PLATFORM · PREDICTIONS
Every customer has a rhythm.
DAYS 31–45
FROM THE INTERVALS BETWEEN HER OWN ORDERS
1.8× HER CADENCE
DRIFT PAST HER RHYTHM — NOT DAYS PAST A STORE-WIDE MEAN
64%
MARKER SITS AT THE ~2.7% NATURAL-RETURN FLOOR
VERY LIKELY
VERY LIKELY · LIKELY · UNCERTAIN · UNLIKELY — BANDS, NOT DECIMALS
THRIVING
5 RANKED FACTORS
CADENCE · FREQUENCY · VALUE · BASKET TREND · REFUNDS
3.2× IN THE WINDOW
12-MONTH BACKTEST OF REAL ORDERS — FLAG PRECISION, NOT CAUSAL LIFT
NEXT 30 & 90 DAYS
MODELLED REORDER REVENUE — AND LABELLED AS MODELLED
01 — CADENCE
Her rhythm, not a segment average.
The window comes from the intervals between her own orders. Lapse risk is drift past that cadence.
RETURN PROBABILITY
64%
READ AGAINST THE FLOOR
~2.7%
CONFIDENCE BAND
Very likely
A WINDOW, NOT AN APPOINTMENT — THE WEEK, NOT THE DAY
02 — THE METHOD
Where a score comes from, and where it stops.
MEDIAN, NOT MEAN
ONE HOLIDAY GAP CANNOT MOVE THE WINDOW
CONSUMABLES DIFFER
A GRINDER AND A BAG OF BEANS ARE NOT ONE CLOCK
STORE TYPICAL
UNTIL SHE HAS A RHYTHM OF HER OWN
11 DAYS PAST
THE WINDOW CLOSED, AND THE PROFILE SAYS SO
NO RATE AT ALL
UNDER TEN BUYERS A REPEAT RATE IS NOT SHOWN SMALL — IT IS NOT SHOWN
MONTHLY TRIANGLE
SURVIVAL BY ACQUISITION MONTH
5×5, REVENUE PER CELL
THE WHOLE BASE, NOT ONE CUSTOMER
ON EVERY SYNC
AS ORDERS LAND, NOT ON A MONTHLY BATCH
REORDERS · WIN-BACK
A RANKED LIST, NOT AN OUTBOX
TODAY'S TOP MOVES
ONE PRINT-CLEAN PAGE EACH MORNING
FROZEN, WITH A CONTROL
YOU WRITE THE MESSAGE — RETRICS PICKS WHO, AND WHO IS HELD BACK
RANK AND FLAG ONLY
NOTHING MOVES ON A MODEL'S SAY-SO
WATCH IT PREDICT
See the reorder
before it happens.
WINDOWS, NOT APPOINTMENTS · RE-SCORED ON EVERY SYNC
8 orders · due in 6 days
3 orders · 71 days quiet
5 orders · 44 days quiet
12 orders · steady
CONCEPT PREVIEW · NUMBERS ILLUSTRATIVE · YOU WRITE THE MESSAGE
03 — READ AGAINST THE FLOOR
Scored against the floor, not hope.
Every return probability is read against the ~2.7% of lapsed customers who come back with no outreach at all. If she'd return anyway, Retrics says so.
THE FLOOR IS WHAT COMES BACK WITH NO OUTREACH AT ALL · CLEARING IT BY A LITTLE IS NOT A WIN
04 — WHAT IT FEEDS
Predictions rank. They never send.
Every score becomes a board, a brief, or an audience frozen with its control held back — waiting on the message you write and the approval you give it.
BEFORE YOU ASK
Straight answers.
Q01What about customers with only one order?
One order isn't a rhythm yet, so Retrics falls back to how that product is typically repurchased across your store — and watches the second-order window closely, because that is where the strongest measured effect lives. Once she orders again, her personal cadence takes over.
Q02How accurate are the predictions?
Honestly: they are windows, not appointments — the week a customer is likely to need to reorder, not the day. The method is validated on a 12-month backtest of real orders, which measures how precise the flag is rather than what a message causes. A controlled send is read against its own holdout's observed return rate; an audience too small to carry a control is read observationally against the fixed ~2.7% natural-return floor instead. The recovered-revenue figure is withheld today. The holdout is still carved and never messaged on every eligible send, so the evidence accumulates — but the comparison is being rebuilt to count only the customers a message actually reached, and Retrics won’t publish a dollar measured over anything looser. The full method — holdout design, the floor, the backtest windows — is set out in the method section on this page.
Q03Do predictions ever send anything on their own?
No. Predictions only rank and flag. A score can freeze an audience and carve the control that will measure it, but the message is yours to write and yours to approve — and nothing sends on your approval either, because what approval produces is a campaign-safe export you send from your own platform.
See your own customers' rhythms.
Connect your store and Retrics scores your real order history the same day.