Skip to content
allanninal.dev/build
Collaborate

Part 4 of 7 · Background check chaser series ~5 min read

How the start date prediction works

The prediction is the smallest piece of arithmetic in this series and the part people want to over-engineer. It resists a model, resists a probability, and resists a dashboard, and the reason is that its entire audience is one busy person who needs to decide whether to move a start date.

Key takeaways

  • Three outcomes: comfortable, tight, will not make it. No percentages.
  • Expected completion is the request date plus your own median for that check type.
  • A chased item that has not moved gets its expectation extended, not reset.
  • Held items add their blocker’s remaining time to their own duration.
  • The day-one prediction is the most valuable one, because the start date can still move.

Three words

How an outstanding check is predicted against a start dateA vertical chain of five steps entered by a box labelled An outstanding check, together with a start date. Step one computes expected completion as the request date plus your own median for that check type, read from your own history. Step two asks whether it is blocked by another check; if so it exits to Chain the durations, adding both in sequence. Step three asks whether it has been chased and is still silent; if so it exits to Add the chase lag, taken from your history. Step four compares the result with the start date. Step five produces one of three words: comfortable, tight, or will not make it. A note says there is no probability, because a hiring manager needs to act and sixty-eight per cent is not an action.AWS ACCOUNTAn outstanding checkand a start dateExpected completionrequested + medianYour own historymedian per typeBlocked by another?add its remaining timeChain the durationsboth, in sequenceyesChased and still silent?extend, do not resetAdd the chase lagfrom your historyyesCompare with the start dateOne of three wordscomfortable / tight /will not make itNo probability. A hiring manager needs to act, and 68% is not an action.
Fig 1. How each outstanding check gets its one-word verdict. The chase-lag step matters: a chased item that is still silent is slower than a fresh one, not faster.
  • App integration
  • Machine learning
  • Management
  • Analytics

Why not a percentage

A percentage invites interpretation, and interpretation is exactly what the reader has no time for. “Seventy-two per cent likely to complete before the start date” requires somebody to decide what to do at seventy-two, and different people will decide differently on different days.

Three words remove that. Comfortable means do nothing. Tight means chase it today. Will not make it means have a conversation about the start date this week. The thresholds that produce those words live in the check list, so if your business wants tight to start earlier that is an edit rather than an argument.

The chase lag

The subtle bit. A reference requested nine days ago with a five-day median is late, and the naive response is to expect it tomorrow. But a check that has already blown through its median and been chased once is empirically much slower than a fresh one — the referee is on leave, the address is wrong, the agency has a backlog.

So a chased item’s expectation is extended by the median time from first chase to arrival, taken from your own completed checks. In practice that is usually another week or more, and it is what turns a cheerful “due tomorrow” into an honest “this will not make the 28th”.

The day-one prediction

The most valuable output of this system arrives on the day the offer is accepted, before anything has been chased, when the start date is still a proposal rather than a commitment.

How a start date is checked for realism on the day an offer is acceptedA horizontal row of five boxes. Offer accepted: a start date twenty-one days out is proposed. Medical: median twenty-six days. Depends on: the offer becoming unconditional, adding three days. Earliest realistic: twenty-nine days. Said on day one: rather than on the Friday before. A note says nobody chased anything wrong here and the date was never achievable.THE MOST USEFUL SENTENCE THE SYSTEM WRITESOffer acceptedstart proposed: 21 daysMedicalmedian 26 daysDepends onoffer unconditional: +3Earliest realistic29 daysSaid on day onenot on the FridayNobody chased anything wrong here. The date was never achievable.
Fig 2. The day-one arithmetic. This is the one case where the system prevents a problem rather than surfacing one, and it costs nothing but a subtraction.
  • App integration
  • Machine learning
  • Management
  • Analytics

There is no cleverness in that calculation at all. It is one addition and one comparison, and the only reason it is not done routinely by hand is that it requires knowing your own median durations, which nobody records. The system records them as a by-product of doing everything else.

What the prediction is careful about

  • It never predicts a result, only an arrival. Whether a check comes back clear is not something it knows or should guess at.
  • It uses your medians, not the provider’s stated turnaround. Those differ, consistently, in one direction.
  • It extends rather than resets on a chase. A silent chased item is slower than a fresh one.
  • It says nothing about the candidate. A slow reference is a fact about a referee.

Next: how the chasing picks who to contact.

All posts