The propensity-to-return model sizes every credit and offer within a cost cap Finance signs; entry is the completed trip, never the score, and the churn-risk score stands in until it is built.
The problem Our best moment is answered with silence.
Why it happens The most positive moment in the whole lifecycle is followed by silence, a broken survey link and referral asks from three senders with no sequencing. The referral mechanism itself is broken: rewards fail to arrive and rejections go unexplained, so the programme punishes the behaviour it exists to create. And the repeat ladder's second cliff has no owner: Journey 1 works the step from the first upgrade to the second; nothing works the step from the second to the third, the last before the habit holds.
The behaviour we are driving They see the saving proved, book the next trip on their own rhythm, rate the app after a good trip, and refer a friend once it is earned, the day after a first-class trip that went well.
The job to be done “Remind me at the right moment, make sharing easy, and ask me for nothing until the trip has actually gone well.”
Entry is the completed trip, never the score. The propensity-to-return model sizes the incentive: which credit a customer gets, and at what level, within a cost cap Finance signs. Until Analytics builds it, the churn-risk score (0 to 100, nightly, null on thin history) stands in and decides one thing: whether the day after the second positive trip carries the £5 credit, at or above a threshold Analytics sets. The two scores never share band names. Coverage at two upgrades is an open Analytics ask.
Not sized in the Q4 marketing model. Counted in Journey 1's row: the model sizes one post-win programme, and both journeys work the same first-time winners for the same second purchase. Referral volume is audience-only until Mention Me reports.
Ladders to Build the Habit Loop: first-time buyers who bid or buy again within 30 days of the first win, 33.6%. Metric moved Second purchase within 30 days %. Evidence .
The planned case, a 30% lift in conversion, as presented to the board on 23 Sep 2026 (the Sizing tab behind the model's Business Case tab): volumes checked by Analytics (Ivi, 8 Sep 2026); organic and target rates measured against holdout and confirmed by Analytics; a buyer's month valued at £30 an upgrade, 1.29 upgrades in a first-purchase month and 1.67 in later months (the 20 Sep change; the earlier £172k counted one upgrade a month); net is gross at the 23% take rate (agreed 8 Sep 2026). Source: Q4 2026 Marketing Model, Business Case (output) and Sizing tabs, exported 2026-09-29 (the live model). Every figure is incremental of the customers who would have returned anyway: worked volume × (target rate − organic rate) × months active × times a year × value per active month.
Each engineering-row chip is coloured by the ask: APP AND JOURNEY OWNED COMPUTED TRAIT · ANALYTICS NEW EVENT · ENGINEERING BUILDS READY · NOTHING TO BRIEF
| The moment | Phase 1, ships now | Phase 2, waits on engineering | Who builds it |
|---|---|---|---|
on the map: 1 · The value proof (the reflection touch) The value proof | braze push Ninety minutes after arrival, one push proves the win with the customer’s own figure: the saving on this trip, true and computed, never an up-to claim. braze content card The same value proof waits in the app for whenever they next open it, and carries the one durable fix for this group: a value-framed opt-in… | Nothing waits | Braze · Braze buildAnalytics · computed trait, logic changeDesign · copy and creative |
on the map: 1 · The value proof (the reflection touch) The return-leg seed, on the reflection touch | email When the customer's own return leg has a live upgrade auction, the day-after email leads with it: the most valuable, most time-bound action available, and usually… | Nothing waits | Braze · logic change, computed traitDesign · copy and creative |
on the map: 3 · The day-after email The day-after slot | email Twenty-four hours after the trip, one email, its job decided by priority: the return-leg bid when one is live (carrying the £5 credit for a second-trip… email The cumulative saving, banked and stated, with a single plan-your-next-trip action. braze push The cumulative saving, banked and stated in one line, with the plan-your-next-trip tap. | email, braze push After the second positive trip, a customer the score flags as unlikely to return unaided gets a credit toward the third upgrade, its level sized by… | Backend engineering · logic change, new event, decision needed, new modelBraze · logic change, new eventAnalytics · computed trait, logic change, native to braze migrationDesign · copy and creative |
| The milestone | email, braze push The day after the trip that lands a rung, one email recognises the count and names the next mark (three in; five is the next mark),… | Nothing waits | Braze · Braze buildAnalytics · computed trait |
on the map: 3 · The day-after email The day-after survey | email Nothing changes on this send: it stays at purchase plus one day as Product's booking-experience feedback, service-classified and never combined with marketing content. Engineering rows D8 | Nothing waits | Design · creative only |
| The survey repair | email, braze push The survey sends become readable and safe: identity restored on the three emails that stripped it, customer identity out of URL query strings everywhere, the survey… Engineering rows D8 | Nothing waits | App engineering · app changeBackend engineering · new event |
on the map: 5 · The re-booking nudge The re-booking nudge | braze push, email The nudge points at the existing Route Watch mechanic on the customer's own route, timed by their own median gap between purchases instead of a fixed… | Nothing waits | Braze · new eventAnalytics · computed traitDesign · copy and creative |
| The referral stand-downs | email The onboarding day-seven referral ask stands down (it asks before any transaction, against the ruled sequencing), seasonal referral broadcasts exclude customers inside a journey episode or… | Nothing waits | Braze · Braze build |
| The reward experience | email Rewards arrive when promised and rejections say why. Engineering rows D6 | Nothing waits | Backend engineering · new event |
| The disrupted trip | Nothing changes here and nothing ever sends: a savings celebration after a disrupted trip is an insult with a send time. | Nothing waits | Marketing · copy only |
| The power user | No nudge: blanket-messaged power users bought 11.8 percent less against a holdout. | Nothing waits | Analytics · computed trait |
on the map: 5 · The re-booking nudge The first-trip forward asks | Forward-looking asks stand down for thirty days: the welcome journey owns the path to the second win, so a first-time winner is never contacted by two… | Nothing waits | Marketing · copy only |
| The next trip's travel window | Relationship comms never fire inside another trip's travel window: the pre-travel and travel-day journey owns the customer there, and this journey's whole spine yields. | Nothing waits | Marketing · copy only |
| The journey holdout | The headline test is a journey-level holdout on the whole spine, read on the next-transaction rate within sixty days. | Nothing waits | Analytics |
The words are on the journey map above. A D number opens that row of the Engineering brief; the engineering board uses the same numbers. Live sends not named here continue unchanged.
| The question | Owner |
|---|---|
| Sign the positive-trip definition, and confirm every component is queryable at send time | Blocks launch Analytics, via Ivi |
| Size the journey: the second-transaction cohort, the gate pass rate, the baselines and the holdout | Blocks launch Analytics, via Ivi |
| Reconcile the purchase-gap medians | Blocks launch Analytics, via Ivi |
| Validate the schedule-derived arrival anchor | Blocks launch Engineering, via Sophie |
| Verify the survey-link failure before the fix is briefed | Blocks launch Engineering, via Sophie |
| Confirm the reward mechanics before any copy promises them | Blocks launch Lisa, and the Braze developer |
| Decide the switched-off nurture Canvas: re-anchor or retire | Blocks launch Lisa, and the Braze developer |
| Arbitrate the arrival window | Blocks launch Product |
Everything Journey 6 needs from Analytics, in one place. Two facts frame it: the journey cannot launch without the positive-trip gate and the savings traits, and it cannot be sized at all until its denominators exist. The answered ledger is first so nothing settled gets re-litigated; every open ask names what comes back and where to start.
What is already settled, so nobody re-opens it here. Everything else on this tab is an open ask.
| Settled | The position |
|---|---|
| The day-after survey email stays | Kept at purchase plus one day as pure feedback. The journey’s own email ships in addition on its trip anchor; neither send moves for the other, and the survey’s completes are a guardrail that must not fall. |
| The episode cap is three | Ratified: the value proof, one arbitrated email, one nudge. The lapse journey’s two-touch cap is unchanged and this is not a precedent against it: the two caps govern different units. |
| The Journey 1 boundary | Ruled: the welcome journey owns the forward asks for thirty days after a first win; this journey owns the trip-anchored touches from trip one and the full spine from the second transaction. |
| Persona is tone, never cadence | The current model classifies all but 1.7 percent of the base; business and leisure purchase medians are near-identical, so persona selects wording only. |
| Broadcast revenue is attributed, never incremental | Ruled: no referral broadcast figure is quoted as incremental, and campaign control groups in the console are labels with zero members, never holdouts. |
Analytics owns each output. CRM signs the definitions, and Sheran ratifies the operating points. Statuses are as of the last publish; the live status is walked at the scoping session.
| # | The ask | The output | What it gates | Status |
|---|---|---|---|---|
| 1 · Sign the gate and the definitions · the launch-gating trait rides on the first row | ||||
| 1 | Sign the positive_trip definition (pin D1, with Engineering and Data): trip_completed AND no disruption, declassification or open Delay Repay / refund / dispute, queryable at fire time. | The signed definition; Engineering builds to it | Launch: the gate on every send | DRAFT, Ivi to sign |
| 2 | Reconcile the converter medians (the 2-8d converter view vs the 6wk all-transition median, carried from the welcome journey open Q3) before the 15-16d cadence fallback counts as evidence-locked. | One reconciled definition | The cadence fallback’s evidence status | Open |
| 3 | Confirm the canonical M0-to-M1 figure: canonical 20.15/7.47 (brain) vs Jun-2026 export 20.22/7.43. | One canonical figure | The north-star framing this journey attacks | Open |
| 2 · Build the traits · definitions signed by CRM; the sync pipes are Engineering’s rows | ||||
| 4 | Build the savings traits (pin D2): per-trip saved £X + cumulative £X across N trips, warehouse-built, nightly; true figures, never "up to". | Two computed traits, nightly | the value-proof moment copy: a missing figure HOLDS the send | Does not exist · LAUNCH-BLOCKING |
| 5 | Build personal_median_inter_txn_gap (pin D3), ONE build shared with Journey 8: it times the re-booking nudge here and sets J8’s slip boundary at about 1.5x. Fallback base median 15-16d below n=3 txns. | One computed trait, nightly, two journeys | Nothing at launch (the fallback exists); phase-2 timing | Priority 2 |
| 6 | Confirm the persona sync (pin D7; the 3 Jul model, 1.7% unclassified, retiring the stale about 44% caveat all four cards carry). | A one-line Braze confirmation | Per-persona tone copy; low confidence routes generic | Model exists · sync to confirm |
| 3 · Make the test readable · measurement design; size it before the Braze build starts | ||||
| 7 | Size the journey-level holdout and the MDE before build; no fixed % prescribed. It must separate journey lift from model targeting once D3 times touches (ruled 20 Jul; holdout structure at the agreed holdout structure). | Holdout % + MDE | The Braze build: do this first | Open |
| 8 | Size the volumes and baselines: eligible positive-trip volume/yr out of 710,616 departures, the first-class completed-trip referral pool under the peak-moment gate, the baseline next-transaction-within-60d rate, the k-factor baseline, and the transaction-count split that powers the referral-quality guardrail. Flag now if the decision read runs past the quarter. | The volume + baseline table | Launch caps, and the read’s runtime | Open |
| 9 | Wire the day-after survey email completes guardrail (RULED 28 Jul, the ruled position: it applies once the J6 slot email ships) and verify the day-after survey email trigger coverage with Lisa (auction wins, or BUN only?). | The wired guardrail + the coverage answer | Launch | Ruled · to wire |
| 10 | Read the survey-send fix pre/post (click-to-complete). A fix read, NOT a test; blocked on the Engineering fix (pin D8). | The pre/post read | Any survey-quality claim | Blocked on the P0 fix |
| 11 | Log the Test & Learn roadmap record (with Marketing): hypothesis, KPI, guardrails, reads. | The logged record | The read | Open |
Decisions the team rules on Analytics’ evidence: the positive_trip definition (ask 1) · the power-user treatment given the -11.8% holdout read · the decision rule at the read, agreed BEFORE launch.
Every customer completing a positive trip, once the gate is signed and queryable; one episode per trip.
A journey-level holdout on the whole spine, sized by Analytics, never a fixed percentage, protected from broadcasts too.
Next-transaction rate within sixty days of a positive completed trip. The do-nothing reference already exists: 28% of first-time buyers repeat within sixty days unaided.
One, two and four weeks in, with the decision read against the sixty-day window; flag now if it runs past the quarter.
Guardrails: the day-after survey’s completes must not fall; opt-outs and complaints watched; sessions must not rise while purchases stay flat (a prior send drove +108% sessions and -12% purchases); total referral volume must not fall net of the broadcast stand-down; and the referral-quality read: referred-friend activation split by the referrer’s transaction count at time of ask, with the gate re-tightened on evidence if trip-one referrals produce friends who activate materially worse.
The power-user programme: the milestone moments, the ladder to three then five, the third-upgrade incentive by score, and the referral cap at once a quarter. Each row is on the Analytics board; the link opens it.
| # | The ask | Hand back | Start here | Board |
|---|---|---|---|---|
| 1 | The lifetime upgrade count as a nightly trait: bid wins, Buy Now and instant wins, refunded and cancelled sales excluded, from the warehouse (auction wins emit no client event). Powers the milestone moment, the ladder line and the referral-quality guardrail. | The count in the daily sync (the pipe is the engineering brief’s D13). | The milestone row on the decisioning matrix. | on the board |
| 2 | Churn-risk score coverage at two upgrades, and the at-risk threshold for the third-upgrade incentive: the score is null on thin history, so confirm coverage at exactly two upgrades, the fallback (the slip trait), the threshold and the monthly size of the at-risk band; restate the second-cliff baseline (68 to 71% of second-time buyers never reach a third) on the same population. | Coverage, threshold, band size, baseline. | The third-upgrade incentive row. | on the board |
| 3 | Four test designs with holdout sizing: the third-upgrade credit against recognition-only (third upgrade within 90 days, the double read); the milestone reward cells at ten and twenty; recognition against silence for the power band (retires the minus 11.8% read as the sole evidence); the referral cap read, opt-outs and complaints per ask number. | Sized holdouts, read dates, kill conditions. | The test blocks on the four rows. | on the board |