Vendor accuracy claims are unfalsifiable at purchase time, which is why they are always high. Everything that actually predicts whether a dataset will work is measurable from a sample, in under two hours, with a spreadsheet and a browser tab.
Run these twelve checks before signing. In our experience most providers fail somewhere between check 3 and check 6, and you will have paid nothing to find out.
Ask for the sample against the exact filter you intend to buy — not a generic sample. A curated demo file tells you about the vendor's marketing, not their data.
Structural checks
1. Fill rate, per column
Count non-empty values in each column and divide by row count.
Registration fields will approach 100% because they come from a structured source; that is not a quality signal, it is arithmetic. The number that matters is fill rate on the contact columns. Phone and email fill varies from roughly 15% to 70% across this market, and it drives your cost per usable record more than the price does.
Record it per column. A vendor quoting one number for the whole file is averaging away the only figure you need.
2. Duplicate rate, on two keys
Deduplicate on GSTIN, then on PAN (GSTIN positions 3–12). The gap between the two counts is your multi-state registration volume — the same business appearing several times.
If you are billed on GSTIN count but sell to businesses, you are paying for duplicates. Decide which key your CRM should use before import, not after your team has called the same firm three times.
3. Status validity
Filter for cancelled, suspended and provisional GSTINs. In a file sold as "new registrations", these should be rare. A meaningful share indicates a stale extract or careless filtering — and each one is a record your team will burn time on.
4. Registration-date distribution
Compute delivery date − registration date for every row. Report median and 90th percentile.
This is the real freshness measurement, and it routinely contradicts the marketing. Two vendors both advertising daily data often differ by weeks. Full method in what "daily" and "fresh" mean.
5. Filter relevance
What proportion of rows actually match the filter you asked for? Off-filter records inflate the count and are not usable, but you will pay for them.
Check for taxpayer types that were never prospects — TDS/TCS deductors, ISD registrations, casual and non-resident taxpayers. Their presence in a sales file is a filtering failure.
Verification checks
6. Independent GSTIN verification
Take ten records at random. Verify each against the official Search Taxpayer tool.
Check that the GSTIN exists, the legal name matches, the status matches and the registration date matches. Twenty minutes of work, and it tells you whether the registration layer is real. Procedure in how to find company details from a GST number.
7. Contact source test
Take five records with a phone number. Search the business name plus city.
If the number you were sold is the top result on a public directory, you now know the enrichment method — and you can assess it properly using how GST contact databases are built. If you cannot find the number anywhere, ask the vendor where it came from. Their willingness to answer is itself a data point.
8. Dial test
Call fifteen numbers. Record: connects and is the right business / connects and is the wrong business / dead or unreachable.
Nothing substitutes for this. Fifteen calls gives you a rough rate — not statistically tight, but enough to distinguish a 70% product from a 30% one, which is the decision you are making. Wrong-business connects are the most damaging category and the one that never appears in a fill-rate statistic.
9. Address quality
Group by exact address string, sort descending by count. Concentrations reveal virtual offices and CA practices where dozens of unrelated businesses share a registered address.
These are legitimate registrations, not errors — but a field team routed to them will waste a day. See state-wise GST registration data.
Governance checks
10. Per-field provenance
Ask for a table: field, source category, collection date, fill rate, known limitation.
Any competent provider can produce this. Very few offer it unprompted. Refusal or vagueness here predicts problems everywhere else, and it is the single most informative request in the whole process.
11. Compliance posture
Three questions, answers in writing:
- What is the lawful basis for the contact data, and where did it come from?
- What is your process when an individual requests correction or deletion?
- What are our contractual usage rights — internal only, client work, resale, retention period?
Buyer-side obligations are in the DPDP checklist for B2B data. A vendor with no answer to question 2 is a vendor whose problem becomes your problem.
12. Cost per usable record
Combine everything above:
usable = rows × fill rate × (1 − bad contact rate)
× filter relevance × (1 − duplicate rate)
cost per usable record = total price ÷ usable
This regularly comes out at 5–7× the invoice rate, and it reorders vendor rankings. Worked example in what GST lead data costs.
Scoring
Run all twelve on every shortlisted vendor and record the numbers in one comparison sheet. The exercise costs one afternoon. It has more effect on the outcome than any negotiation you will have on price, and it converts a sales conversation into a procurement decision.
Re-audit at renewal
Data products degrade quietly. Collection pipelines break, enrichment sources dry up, and a file that was 60% filled in January can be 35% in July with no announcement.
Re-run checks 1, 3, 4 and 8 before every renewal. Renewal is the only moment you have leverage, and arriving with measurements rather than impressions changes the conversation entirely.
Common questions
Run this against every shortlisted vendor — MCA-derived platforms like Probe42 and InstaFinancials, contact platforms like Apollo.io, and GST-registration feeds including FinScreener, which is built by the team that publishes this site (see our disclosure). The checks are provider-agnostic by design.
How big a sample do I need? 500–1,000 rows is plenty for fill rate, duplicates and date distribution. The dial test on 15 is rough by design — you are separating good from bad, not estimating a precise rate.
What if a vendor will not give a sample matching my filter? That is your answer. Every serious provider can generate a filtered sample; unwillingness usually means the filter returns less than the sales conversation implied.
Which check matters most? Number 8, the dial test. It is the only one that measures what your team will actually experience, and it is the one buyers skip most often because it takes an hour of somebody's time.
Should I run this on free samples from several vendors at once? Yes. Comparative numbers across three samples are far more informative than absolute numbers from one, and the whole exercise costs nothing but time.