A pipeline that tells you what is slipping

A candidate who has been in your database for two years is not a problem. A candidate who has been at submitted for nineteen days is. Ageing is measured from the last stage change, per stage, and the threshold is set per stage rather than globally, because a three-day-old offer is fine and a three-day-old interview confirmation is not.

Thresholds can be learned from your own history rather than guessed. If your average submittal-to-response for a given client is four days, an eleven-day silence is worth a call.

The stages differ in what a delay actually means, which is why one global number never works. Silence after a submittal is usually a client attention problem. Silence after an interview is usually a feedback problem. Silence after an offer is a candidate problem and frequently a counter-offer. Same elapsed time, three different phone calls, and a system that treats them identically sends the recruiter to the wrong one.

Every requisition carries a health state derived from real signals: how many candidates have been submitted against it, how many were rejected and for what coded reason, how long it has been open relative to your own time-to-fill for that role type, and whether the client has responded to anything recently.

A requisition with eleven submittals and nine salary rejections is not a sourcing problem. Surfacing that early is the difference between renegotiating the brief in week two and losing the job in week seven.

The health state is a prompt rather than a verdict, and it is worth saying so because a red badge invites people to argue with the badge instead of acting on it. What the state is claiming is narrow: on the evidence in the system, this requisition is behaving unlike the ones that fill. The recruiter frequently knows a reason that is not in the system. The point of the flag is to make sure the reason gets said out loud rather than assumed.

Revenue forecasting uses your own conversion rates by stage rather than a probability a recruiter types in. If your desk converts thirty-one percent of first interviews to offers and eighty-four percent of offers to placements, the forecast follows from where the candidates actually are.

The forecast is shown as a range with the assumptions visible, because a single number invites a false confidence that a staffing pipeline rarely deserves.

It gets less useful the further out you look, and the interface says so rather than extending the line. A forecast for the current month is largely arithmetic on candidates who already exist. A forecast for the quarter depends on requisitions nobody has won yet and candidates nobody has met, which is not forecasting so much as budgeting. We would rather show a narrow claim you can rely on than a wide one that gets quoted in a board pack.

A recruiter sees their own board and their own exceptions. An owner sees the same data rolled up by desk, by client and by job type, and can drill straight into the underlying records.

Both views are the same data. Nobody is maintaining a separate management report, which is what makes the numbers trustworthy.

That shared basis changes the politics of reporting as much as the mechanics. When the management number is produced by a separate process, every conversation about performance starts with a dispute about the number, and the recruiter is usually right that it is missing something. When both people are looking at the same records with the same definitions, the disagreement moves to what to do next, which is the conversation that was supposed to be happening.

Exceptions can be pushed to a daily digest, to a Slack or Teams channel, or into the task list a recruiter already works from. What matters is that the alert carries the next action, not just the fact.

An offer past its response deadline arrives as a task to call the candidate, with the candidate, the client contact and the offer detail attached.

Volume is the thing that kills an alerting system, so the defaults are deliberately quiet. A digest once a day beats a notification per event, exceptions are grouped by the action they need rather than by the record they came from, and anything a recruiter dismisses repeatedly is a signal that the threshold is wrong rather than that the recruiter is ignoring it.

Every desk that switches ageing thresholds on discovers the same thing in week one, which is that the backlog is real. Dozens of submittals genuinely have been silent for a fortnight, and several offers genuinely did pass their deadline. The instinct is to raise the thresholds until the noise stops, and that is how a monitoring tool gets quietly turned off in its first month.

The sequence that works is to treat the first pass as a clean-up rather than as an alert stream. Work the backlog once, close or chase everything in it, and only then set thresholds from each client's own response history rather than from a single global number. Starting with two exception types, usually stalled submittals and offers past deadline, gets a desk into the habit without burying it.

After that the useful measure is not how many exceptions there are but how quickly they get closed. A board that carries fifteen open exceptions on a Monday and three on a Thursday is a working desk. One that carries the same fifteen all month is telling you something about capacity, and it is better to see that on a board than to infer it from a bad quarter.

It does not fill requisitions. Everything on the board is a description of work that has already happened, arranged so the gaps are obvious. The board can tell you that a requisition will not fill on its current trajectory three weeks before that becomes undeniable, which is genuinely valuable and is also not the same thing as doing something about it.

The forecast is only as good as the history behind it, and on a new desk there is no history. Until enough placements have gone through the stages, the conversion rates are computed from a handful of outcomes and the interface says the sample is too small rather than showing a confident figure. Source-level forecasting is thinner still, because source-of-hire attribution is in build for the Wave 1 release, so today the forecast decomposes by desk, client and job type but not yet by where the candidate came from.

It also cannot see the things that are not in the system. A client who has told your recruiter over lunch that the budget is frozen looks, on the board, like a client who has stopped responding. The health state will flag it, and the flag will be right for the wrong reason. That is an argument for logging the conversation, not an argument against the flag.