Every contact-data vendor advertises accuracy in the mid-90s. Those claims are not usually lies; they are answers to a question you didn't ask. "95% accuracy" almost always means: of the addresses we returned, 95% passed a deliverability check. A tool can hit that number by returning addresses only when it is confident — and returning nothing for half your list.
The number you actually care about is different: of the specific people I want to contact, what fraction can I reach? This guide defines the metrics properly, gives you a test protocol you can run in an hour during any trial, and sets realistic expectations by segment.
Four different numbers, all called "accuracy"
| Metric | Definition | Who quotes it | Why it matters |
|---|---|---|---|
| Coverage (hit rate) | % of requested people for whom any contact is returned | Rarely quoted | Determines how much of your list is actionable |
| Accuracy | % of returned contacts that are valid | The marketing number | Determines bounce rate and sender reputation |
| Verified-contact rate | Coverage × accuracy | Nobody | The number that decides your workflow |
| Freshness | How recently the record was verified | Occasionally | Predicts decay on job changes |
The arithmetic is unforgiving. A tool with 60% coverage and 95% accuracy gives you a 57% verified-contact rate. A tool with 85% coverage and 85% accuracy gives you 72% — worse on the marketing metric, materially better in practice, at the cost of more bounces. Which tradeoff you want depends on whether your bottleneck is list size or sender reputation.
One more distinction worth insisting on: personal email versus work email versus mobile. Vendors often blend them into one figure. Work emails are the easiest to derive (pattern plus domain) and the fastest to decay when someone changes jobs. Personal emails survive job changes and are much harder to source. Mobile numbers are the weakest category everywhere — treat any coverage claim above roughly 40% for mobiles with scepticism, and check the legal basis for calling them in your jurisdiction.
The 60-minute test protocol
Run this in every trial, on your own roles. It is the single highest-value hour in a sourcing tool evaluation.
Step 1 — build a 50-person gold set (20 min). Pick 50 people you genuinely want to contact for a live role, drawn from your actual market — right geography, right seniority mix, right industries. Do not use the vendor's sample list, and do not use only famous companies; big-tech coverage is uniformly good and tells you nothing about your niche.
Step 2 — run all 50 through the tool (10 min). Record for each: work email returned (y/n), personal email (y/n), mobile (y/n), and any confidence score the tool exposes.
Step 3 — verify without sending (10 min). Use an independent verification service to check syntax, domain, and mailbox acceptance. Verify with something other than the vendor's own validator — self-graded homework is the thing you are trying to avoid.
Step 4 — compute the three numbers (5 min). Coverage = contacts returned ÷ 50. Accuracy = valid ÷ returned. Verified-contact rate = valid ÷ 50. Do this separately for work email, personal email, and mobile.
Step 5 — send a real batch and measure bounces (15 min of work, days of waiting). Verification tools disagree with reality often enough that a live bounce rate is the ground truth. Above 3% is a sender-reputation problem, not just a data problem.
Run the identical gold set through every vendor on your shortlist. Directionally, in the evaluations we track, coverage on US technology roles clusters tightly across the main vendors, and the spread widens sharply on non-US, non-technical, and non-desk populations — which is precisely where the decision should be made if that is your market.
Where coverage predictably collapses
| Segment | Expected difficulty | Why |
|---|---|---|
| US tech, mid-level | Easiest | Dense professional-network data, predictable email patterns, heavy prior scraping |
| Non-US, especially EU | Harder | Lower profile density plus GDPR constraints on collection and retention |
| Non-desk / frontline (nursing, trades, logistics) | Hard | Fewer professional profiles; personal email and mobile are the only routes |
| Very senior (VP and above) | Hard | Gatekeepers, generic role addresses, deliberately low public footprint |
| Recent job changers | Hard | Records lag reality by weeks or months; work email may be dead |
| Small and mid-size employers | Medium | Non-standard email patterns and less pattern data to infer from |
Practical consequence: your niche determines your vendor, and no ranking — ours included — can substitute for testing on your own population. Healthcare and skilled-trades teams in particular should assume the general-purpose tools will underperform and should test specialist sources; see the healthcare segment ranking.
Data decay, and what to do about it
Contact data is perishable. People change jobs, companies migrate email domains, and records age. Widely cited B2B data estimates put annual decay at roughly 20–30% — treat that as directional (it is vendor-published and varies by segment), but plan for it structurally rather than arguing about the exact figure:
Re-verify at use, not at import. An address verified when you scraped it eight months ago is a guess. Tools that verify on retrieval are worth a premium over ones that serve a cached record.
Prefer personal email for long-cycle relationships. If you are building a talent pool you'll re-engage in a year, work addresses will be substantially dead by then.
Track your own bounce rate weekly. It is the earliest signal that a vendor's index is degrading — and the most common trigger for a renewal renegotiation that actually works.
Don't let a sequencer send to unverified addresses. Bounces damage domain reputation, which quietly lowers deliverability for the addresses that are valid. This is the most expensive form of bad data, because the cost lands on your good outreach.
Compliance is part of the evaluation
Contact-data sourcing sits inside privacy law, and "the vendor said it's fine" is not a defence. Nothing here is legal advice, but four questions belong in your evaluation:
What is the lawful basis for the data? Under GDPR, legitimate interest for recruitment outreach is the usual argument; it requires a balancing assessment, not a checkbox.
Can the vendor honour deletion requests end to end? Including copies that have already synced into your ATS.
Does the vendor provide notice to data subjects, or is that your obligation? Get the answer in writing; in most setups it is yours.
What are the rules for the channels you plan to use? Unsolicited email, SMS, and calls are regulated differently by jurisdiction, and mobile outreach carries the most exposure.
What to shortlist
Contact data is bought as a layer rather than a platform, and it is the cheapest layer in the stack — our estimates put it at $0–$4K a year, with usable free tiers. See the pricing benchmark for the per-vendor basis.
ContactOut is the browser-extension default for recruiters working profile by profile. Apollo and Lusha come from the sales side, with strong company data and weaker recruiting-specific coverage. RocketReach and SignalHire compete on credits per dollar. Full-stack sourcing platforms — hireEZ, SeekOut — bundle contact data with search, which is convenient and usually not better on coverage in narrow niches.
Two useful comparisons: ContactOut vs RocketReach and hireEZ vs SeekOut. And if contact data is where your hours go, note that it is a symptom worth checking: teams running autonomous agents generally stop shopping for contact tools, because the outreach layer owns that problem. The overall field is in our 2026 ranking.