Clayface is an AI-native market-intelligence platform: it watches every governed KPI × scope against event-adjusted expectations and dispositions everything that moves — so leadership attention only goes where it pays. This demo runs on a Bloomingdale's tenant: real org shape, governed metrics, synthetic [sample] numbers. Across the last few weeks five things moved. The queue sorted all five — and this deck shows the sort, then follows #1 end-to-end.
Release R-1142 ships to mobile checkout. Nobody thinks about it twice.
Summer Glow begins. Traffic climbs — a promo week like six before it.
Data certified through Sun Jul 05. Somewhere in the noise, a number lands wrong.
You are here. The WBR is out, the trading call is at 9:30 — and the week must explain itself.
For a cold audience — 60 seconds here"Name the tenant honestly: Bloomingdale's shape, synthetic [sample] numbers, the pilot's exact four sources. Then make one promise: by the end of this week, Monday's argument becomes Monday's action."
In the old world, every one of these becomes a fire drill: a dip to investigate, a spike to explain, a drift nobody sees, a pipe that impersonates a crisis. Clayface gives each a verdict with evidence — real anomaly, normal variation, or not a business problem at all. Click any card. All figures [sample].
Broke below what six comparable promos support — the promo can't excuse it; concentration named: mobile × Northeast × delivery-promise step.
Carousel "Surfer Chic" + designer drop is first-of-kind — zero comparables exist, so nothing may suppress it. The envelope flags upside with the same rigor as downside; full-price sell-through 84%, AUR +6% — real demand, not promo froth.
The trap: the sales line looks fine — returns reason-mix shifts to fit/size (41→58%), concentrated in 34B/34C. WBR structurally can't see family-grain returns. Projected to cross the gate ≈ W31; buyer asked to confirm a supplier change.
The same math that flagged #1 suppressed this — six comparable events say +22…+41% is normal. Suppression is the model working in both directions, and it's auditable: the tray keeps the evidence.
The paired artifact: traffic "falls" and conversion "jumps" simultaneously — a source symptom, not two business signals. Intact orders are the tell. Affected tiles get a "partial data" chip; the gap is excluded from model training.
The line that lands"Notice what's not here: no hand-tuned thresholds, no alert fatigue. The same envelope logic ran both directions in the same week — it suppressed Beauty's +32% and flagged RTW's dip. That's 'anomaly versus normal variation,' proven."
Mid–Summer Glow promo, Women's RTW digital orders quietly land 7.5K below the lower bound of what six comparable promos support. The loss hides inside promo noise, concentrates in mobile checkout in the Northeast, and traces to a carrier disruption stretching delivery promises from 2-day to 5–7 days. A site release looks guilty — and is exonerated by the evidence.
Open with"This is the required use case, verbatim: separate anomalous signals from normal variation — during the noisiest possible week to do it: a sitewide promotion." Then tap the Priority tile to show the math is governed, not vibes.
The WBR shows Women's RTW digital soft vs. plan — but the Summer Glow promo gives everyone an excuse to argue. Click each door to see where the morning ends up; the quotes arrive as you advance.
Soft number, promo week, five dashboards, zero shared baseline. Every function reads its own tool and defends its own turf.
The week landed below what comparable promos support. Concentration named: mobile × Northeast × the delivery-promise step. Strongest supported driver: carrier disruption. R-1142? Evaluated and cleared.
Framing"Keep both doors on screen. The room has lived the left one — let them recognize it before you show the race."
One clock, two organizations. Watch the margin meter on the right while the left side argues. Drag the scrubber to any moment.
One row in a 40-tab workbook. The promo makes the miss deniable.
Marketing cites sessions +9%, merchandising blames promo mix, finance sees an 18% gap. Three tools, three truths.
"Didn't engineering ship a release?" R-1142 becomes prime suspect; tech benched to defend itself.
Order data, clickstream exports, promo calendar, WBR files — reconciling definitions before analyzing.
Every function re-cuts the data to defend its own story. The miss keeps compounding.
It was never the release — a carrier disruption stretched NE promises to 5–7 days. Mitigation starts next week.
Breach vs the event-adjusted envelope · $1.4M/wk at stake · confidence High · maker/checker ✓.
Mobile × NE −11.2K · delivery-option step · promise slip 2d → 5–7d · R-1142 evaluated & cleared.
Priya validates method in the evidence drawer instead of rebuilding it. Tech never gets accused.
fulfillment-mitigation → Digital Commerce lead + Delgado · ship-from-store, pickup at checkout, Loyallist expedite, honest promises.
✓ Done since Monday. Engineers on roadmap. Analysts analyzing.
Mitigation already recovering the funnel.
Land this line"Detect → explain → assign → approve → measure. And notice the platform spent its first breath clearing your tech team, not accusing it."
Not a static plan — an event-adjusted envelope trained on the six comparable promos in the calendar's own history. Hover the points; flip the toggle to see why the old way sleeps through this week.
The differentiator"A promotion is precisely when static thresholds go blind — the miss is measured against the event's own history, so the event cannot excuse the miss."
Click each source for its contribution — then drill the decomposition to find where the orders went.
Summer Glow W27–28 · 6 comparables → envelope lifted to [98.7, 125.1]K.
Demand −18% vs adjusted median · AUR flat · BOPIS +3.1pp in NE doors while ship-to-home falls.
Sessions +9% — promo working! Funnel healthy until mobile completion −7.8pp at the delivery-option step.
W27 disposition "monitor — promo noise expected" captured · W28 certified actuals reconcile at Δ 0.00%.
Delivery-option step: NE promise slipped mid-W27. Mobile completion 87.4% → 79.6%; desktop only −1.2pp.
The demo's political high point: the platform evaluates every plausible explanation including the one the room would jump to — and shows its work. Click each suspect for the verdict stamp.
For Miller (Product & Tech Strategy)"A tool that ends 'blame engineering first' war-rooms is a tool Product & Tech Strategy sponsors. Your release got accused at 11am and acquitted by the data before lunch."
The signal becomes an owned, approved, deadline-bearing action. Approve it as Adrienne — then watch the ledger close the loop. Owner ≠ approver, by design: approval can't be bypassed.
Signal fires. Breach z = −4.1 · $1.4M/wk at stake · action approved same day.
Mitigation live. Ship-from-store routing on; pickup surfaced at checkout; NE promises truthful.
Completion recovered to −0.9pp vs expected · $0.0M/wk recovered.
Attribution caveat recorded: "carrier lanes partially restored in parallel; recovery not solely attributable."
Trust beat"Read the caveat out loud. A platform that hedges its own success honestly is a platform you can believe when it claims a $1.4M problem."
Signals don't end at insight. Each got an owner, a decision, and a measured outcome in the ledger — with caveats recorded, because trust compounds. All figures [sample].
RTW digital fix live in 48h. Completion recovered to −0.9pp · ≈$0.9M/wk back · ship-from-store share holds +6pp after the disruption clears — a structural omnichannel win.Caveat recorded: carrier lanes partially restored in parallel; recovery not solely attributable.
Chase + transfer approved Monday. +$2.1M incremental demand captured vs do-nothing · sizes held through W30 · rotation window extended.Caveat recorded: rotation extended and chase landed — effects not separable.
Buyer confirms the supplier change on AQ-1147 POs → "runs small" PDP nudge + size-curve correction on next receipt. Caught ≈12 weeks before season-end discovery — ≈$1.0M margin difference.
Beauty stayed suppressed. Zero analyst-days spent proving the obvious; the event's actuals joined the comparable set — next quarter's suppression just got smarter.
Feed backfilled, checks green, chip cleared. The gap stays excluded from model training, so the model never learns that low traffic is "normal" — and the next real alert arrives untarnished.
Trust beat"Every number on this screen sits next to its caveat. A platform that hedges its own wins honestly is a platform you can believe on a Monday morning."
This isn't a dashboard upgrade. It's a different operating rhythm for Bloomingdale's: agents do the watching, assembling, and explaining — continuously — so the leader and the teams spend their week deciding and acting, not proving and arguing.
Dashboard archaeology: 40 tabs, five tools, three versions of the truth
One ranked queue, dispositioned with evidence — reading, not assembling
The meeting argues about whether it's real
The meeting starts at the fix and the chase
Analysts hand-stitch extracts; engineering defends a healthy release
Analysts validate and extend models; engineering ships roadmap
The truth arrives four days late; the good news already capped itself
The ledger reports outcomes with caveats — and the models learn the week
Attention becomes the optimized resource. Monday opens on decision #1 and #2 — not on "is it real?". Five moves, two meetings, zero chased ghosts.
3 days of arguing → 30 minutes of decidingSignals arrive at their grain — style × size, door, funnel step — with owners, deadlines, and runway: chase while sizes last, fix while the promo still runs.
Decisions on the timeline where they still matterA platform that defends its own rigor — clears releases with evidence, quarantines bad pipes before they lie, and treats feed owners as first-class users.
No more war-rooms that end in "it was tagging""Five things moved. Two were explained away with evidence, one was quarantined, one went on watch with a named owner — and leadership attention was spent only on the two that deserved it. That discipline, every single week, is the new era."
Clayface — agentic market intelligence for Bloomingdale's. Questions welcome.