refreshes quarterly and quotes an accuracy figure measured immediately post-refresh is describing the best day of the cycle. Ask when in the refresh cycle the measurement was taken, and what the figure is on the last day before the next refresh. The gap between those two numbers is the more honest description of the product, and it connects directly to the freshness question in what "daily" and "fresh" actually mean.
Volume claims have the same problem
"25 crore GST records" and "1.2 crore businesses covered" are not the same kind of statement, and neither is directly comparable to another vendor's count without knowing:
- Whether the unit is GSTIN or PAN — one business with registrations in six states is six GSTINs
- Whether cancelled and suspended registrations are included in the total
- Whether TDS deductors, input service distributors, casual and non-resident taxpayers are counted as businesses
- Whether historical rows and current rows are summed together
A large number assembled by counting rows rather than businesses is not evidence of coverage. Ask for the count on the definition you care about — usually active, regular taxpayers, deduplicated by PAN — and expect it to be substantially smaller than the headline.
How to run a comparable test across vendors
The point of a sample is not to see whether the data looks reasonable. It is to produce one number, defined by you, that you can compute identically for every vendor on your shortlist.
- Fix the segment first — your states or cities, your business types, your registration-age window. Never accept a sample the vendor selected.
- Fix the metric — for most sales teams, the right one is usable records per thousand delivered: rows that are active, in segment, with a contact that reaches the intended business.
- Request the same sample specification from every vendor. Most offer samples; the ones that will not are answering the question.
- Verify status independently against the GST portal's public search on a random subset. This is free and it is the only reference that settles the status question.
- Dial fifty numbers and record who answers — the business, an accountant, an office, or nothing. This distribution predicts your pipeline better than any vendor figure.
- Compute cost per usable record, not cost per record. Pricing structures differ enough that headline rates are not comparable — the pricing models are broken down in GST leads pricing in India.
- Repeat at renewal. A vendor whose measurement holds up over two cycles is running a pipeline; one whose numbers move sharply is running periodic bulk refreshes.
What a credible provider will tell you
Not a higher percentage — a more specific one. The signals that a quality claim is real are structural rather than numerical:
- Fill rate and correctness quoted separately, per field
- A stated measurement date and reference source
- A field-level provenance table, so you know which register each column came from
- Willingness to be tested on a sample you specify
- A stated deduplication key
- An account of what the data does not cover
A provider offering all six with a modest headline number is a better purchase than one offering 99.9% and none of them.
Most providers in this market publish samples. FinScreener — built by the team that publishes this site, see our disclosure — works from new GST registrations with dated rows. GSTDataProvider, IndiaDatabase and EMarket Zone sell prebuilt segmented databases and advertise samples; Probe42 and Tofler work from MCA filings, where the underlying record is filed by the company itself. Run the same specification against each. The full landscape comparison is in where to buy B2B leads in India.
Common questions
Is a 99.9% accuracy claim a lie? Usually not — it is typically a true statement about format validity or status, presented where a reader will assume it refers to contact data. The problem is the missing denominator, not the arithmetic.
Should I prefer the vendor quoting 72–78%? Not automatically, but a figure in that range is more consistent with having measured something meaningful. Judge it on whether they will define the measurement, then test it yourself.
What accuracy should I actually expect on phone numbers for new registrations? Test it rather than accept a number from anyone, including this page. Businesses registered in the last few weeks have the thinnest web and directory footprint, so contact fill rates are structurally lower for genuinely new records than for aged data — the mechanism is in where GST mobile number data comes from.
Why will some providers not give a sample? Sometimes genuine commercial caution about a file being copied. Often it is that the sample would be measured. Either way you can only buy on the evidence you were given.