Analytics
Are we on track?
Executive summary
Open orders
Status is Active · one order is one hire
Why this number
A bare count of open orders reads as that many seats with nobody on them, and only about a quarter of them are — hence the split. Paused orders and near hires are NOT open and are shown beside the count rather than folded into it: a paused order is coming back, and a near hire already has a named person on it.
Finalists
Interview or later, on pipelines with an open order
Why this number
Split because the two halves need different things: the pre-offer group needs interviews run, the at-offer group needs paperwork closed. Review is excluded deliberately — most funnel washout still happens there, so counting it would overstate how close the book is to filling. Counted as distinct people: one person can be a finalist for several seats, so this is not the number of seats that can be filled.
Candidates in play
Active in the last 90 days, on pipelines with an open order
Why this number
Only candidates active in the last 90 days — Salesforce doesn't expire idle applications yet. The second number is the one anybody acts on.
Starting in 30 days
Accepted, with a start date booked
Why this number
A logistics number, not a health metric. Onboarding detail sits in the action queues below.
Orders filled vs raised
Since 2026-07-01 · there is no headcount plan, so demand is the denominator
Why this number
Below 1.0 the book is growing. The current month is still accruing hires and always reads short.
Offer acceptance
Hires ÷ (hires + candidate withdrawals)
Why this number
Our own rejections are out of the denominator: when an offer goes out we intend to hire, so a withdrawal after offer is the candidate's decision and a rejection is ours. Counts offers MADE since 2026-07-01 — an offer's fate can land long after it went out, so the tile follows the offer rather than the outcome.
Time to a passing interview
Median days from ready to a candidate passing · the measure Crossover owns
Why this number
This is the cleanest attribution boundary for a duration, not the edge of the remit — the work continues through offer, placement and start. Time to offer and time to hire sit beside it because both end on the customer's side of the line. Counts interviews PASSED since 2026-07-01, not orders raised then: cutting on the order's own clock would drop the long cycles this measure exists to show.
Then to offer, and to hire
Downstream of the passing interview · customer-owned latency
Why this number
Both clocks start at Date Ready, not at requisition approval, so neither is comparable to an industry time-to-fill — the benchmark is a footnote here, never a line on a chart. The negative durations are kept: an offer that predates its order's own clock is a retroactive R2 on a seat already being worked, and dropping those moved the median by three days in an earlier pass. Counts offers MADE since 2026-07-01, so an order ready long ago whose offer landed inside the range is in.
Readiness gap
Days an order waits for its requirements before it can be sourced
Why this number
The same-day orders are excluded from the median on purpose: Date Ready defaults to the creation date, so most zeros are the default firing rather than a turnaround anyone achieved. Counts orders that became READY since 2026-07-01; one still waiting for its requirements is not yet measurable and is in neither number.
Demand and supply
Orders arrive continuously; there is no annual plan
The order book over time
Everything ever opened, split by how it left the book and what is still on it
Why this number
The top edge is every order ever opened; the top band's thickness is the book still open at that moment. It starts from the 0 orders already open when the window began, not from zero — a hiring book does not start empty. An order leaves the book two ways and the two lower bands say which: its hire lands, or it is cancelled. Nothing is in both. An order that is merely PAUSED has not left — one on hold or awaiting budget stays in the open band, because it is coming back, and the open-order tile above counts those separately. TWO STORES: the raised and filled counts are the warehouse mirror, which currently holds a slightly larger open book than the live wire — the difference is mirror lag, not a different definition. The cancelled band is read LIVE, because no cancellation date exists in the mirror or on the order itself; its dates come from Salesforce field history, which records when the reason itself changed rather than when the order was last touched — the last-modified stamp it replaced drifts on any later edit of any field. THAT IS BETTER, NOT CLEAN. Field history dates the SYSTEM EVENT, and a bulk status change is a system event: it writes one history row per order inside the same few seconds, so a batch update still lands a pile of cancellations on one day. Field history on hiring orders also begins on 2026-04-28, so any month before that draws an empty cancelled band however many orders were really cancelled in it. Read the band as the month a cancellation was RECORDED, not the month it was decided.
Orders raised against orders resolved
By month · an order is filled when its hire lands, not when the offer goes out
Pipeline health
Where the funnel loses people, and how the book is ageing
Hiring funnel
Applications created since 2026-07-01, by the furthest stage they reached
Why this number
Reconciles: 0 on the ladder plus 0 off it equals 0 applications. Stage is taken from the furthest stage reached, not from where an application stands — a rejected application reads Rejected, which says nothing about where it died. Covers every pipeline, not only those with an open order, so it is a wider population than the tiles above. Not every pipeline runs every stage, so a rung can read high simply because few pipelines enable it.
Open orders against their own SLA
Days remaining against each order's own goal
Why this number
Goals run from 42 to 140 days, so raw age cannot be banded: 60 days is healthy against one goal and blown against another. Within SLA is the working state, not a warning.
Late-stage candidates by campus
Distinct people matched to an open order, by stage
Why this number
Early stages are excluded on purpose: including them turns every bar into a proxy for ad spend rather than for pipeline health.
Hires by vertical
Hires since 2026-07-01, by the pipeline's vertical
Velocity
Where the time actually goes
Days in stage
Median days from stage entry to result · bottleneck finder
Why this number
Measured from STAGE ENTRY, not from when the step was picked up: every step of a stage is created when the candidate enters it, so for a stage running several steps this is an upper bound.
Time to offer and time to hire
Monthly median, in days
Why this number
Direction matters more than level here, because both clocks start at Date Ready and neither is comparable to an industry benchmark.
Channel ROI
Judge a source on hires and yield, never on applicant volume
Applications and hires by source
Applications created since 2026-07-01 · lower applications-per-hire is better
Why this number
Candidates whose account first arrived from LinkedIn or Indeed are carved out of Unknown and Search Engine only, never out of a named paid channel. Retargeting is deliberately left whole: it is judged on whether re-engaging the pool worked, and folding it into its own feedstock would erase that.
Action queues
What makes someone open this page daily
Stalled
Ranked against each stage's own tempo
Offers awaiting reply
Sent, no answer yet
Why this number
Amber past 7 days, red past 14 — a policy target, deliberately not fitted to what we do today. The median open offer here is far older than that, and that is the thing to fix rather than a tempo to normalise. Rows marked at risk hold a FAILED offer step: that verdict is recorded rather than inferred and it fires before the stage moves, and recovery from here is rare. 0 applications carry one.
Outstanding feedback
Interviews submitted, not yet graded
Why this number
Scoped to the last 30 days of submissions, and narrowed to the pipelines your filter selects.
Hired, not started
Accepted, with a start date ahead
Why this number
Open onboarding tasks are deliberately not counted here. There is no usable task vocabulary to count them by — the work-unit type is null on every task created this year, and the open-task pool is dominated by bulk rows owned by deactivated recruiters.
Exceptions
Long-open orders, split by who can actually fix them
Starved — past SLA with nobody close
Crossover's to fix: these seats need sourcing
Why this number
This is the small table on purpose. The large majority of long-open orders already have a passing candidate; pooling them sends the wrong people to fix the wrong problem.
Past SLA, but somebody is close
The wait here is paperwork and decisions, not sourcing
Why this number
Matched, never filled: supply is inferred from the pipeline, and a matched finalist is not an accepted offer.
Near hires — the last mile
A named person is already on the seat
Why this number
These orders are NOT open — they leave the open count at offer, which is why the headline above does not include them. A third group exists that this table cannot show: seats where a matched candidate has passed interview but no application is linked yet. That link is inferred rather than recorded, and no store here carries it.