Every week somebody asks us which data source to buy. The honest answer is that each of the four we run answers one question well, and the useful part is knowing which question you are asking. A people database tells you who exists. A live profile tells you whether that person is still there. A company's own careers page tells you what is happening at the company right now. The activity feed tells you who is awake and reachable this week.
Numbers below come from our own database, covering 28 July to 18 September 2026: 6,320 people sourced across four sending operations, 2,628 invitations sent, 582 accepted, 136 first replies. Where a rate is quoted, it counts only invitations at least seven days old, because acceptances keep arriving for about a week and a fresh batch flatters itself.
A people database answers "who exists"
This is the source that builds a list. In our window, company search plus contact enrichment produced 2,826 people, and every single row carried an employer, a title and a location. That completeness is the product. Nothing else on this page gives it to you.
Those people generated 1,716 mature invitations, 22.0 percent accepted, 2.9 percent replied. That is the baseline everything else gets compared against.
Two things to plan for. First, cost scales with your shortlist rather than your batch. Roughly a quarter of enriched candidates get dropped after enrichment, because the role turns out to be wrong or the person has moved, so budget about 2.5 credits for every person you actually intend to contact. One batch of fourteen people cost us 36 credits. Second, the meter runs out. Our monthly allowance of 4,050 credits hit zero on 17 September, thirteen days before renewal, and every search across every account stopped in the same hour. Plan the month by the shortlist, and know in advance what runs when the credits are gone.
A live profile answers "is this person still there"
A database record is a snapshot of something that was true when it was collected. The copy you send is about the person as they are today, and the gap between those two moments is where the first sentence of a cold message goes wrong.
We check employers against the live profile before writing. Of 409 checks run this quarter, 382 confirmed the record and 27 contradicted it. Around one in fifteen is a small number until you notice which fifteen: the contradicted ones are disproportionately the people we most wanted to write to, because a recent change is exactly the reason we picked them.
It gets sharper on records where the only evidence of the employer was the stored employment history with the role marked current. We pulled four of those and read the live headline. It named the same employer for one of the four. One person was listing consulting and contract work. Two headlines named a title with no company at all, which is the ordinary case and which counts as unconfirmed rather than contradicted.
Two checks cost nothing and close most of these. The first is the person's own recent activity: their posts and reposts name their employer constantly, in sentences like "join my team at". On eight people whose headline stayed silent, one call to their recent posts resolved seven. The eighth resolved in the other direction, with the most recent activity dated 2019, which is its own answer: that person will not see any message, however good the copy is. The second check is the title of their public profile page as it appears in a search result, which the platform assembles from the current position. Take it when it names the company in full.
The rule we run now: an employer from a stored record is a claim, and the live surface promotes it to a fact.
A job board answers "what is happening there right now"
A dated line from a company's own careers page is the strongest opening material we have, because it is quotable, it is timestamped, and it came from them. Invitations built this way accepted at 24.6 percent across 203 mature sends, the highest accept rate of any list-built source we run, with 3.0 percent replying.
The constraint is coverage, and it is worth measuring before you build a quarter's plan on this line. Our collector, described in reading hiring signals from job boards, reads the six common applicant systems. In continental Europe, out of 55 untouched target companies with active cross-border hiring, a readable board existed at 8. That is about 15 percent, against roughly half on the same check in the United States. Large employers of any nationality sit on enterprise systems that render nothing without a browser. Follow it through to the end and the yield lands near 4 usable leads per 100 companies reviewed, so this is a line you plan in hundreds of companies rather than thousands of rows.
One more distinction that saved us real credits. The "this company posted a job in the last 30 days" field inside a people database is a fine coarse filter for narrowing a search. As the sentence you quote in a message, it needs support: on seven consecutive checks, we went to the company's own pages to find the posting and came back with either an empty board or roles published three and five months earlier. Aggregator counters are counts, and a count carries no publication date. So we use the field to shrink the search and take the quotable line from the company itself. When the company has no readable board, we write from the person's own material instead, and on one batch of nine that substitution produced usable material on 17 profiles out of 18.
The activity feed answers "who will actually answer"
The strongest result in this whole dataset comes from the cheapest source. People collected from live engagement with posts and shared lists, meaning those who commented on or reacted to posts our buyer also reads, produced 94 mature invitations with 37.2 percent accepted and 14.9 percent replying. Of the people who accepted, 40 percent went on to reply. Set that against the 22.0 percent and 2.9 percent from list-built sends over the same weeks.
The reason is plain once you see it. Everyone in that pool was on the platform within the last few days, by definition. Deliverability on a social channel is mostly a question of whether the person still opens the app.
The price is qualification. Of 1,887 people collected this way, 1,790 arrived with no employer field, because engagement data carries a name and a profile, not a company row. That is why only 128 of them ever received an invitation and 1,759 are still sitting in the pool. Every one of them needs a separate resolution step before you know whether they match the buyer profile, or whether they sit on the client's do-not-contact list. Volume here is limited by how many relevant posts your market produces in a week, which makes it a first line rather than the whole plan.
What Sales Navigator adds on top
Two things that the others do not carry: finer spotlight filters, including a job-change view that is closer to the platform's own record than a resold field, and the ability to message somebody who has not accepted a connection.
The second one is where we have measurements, and they are about accounting rather than copy. Messages to open profiles are supposed to be included, and whether they are depends on the sending account. On one of our accounts such a message went out and was delivered. On another account, the same day, the platform asked for credits and refused. Worse, the stored open-profile flag drifts: on one account a live re-check contradicted it 19 times out of 19, while on another it held 7 out of 7. So the working rule is to re-read that flag live at the moment you send, and to check whether that specific account has been refused before. Spending credits is a decision the account owner makes in advance, and our engine asks rather than assumes.
Order them by price, and let each one hand over
The failure worth avoiding is the single source. On 17 September our credits ran out, the standby proxy for the same data was down, and search stopped for everybody. A different provider inside the same subscription was answering correctly the whole time and nobody had asked it.
So the pattern we now run is a cascade (short definitions of these terms live in our outbound glossary): cheapest and most accurate first, and the next step starts on any of refusal, timeout, exhausted credits, an empty answer, or fewer rows than requested. An empty answer describes the provider that gave it, so the next step asks the same question anyway. Only when every step has spoken do you report honestly, naming what each one said.
One implementation detail from building ours, because it cost us a batch. Asking for people at 23 company domains with a limit of 25 rows returned 25 rows, all of them employees of the first domain in the list. The remaining 22 companies were never queried. If your plan is one person per company, ask one domain per call with a small limit and walk the list. Check what your provider does with a multi-value input before you trust a count.
What to do this week
Take your current list and label each row with the question it answered. If everything traces to one search in one tool, you know your exposure, and you know what happens the day that meter reads zero.
Then pick the two cheap upgrades. Re-read employers on the live profile before you write, because a claim promoted to a fact costs one call and protects the only first message you get. And start one line from live engagement, even at low volume, because the acceptance gap in our data is wide enough to be worth the qualification work.
If you would rather have the cascade, the checks and the sending run as one system, our service levels describe the split. Teams who prefer to drive their own model against the same sourcing tools can do that through the MCP tier and keep the choosing in house.
Questions buyers ask
Which data source is best for B2B outbound lead sourcing?
Each source answers one question. A people database tells you who exists and gives employer, title and location on every row. A live profile tells you whether the person is still there, a company careers page tells you what is happening now, and the activity feed tells you who is reachable this week.
How accurate is employer data in a B2B people database?
We checked employers against the live profile before writing: of 409 checks, 382 confirmed the record and 27 contradicted it. That is around one in fifteen, and the contradicted records are often the people with a recent change, which is why they were picked.
Do job postings work as a reason to write in outbound?
A dated line from a company's own careers page accepted at 24.6 percent across 203 mature invitations, with 3.0 percent replying. Coverage is the limit: in continental Europe a readable board existed at 8 of 55 target companies, so the yield lands near 4 usable leads per 100 companies reviewed.
What happens when my data provider runs out of credits?
Search stops unless a second source is ready. We run a cascade ordered by price and accuracy, where the next step starts on refusal, timeout, exhausted credits, an empty answer or fewer rows than requested. Only when every step has answered do you report, naming what each one said.