Four questions, answered in numbers your board already knows how to read.
Every output is aggregate, every estimate carries a confidence interval, and no individual appears anywhere in any of it.
001
Are equally qualified people paid differently because of where they started?
( Pay Equity )
The model holds job level, tenure, performance rating and age constant, then asks whether background still predicts pay. If it does, the size of that effect is your structural pay gap.
| Background premium The gap in percentage terms and in pounds, with a confidence interval attached |
| Share bypassing credentials How much of your gap education could never fix |
| Glass floor ratio Whether background protects your weakest performers more than it lifts your strongest |
| Function decomposition Which parts of the business carry the effect, wherever group sizes allow it to be reported |
| Gap closure budget What removing the effect would cost, phased across pay review cycles |
On the closure budget.It sizes a policy investment. It is not a list of employees to be given rises, and no individual entitlement follows from it. The instruments it informs are starting salary bands, pay review guidance, limits on manager discretion and progression criteria. Nobody should ever receive a pay award keyed to their parents' occupation, and PRISM will not recommend one.
002
At which stage of recruitment does background advantage enter?
( Hiring Pipeline )
Advancement is modelled at every transition, from application through to accepted offer, holding qualifications constant. The output tells you which specific stage of your own process is doing the filtering.
| Stage by stage odds Where in the funnel candidates with identical qualifications begin to diverge |
| Credential return Whether the same degree is worth more in some hands than others |
| Sourcing channel audit Which channels bring which backgrounds, and at what conversion. Referral and executive search usually look very different from open advertising. |
| Assessor variation How much of the outcome depends on who happened to run the interview |
| Quality of hire check Whether any pattern found is actually justified by later performance |
The recruitment firewall
Candidate background data is voluntary, and declining has no effect on an application. It is stored apart from your applicant tracking system. No recruiter, hiring manager or interviewer can see it at any point, and that is enforced by access control rather than by policy. Analysis runs only after decisions for a cohort are complete. Nothing is ever inferred from a name.
On assessors.Individual interviewers are never named in any output, and no such report can be produced. A coefficient estimated from a few dozen decisions carries an error wide enough that ordinary variation looks like signal, and testing a panel of forty assessors at conventional thresholds will produce two apparent findings from noise alone. An organisation acting on that list would be disciplining people for randomness.
003
Does background accelerate careers, and does it shield underperformance?
( Progression Velocity )
Two people join at the same grade in the same month. This module measures where they are a decade later, and how much of the difference performance actually explains.
| Advancement speed by band Levels gained per year, controlling for performance and tenure |
| Time to first promotion The median wait, by background band |
| Glass floor index Whether protection for weak performers exceeds acceleration for strong ones |
| Exit and retention Whether departure and redundancy risk vary at equal performance |
| Career ceiling The highest level each group actually reaches, by tenure |
Why executives find this module the most uncomfortable.The redundancy finding in the research runs above twenty per cent at equal performance. If that holds in your organisation, then every restructure you have run has been quietly selecting on background, and the effect will show up in the seniority profile you have today.
004
Did the money you already spend on this actually work?
( Intervention Return )
Most equity programmes are measured by attendance and satisfaction. Neither tells you whether the gap moved. This module establishes a baseline, measures again afterwards, and reports the difference with a significance test attached.
| Movement before and after Whether the coefficient actually shifted, and whether the shift is statistically real |
| Return in pounds Value created against programme cost |
| Decay rate How quickly the effect fades without reinforcement, and therefore when to reinvest |
| Mechanism attribution Which dimension of background the intervention actually touched |
| Programme comparison Head to head return across everything you run |
Where a programme reached some parts of the business before others, the comparison is estimated against the untreated group, which is a considerably stronger claim than a before and after average.
005
What appears in every deliverable, whatever it says
( Standards )
adjusted to suit
A confidence interval on every estimate
A coefficient without an interval is not a finding. Where the interval is wide relative to the estimate, that is itself the result, and the right recommendation is more measurement rather than more spending.
The typical figure, not the extreme one
Headline premiums are reported between quartiles, describing employees who actually exist.
Non-response reported whatever it shows
Who answered is compared with who did not, on the workforce data available for everybody, and that comparison appears even when it is inconvenient.
Suppressed groups shown as suppressed
Where a group is too small to report safely, the gap is drawn and labelled rather than quietly dropped from the chart.
Scope declined where numbers do not support it
If your population cannot support a breakdown at a credible level of certainty, we remove it at proposal stage instead of delivering noise with a decimal point on it.
Disagreement presented as disagreement
Where two methods point different ways, both appear, with the reason.