AI-native operating decision infrastructure for retail

From messy retail data to margin-protecting decisions.

Your planning systems recommend. Your dashboards report. Outturn starts from the fire they already show, assembles the context, routes the decision to a named owner, and measures whether the action paid — while exposing where ownership, approval or execution leaks.

Request a 4-week pilot See the decision loop

every number computed in code — and cited
a named human owns every decision
impact measured, never assumed

ACTIVE DECISION · RD-0042 · LIVE

Stockout risk — Harper poplin shirt · Dallas NorthPark $45,288 sales at risk — decision routed, awaiting the owner's yes

Detected
Context reconciled
Scenario prepared
Routed
Impact measured

SIGNAL Dallas sold its last unit — was selling 18 a week 0 on hand ↳

SIGNAL Houston sits on 240 units it sells 6 a week 40 wks of stock ↳ same region, one courier run away

SCENARIO Move 54 units Dallas sells that Houston won't $45,288 ↳ 18/wk × 17 weeks left × $148 — the transfer executes in your allocation tool

ROUTED Allocation owns it · buyer notified — context attached 1 yes away

Demo-seeded figures, not customer data.

The problem

The gap is not visibility. The gap is decision-to-execution leakage.

Retail teams see the fire in reports. The margin leaks in everything after visibility: gathering the right context, deciding what to do, routing it to whoever can act, tracking that they did — and knowing whether it worked.

Context is scattered

The six signals a decision needs live in BI dashboards, Excel files, planning tools, Teams threads, emails — and someone's memory. Assembling them is the weekend job nobody invoices.

cover in one tool · open POs in another · the vendor's delay in an inbox

Decisions are unstructured

The critical calls happen in side conversations and Monday meetings. The rationale never becomes data — so the same fire gets re-argued from scratch next season.

"didn't we decide this in March?" — nobody can check

Impact is not measured

Which decisions were delayed, missed, or executed? Did the transfer protect availability? Did the markdown protect margin? Almost no team can answer with numbers.

approved ≠ executed ≠ worked

Everyone sells you visibility. The margin leaks after visibility.

How it works

One loop. Five stages. Every fire, every time.

Every issue becomes a decision record — RD-#### — that moves through the same lifecycle, with the money attached at every stage.

Detect / ingest

An existing report or a bounded deterministic check creates the record and preserves when the issue became visible. Separate monitors surface the leakage after visibility: nobody accepts ownership, an approval goes overdue, an approved action never starts, or the outcome is never measured.

Reconcile

The context that lives in six places lands on one card: stock cover, sell-through against plan, open orders, supplier lead times — and the vendor email that quietly moved the truck. Every signal cited to its source row or document. Nothing asserted without a receipt.

Scenario

Options prepared before anyone asks, from your own playbook: open a transfer scenario, request a buyer review, prepare a markdown review, accept the risk and monitor. Each option carries its arithmetic — never a black-box score.

Route

Decisions get owners, not threads. The buyer approves reorders; allocation runs the transfer; store ops confirms the floor. Routed with the full context attached, tracked until the action actually happens — in your systems, with a named approval.

Measure

Did it pay? Recovered sales, availability, markdown avoided — measured on your numbers and written back to the record. Delayed and missed actions surface instead of vanishing. The graded history of every call becomes an asset your team owns.

Inside the product

An operating layer, not another dashboard.

Three surfaces turn retail signals into decisions with owners, actions, and measured commercial impact.

Retail Fire Boardevery detected issue, decision request, and execution status

Week 287 active

Detected

Harper poplin — Dallas stockout risk$45.3K sales at riskNew

Aria linen blazer overstock$68K markdown exposure

Context reconciled

Sutton rebuy — delivery slippedvendor email + open PO6 signals

Scenario prepared

Nova midi dress — dead size runclear vs. consolidate vs. hold3 options

Routed

Lyra wide-leg denim — rebuybuyer approval pending1.2d

Action in progress

Harper transfer — 54 units to Dallasexecuting in your allocation tool

Impact measured

Camden trench — second markdown$36K kept vs. outlet binClosed

Demo-seeded figures, not customer data.

Why is the Harper shirt flagged?

Decision agent · grounded in RD-0042

Dallas NorthPark sold its last unit and was selling 18 a week. Houston Galleria, in the same region, holds 240 units — 40 weeks of stock. Moving 54 units protects $45,288 of season sales; no reorder needed.

Who should decide?

Decision agent · grounded in RD-0042

Transfers are allocation's call, and this one crosses your taught $10K line — so the buyer is notified too. I can route it to allocation with the full context attached.

Grounded in one record: the copilot explains and routes. It never detects, and it never invents a number — every figure above is computed in code and opens to its rows.

RD-0042 — Stockout risk · Harper poplin shirt · Dallas NorthParkdetected Sun 23:41 by never_ship_a_stockout@v3 — your rule, your threshold

$45.3K at risk

Detected
Context reconciled
Scenario prepared
Routed
Impact measured

Context — 4 signals, each cited: Dallas: 0 on hand, 18/wk · Houston: 240 units, 6/wk · no open PO · courier run $150
Options — your plays, with the math: Transfer 54 units ($45,288 protected) · reorder (lead 2 wks — too slow) · accept & monitor
Owner & routing: Allocation owns it · buyer notified over the $10K line · store ops confirms receipt
Impact — written back when it lands: Recovered sales, availability, days-to-execution — measured on your numbers, kept forever

Similar past case: RD-0017, June — same store pair, transfer executed in 2 days, sales recovered. The graded history is the asset your team keeps.

Connected

Plugged into every system where retail decisions happen.

An agentic merchant that works in the tools your allocation, merchandising, and buying teams already use — planning, ERP, BI, spreadsheets, chat, email. It reads the context, prepares the decision, and routes it to the owner in the tool where the work actually happens.

SAPSAP OROracle Retail RxRELEX BYBlue Yonder BIPower BI TbTableau LkLooker SfSnowflake BQBigQuery DbDatabricks XExcel GSGoogle Sheets SPSharePoint TTeams #Slack @Outlook MGmail JJira AsAsana NNotion CfConfluence SFSalesforce POSPOS / Store SShopify MgMagento EDISupplier EDI

Check Power BI sell-through for the Harper poplin shirt, reconcile with RELEX stock cover and SAP open POs, prepare a transfer scenario, and route it to the buyer in Teams with the sales at risk.

Attach: decision memory, past transfers, buyer notes…

Illustrative — these are integration targets, not claims. The pilot starts from exports and the tools you already have; deeper connections come after the loop proves itself.

Where we sit

Not a forecasting engine. Not a dashboard. Not a task tracker.

Category	Where others sit	Outturn — where we sit
Forecasting / planning systems	They recommend what should happen.	We capture what teams actually decide when reality deviates from the plan.
BI dashboards	They show what happened.	We turn issues into decisions, actions, and measured outcomes.
Workflow tools	They track whether a task was completed.	We track whether the commercial decision worked.
Generic AI assistants	They summarize information.	We understand retail decision context, KPI mapping, ownership, and outcome attribution.

We do not replace planning or allocation systems. We sit above them as the active decision layer.

Built-in retail intelligence

It arrives knowing the workflow. You confirm what makes your company different.

Every stockout-response pack starts with cited metric definitions, required signals, common actions, ownership patterns and outcome methods. Outturn calibrates the defaults on your data; your team only corrects the parts that truly differ.

✎

Industry definition

The pack begins with a cited definition of the decision, the minimum context required and the commercial outcome to measure.

✎

Segment default

Apparel, grocery and specialty retail do not use identical signals or deadlines. The pack starts from the relevant segment pattern, not a generic prompt.

✎

Deterministic calibration

Your history calibrates thresholds and deadlines in code. The record shows the definition, formula and evidence used — never an unexplained model score.

✎

Retailer correction

If your sell-through includes outlets, Texas transfers use Dallas first, or reorders over $50K need two approvals, your correction becomes a versioned company policy after review.

proposed policy Policy proposal: increase minimum donor coverage from 82% to 88% — based on two negatively affected donor stores, with replay, trade-off and affected scope shown before approval.

A policy proposal must explain why: the Decision Records that triggered it, how those cases replay under the new rule, the expected benefit and downside, where it would apply and what the evidence cannot yet prove. Nothing changes until a human approves it.

Honest AI — our favorite sales pitch

No black-box forecasts. No autonomous agents inside your systems. The math is published.

AI models are not reliable reasoners — they are excellent accelerators and researchers. So no model ever computes your money. It appears exactly three times: reading and reconciling messy sources, proposing scenarios and rules a human adopts or edits, and narrating results after the numbers are final.

The part that earns trust: the layer knows when not to act.

ON HOLD Miller graphic tee — don't decide on bad numbers $32,640 ↳

NEEDS A HUMAN Vega crop cardigan — no rule explains it $22,080 ↳

210 styles trading to plan — nothing needed, nobody paged ✓

read-only queries, enforced zero model calls in the decision math human-approved writes append-only audit log

"AI can't reason" is a contested claim, and we don't make it. Ours is narrower and holds up: today's models are not reliable reasoners — so we keep them out of the loop that computes your money entirely. Call it an agentic merchant if you like; we call it code you can audit.

ROI

ROI starts with decision leakage.

The stockout is the source issue. We measure what fails after it becomes visible — ownership, context, approval, execution and outcome — then close that leakage.

Sample 4-week pilot Action completion 68% → 84%

78issues detected

42decisions required

16delayed or missed actions

$240Ksales at risk

$78Krecovered or protected margin

31hmanual follow-up saved

Illustrative sample — not customer results, and we'll never pretend otherwise. Your report is computed from your own sales, inventory, margin, and demand assumptions.

Commercial risk proxy

sales_at_risk = recent_sales_velocity
  × expected_stockout_gap
  × selling_price

gross_margin_at_risk = sales_at_risk
  × gross_margin_%

Grounded in your existing sell-through, cover, and margin assumptions — not a black-box forecast. If you can't recompute our number from your rows, we've failed.

The 4-week decision-leakage pilot

Start by measuring the leak.

No heavy IT integration. Week one starts from existing reports, Excel exports, and the Teams threads where your decisions already happen.

Week 1 — map your store-stockout response workflow and connect the signals
Week 2 — open live stockout records, assemble cited context, stand up the board
Week 3 — your operators run the workflow; Outturn observes product gaps without joining the daily decision loop
Week 4 — continue vendor-independent operation and review decision lead time, delayed or missed actions and agreed commercial outcomes

We publish metric definitions, not invented results — the first pilot numbers you'll ever see from us will be your own. Built by an operator who has built and run a production retail-ops AI system; independent and clean-room. No logos or outcome claims here until a customer earns them and consents in writing.

Name

Please enter your name.

Work email

Please enter a valid work email.

Company

Please enter your company.

Role

Please tell us your role.

Tell us about the last fire that cost you margin — what was decided, who owned it, and did anyone measure the outcome?

A sentence or two helps us prepare a relevant pilot.

We only use this to evaluate fit and reply — from a real person, not a sequence.

FAQ

What your operators — and your security team — will ask.

Is it autonomous? Does it act inside my systems?

No — by design. Outturn ingests the visible issue, assembles the context, prepares options and routes a human-owned decision. The approved action executes in your existing systems: the transfer runs in your allocation tool, the reorder goes to your buyer. We record it to completion; we do not improvise actions through your systems.

Do you forecast demand?

Never. The risk math is published arithmetic over your own sell-through, cover, and margin assumptions — not a black-box forecast. If your planning system forecasts, we happily read its output as one more signal. We compete with nobody's demand model.

How do I know it's not hallucinating my numbers?

Because no model computes them. Ingestion checks, leakage monitors and decision math run as deterministic code; every number carries its formula and opens to its source rows. Models may extract from messy documents or draft a proposal. Extracted, conflicting or unusual facts require review.

Is this another dashboard?

No. Dashboards show what happened; the leak starts after visibility. Outturn turns each issue into a decision record with an owner, options, a deadline, and — the part nothing else does — a measured outcome. It's an operating layer: things move through it, they aren't just displayed on it.

Does it replace our planning or allocation systems?

No — it sits above them and makes them worth more. Their recommendations become signals on the record; their screens are where approved actions execute. We also catch the seams between them: the allocation built on a delivery that slipped, the plan that disagrees with the warehouse.

Do we have to teach it everything first?

No. It arrives with a Retail Workflow Pack: cited metric definitions, required signals, deterministic checks, common response options, ownership patterns and outcome measures. Your team confirms only company-specific differences. Approved corrections become versioned policy; they are not prompt memory.

What data does it need to start?

What you already have: the stockout or availability report plus sales, stock, inbound and transfer exports at the relevant product-location grain. Authoritative structured fields flow through governed contracts; forwarded emails and notices are cited but reviewed when they introduce a new or conflicting fact. No heavy integration for the pilot.

Where does our data go?

Per-tenant isolation, your choice of EU/US residency, a zero-retention option, and no shared models trained on your data. Read-only by default; writes leave as drafts with a named approval. Full audit trail, kill switch. We'll run a security review with your team before any data is connected.

What does it cost?

The four-week pilot is $6–8K for one stockout/action-leakage workflow, normally creditable against the first annual subscription. After that: a predictable, capacity-based subscription, not token pass-through.