Find out which sources actually produce placements

Attribution fails at the point of capture far more often than it fails at the point of analysis. A candidate arrives through a board feed and is tagged correctly. The same candidate is re-uploaded three weeks later by a recruiter working from an email, and now the record says manual entry. A referral is entered as a database rediscovery because that is the default in the dropdown.

The Wave 1 build treats source as a required, structured field written at the moment a record is created, with the channel, the specific source and the campaign held separately. A candidate from a board carries the board and the job advert. A candidate from a referral carries who referred them. A candidate from the extension carries the platform and the recruiter who captured them.

Where a record already exists, the original source is preserved and the new touch is recorded as a re-engagement rather than overwriting the origin. Overwriting origin is the single most common way an attribution report ends up telling an agency that its own database sources everything.

Reporting a placement count per source is useful but shallow. The value sits in where each source loses candidates. A source that produces plenty of submittals and no interviews is producing candidates that look right on paper and do not survive a client screen. A source that produces interviews and no offers may be delivering people whose salary expectations were never realistic.

The report will show the full ladder per source: entered, submitted, interviewed, offered, placed, with the coded disposition mix at each drop-out. Because dispositions are a fixed list rather than free text, the reason a source fails is legible rather than anecdotal.

This also protects a source that looks bad on a headline number. A niche board that sends eight candidates a quarter and places two of them is one of the best sources you have, and a volume-based report would have you cancel it.

Attribution is usually framed as a marketing question about channels. In a staffing business the recruiter is a channel. Two people working the same board with the same database produce different placement rates, and knowing by how much is more actionable than any board comparison, because it is coachable.

The build separates the source from the recruiter so both can be read independently. You will be able to see that one desk converts board candidates at twice the rate of another, which is a training observation, and separately that the board itself underperforms your database across every desk, which is a purchasing observation.

The obvious next step is a multi-touch model: a candidate saw an advert, was later found by a sourcer, then responded to a campaign, so distribute the credit across the three. Marketing attribution has spent fifteen years building those models and the honest summary is that they are extremely sensitive to the quality of the underlying touch data.

Recruitment touch data is worse than marketing touch data. Touches happen on the phone, on WhatsApp, on a platform we cannot legally log, and in a conversation at an industry event. A fractional-credit model built on that foundation produces a number with a decimal point and no basis, and the decimal point is what makes people believe it.

So the Wave 1 build does last-meaningful-source with the full stage ladder, and says so. Multi-touch is only worth building once source capture has been proven reliable in real tenants over real quarters, and we would rather add it when the data earns it than ship a model that flatters whichever channel happens to log the most events.

An agency owner who cancels a job board on the strength of a wrong attribution report has lost real money, and will not find out for two quarters. An owner who is told plainly that the report shows the source of record and the stage conversion, and not a fractional credit allocation, can weigh it correctly against what they already know about their own desk.

The failure mode we are designing against is a dashboard that presents modelled output in the same visual language as counted output. In Surhires, counted things and derived things are labelled differently, and a report that cannot support a conclusion says so rather than rounding to a confident number.

Attribution is only worth the build if it changes a decision. Three decisions are in scope: which job board contracts to renew and at what volume, where to point sourcing effort next quarter, and which desks need coaching on which channel.

For the first, the report gives you cost against placements rather than cost against applications, provided you have entered what the contract costs. For the second, it tells you whether your own database is outperforming paid sourcing, which for most established agencies it is. For the third, it separates the channel from the person working it.