Your systems know. Ours decide.
Audiences AI builds decision systems: the working layer between a business’s knowledge and its next move.
What a decision system is
Data tells you what happened. Models estimate what may happen. A decision system determines what the business should do next, why, who owns the call, and how the outcome changes the next decision.
The loop is the same whether the setting is a merchandising review or a morning production meeting.
retail
A cart held for three days by a shopper who is deep in the reviews reads as distrust. The right treatment is reassurance. A discount would cost margin and answer a question nobody asked.
manufacturing
Three idle hours coded NO MATERIAL while the order book is loaded read as a readiness failure. The fix belongs to the supplier process. The machine was never the problem.
The decision layer is what we build.
signals→ decision context→ decision→ action→ outcome
A decision may recommend, re-sequence, route, approve, re-price, or hold. What each action produces feeds the next read of context.
The two shapes we watch for
A STATE problem: the event is observed correctly while its meaning stays unknown, and the same behavior with a different underlying reason deserves a different action. A TRIAGE problem: the event is recorded correctly, and someone still has to decide whether it is worth acting on, what caused it, and where it should go.
Both shapes recur across industries, and both deploy as lightweight decision logic sitting beside the enterprise systems you already run.
missed protocol visit window · clinical trials
Patient hospitalization. Clinically consequential; goes to review.
A one-hour scheduling shift on a noncritical assessment. Documented, and nothing more.
case on tomorrow’s OR schedule · surgery
Authorization missing and recoverable by the afternoon. Chase it; hold the slot.
Clinical clearance unresolved. Out of administrative hands; a clinician reviews.
repeated HVAC work order · property
A cheap component keeps failing. Repair it again.
The asset is past its replacement threshold. Stop repairing; open a capital review.
Industries under active research. Read the longer map →
The products
One system per shape: Interpret for STATE, Triage for TRIAGE. Built first for the two rooms we know well: the category review and the manufacturing plant’s morning meeting.
Interpret · Retail & D2C
Your stack sees customer behavior. Interpret reads it in context before deciding which treatment fits, and it holds back when the honest answer is to sell nothing today.
Triage · Manufacturing
Your plant measures operating loss. Triage decides which loss is recoverable, what it is worth, and which fix fits, and it says so when the records cannot support a diagnosis.
| Interpret | Triage | |
|---|---|---|
| Input | Customer signals | Production and cost data |
| Missing judgment | What does this behavior mean? | Which loss is worth attacking? |
| Decision output | Which treatment fits | Which intervention fits |
| Guardrail | Hold, suppress, or decline to sell | Refuse, decline to route, or request better data |
Pilot-first by design. Both products begin against your own data, in a controlled and reviewable environment, before any production commitment.
The advisory practice
Some decisions fit neither product, and some businesses need the diagnosis before any build. The advisory practice covers both: custom decision systems designed around the decision itself, decision diagnostics, and the AI Decision Workshop for leadership teams choosing where to start.
Recent decision territory spans customer lifecycle and retention · promotion and margin · merchandising and assortment · buying and procurement · production and delivery · planning and scheduling · store formats and expansion · quality and complaints · AI governance and prioritization.
Where an engagement starts
Product pilot
The decision fits Interpret or Triage. Validation and data sufficiency are built into the pilot path before any production use.
Custom engagement
The decision is specific to your business. The engagement establishes what must be true before anything gets built.
AI Decision Workshop
You do not yet know which decisions AI should touch. The engagement maps them and prioritizes the first few. Scope a workshop →
Whichever route fits, no build executes on data that cannot carry the decision.
The record behind the firm
incremental revenue impact across the founder’s operating record
decisions operationalized inside large retail and banking organizations
senior data science and analytics leadership
Senior Director of Data Science, Macy’s. VP of Analytics, Citi. Columbia MS, Operations Research. HBR Advisory Council. Frameworks and outcomes are real. Client names are kept anonymous by design.
What Audiences AI does today
Two decision-system products, Interpret and Triage, available for pilot. And an active advisory practice: custom decision systems, decision diagnostics, and AI Decision Workshops for leadership teams deciding where to act first.
Every build carries the same working discipline: every inferred figure carries its confidence, refusal is a designed outcome with its own deliverable, and automation is offered only after a human acceptance record earns it.
From the articles
-
Two shapes, twenty industries
For most of this year we described our work as two products: a decision layer for retail demand signals and a triage system for the plant floor. In August we sat down to map…
-
Recoverable is a verdict
The downtime sheet that opens the morning production meeting reads like a settled matter. Equipment failure sits at the top of the loss column, large enough that nobody argues…
-
Fourteen Pages, No Holes Showing
The deck arrived on a Sunday night with a one-line message: would love your read before Thursday. Fourteen pages. Five recommendations. A founder I have known for years had…
Bring us a decision.
The one that keeps returning to a meeting, a spreadsheet, or someone’s judgment.
Thirty minutes to understand the decision and determine whether a product pilot, a custom engagement, or no engagement at all is the right next step.