Enterprise Intelligence Group
Your AI agents need an operating model, not another collection of pilots.
An operating model for the AI agents a retailer already runs: shared objectives, decision rights and precedence rules, from an independent, technology-agnostic adviser.
Retail AI Operating Model
EIG helps retailers, CPG companies and distributors align agent roles, decision rights, shared knowledge and executive strategy across merchandising, supply chain, pricing, marketing and operations, so local automation compounds into enterprise performance.
Book the Executive Agent Coordination Review
Not ready? Take the free AI Maturity Assessment
An open working session with an independent adviser. No platform to sell, no deck to defend.
The argument
A faster silo is still a silo.
Retailers spent a decade fixing the integration problem. Point solutions per function. Data that would not move between them. Years of work to make one number mean one thing in two systems. We are now rebuilding that problem, much faster, with assistants.
Every vendor you already pay is an AI company now. Individually most of their agents work. Collectively they hold slices of your company's judgment in formats you do not own, behind roadmaps you do not control. Shared data was never the hard part. Shared understanding is. Small pockets of enterprise SaaS may deliver quick wins now; they will not enable success in the future.
The EIG proposition this page serves: retailer, CPG and distributor coordination. The intelligence that passes between merchants, planners, suppliers, warehouses, brands and stores, not just the data inside one department.
A faster silo is still a silo
Pockets of AI are the new silos. Without common objectives, decision context and rules for resolving conflicts, disconnected agents reproduce the silos of disconnected enterprise software, only faster.
Every vendor's agent holds a slice of your judgment
Planning, ERP, category management, pricing, workforce, commerce: each system you already pay for has grown an agent. Each one holds part of how your business decides, in a format you do not own, behind a roadmap you do not control.
Nobody sells the coordination layer
Every vendor supplies endpoints. Not one supplies your objectives, decision rights, precedence rules or the threshold at which an agent must stop and a person decides. That is the operating model, and it stays yours whether you set it deliberately or by accident.
The private-label launch
When six AI agents touch one retail decision.
A retailer can deploy six excellent AI agents and still make a worse decision. Consider a private-label launch.
Sees the assortment opportunity and identifies the whitespace.
Predicts margin and elasticity and sees an attractive price point.
Forecasting agent
Estimates demand and the cannibalisation of the lines already on shelf.
Finds the vendor capacity and distribution-centre constraints.
Tests shelf feasibility and finds the product will not fit consistently across the estate.
Marketing and stores agent
Plans the investment to stimulate demand and flags the execution limits in store.
Every answer can be technically correct. The combined recommendation can still work against the retailer's strategy.
Individually each recommendation is plausible. Collectively they produce a bad decision unless the agents share the same strategy, assumptions, constraints and escalation rules. It is no longer enough to put an agent inside every application or department.
AGENTS NEED MORE THAN COMMON DATA
- Shared strategic objectives
- Decision context and commercial judgment
- Defined roles and precedence
- Rules for challenge and escalation
- Cross-functional decision memory
- A named human accountable for the outcome
The model
One intelligence layer the whole business decides from.
Your vendors' systems below. Agents above. Between them, the layer you own: what the business is trying to achieve, who may decide what, the knowledge every agent must share, and the rule for when two agents disagree about the same shelf. Decision rights and earned autonomy are the rails it runs on.
ENTERPRISE INTELLIGENCE LAYER
- Objectives and trade-offs
- Decision rights
- Shared knowledge and precedents
- Precedence and escalation rules
- Decision memory
YOUR VENDORS' SYSTEMS
- Category management
Earned autonomy
The Retail AI Operating Model. Agents keep their specialist jobs. The layer between them carries the objectives, decision rights, shared knowledge, precedence rules and decision memory that no single vendor supplies. Autonomy is earned per decision, not granted per tool.
Where it sits
Not governance. Not orchestration.
Governance and orchestration are chapters inside the operating model, not the title. Both matter. Neither answers the question the leadership team owns.
PRIMARY CONCERN
Risk, policy and compliance
Technical orchestration
Routing, integrations and model execution
THE OPERATING MODEL
EIG Retail AI Operating Model
Which decisions matter, how agents cooperate, what judgment they require, and who remains accountable
How we work with you
Two offers, and the method behind them.
One decision before one platform. The path starts with a conversation, moves through a review of the agents you already run, and ends with one workflow instrumented and moving toward production.
Executive Agent Coordination Review
The natural first meeting. A working session designed for an executive team to have an open discussion and take direction and advice from an independent adviser: where your agents already are, where they will disagree first, and which decision is worth coordinating before any platform is chosen. Lighter than the Sprint, and the place most conversations begin.
Book the Executive Agent Coordination Review
Executive Agent Coordination Sprint
The entry engagement. It opens with an Executive Agent Coordination Review of the agents, human roles and SaaS platforms you already run, then produces seven working assets. The Sprint ends with one instrumented cross-functional workflow moving toward production, not a set of documents.
THE SEVEN DELIVERABLES
- 1 Executive AI Strategy and Value Thesis: the enterprise outcomes agents must serve, with explicit priorities and trade-offs
- 2 Enterprise Decision Architecture for coordinated retail AI: a cross-functional map of important decisions, dependencies, inputs, owners and failure points
- 3 Agent Portfolio and Coordination Map: existing and planned agents, human roles, SaaS platforms, overlaps, gaps and required handoffs
- 4 Shared Enterprise Intelligence Model: policies, thresholds, commercial judgment, precedents, assumptions and operational context agents must share
- 5 Decision Rights and Autonomy Matrix: what each agent may observe, recommend, prepare, execute or escalate, and under which conditions
- 6 Coordination and Exception Playbooks: precedence rules, conflict resolution, challenge protocols, human escalation, audit and rollback
- 7 Value Scorecard and 90-Day Activation Roadmap: one prioritised cross-functional workflow, baseline measures, instrumentation and sequence
Enterprise Decision Architecture for coordinated retail AI
The signature method inside the Sprint. It maps the decisions that matter across functions, with their dependencies, inputs, owners and failure points. It sets the decision rights and autonomy matrix: what each agent may observe, recommend, prepare, execute or escalate, and under which conditions. And it writes the coordination and exception playbooks as concrete handoff rules, not as a promise that agents will collaborate.
AUTONOMY, EARNED PER DECISION
One agent can hold a different level on each decision it touches. The level is set by the cost of a wrong call.
PLAYBOOKS WRITTEN AS HANDOFF RULES
- Conflict resolution
- Human escalation
For example: when the forecasting agent and the margin agent disagree about the same line, which one takes precedence, at what threshold the planner is called, and who signs the outcome.
WHO THIS IS FOR
- Chief Operating Officer
- Chief AI, Digital or Transformation Officer
- CIO or Chief Data Officer
- Chief Merchandising Officer
- Chief Supply Chain Officer
- CFO, as value and risk sponsor
- CEO of a mid-market business
The board is the audience and a governance stakeholder, not the day-to-day buyer.
WHAT WE MEASURE
Coordinated AI is measured on outcomes across functions, not on the accuracy of any one agent. Baselined before the workflow goes live.
- Conflicting-recommendation rate
- Exception backlog
- Time-to-decision
- Manual reconciliation
- Override quality
Where it gets applied
An agent in a silo is still a silo.
Coordinating agents around shared objectives, decision precedence and evidence is where the operating model gets implemented, one live decision at a time. The Retail Agent Coordination Review is that working session: you bring the workflow and the people who own it, and you leave with the map.
Retail Agent Coordination Review
The landscape
What the vendor landscape tells you.
Read from the vendors' own published documentation, not from analyst percentages. Five implications for a retailer, CPG company or distributor deciding what to buy next.
The reasoning engine is becoming a commodity.
AWS, Microsoft and Google all offer Claude, GPT and Gemini side by side, and the open-weight options are self-hostable. The model is now a procurement choice you can change. The intelligence layer you own is not.
Openness is answerable, vendor by vendor.
It can be read from published documentation, and it differs by direction. Some retail planning and execution vendors publish exactly how an external agent connects to their systems, in both directions. Others publish nothing at all for their agents. Ask before you buy.
The protocol layer is neutrally governed.
MCP and A2A are Linux Foundation projects with AWS, Google, Microsoft, Salesforce, Shopify and Snowflake as members. Requiring them of a vendor is not a bet on one supplier. It is a bet on not being locked to one.
Your data platform is already the pattern.
Snowflake and Databricks expose governed MCP access enforced by the permissions you have already set. That is proof that one layer, centrally governed, can serve many agents.
Nobody sells the missing piece.
Every vendor supplies endpoints. Not one supplies your objectives, decision rights, precedence rules or escalation thresholds across them. That is the operating model, and it stays yours.
The test to apply before buying any of them: can my own agent call your system? Eight questions for your SaaS provider is the one-page version, free in the Library.
Why EIG
Independent by design.
EIG is technology- and software-agnostic by design. We are not backed by a software company, we are not tied to a particular LLM, and we have no orchestration platform to sell. So the answer to which decisions matter and who stays accountable is never bent toward a product.
Retail, supply-chain and CPG knowledge from the operator's side of the table: extensive enterprise software implementation and programme delivery experience in retail and CPG, on both the retailer side and as an enterprise software and consulting partner, and a working knowledge of what an agent's recommendation costs when it is wrong.
The work is complementary to the SaaS, data and agent investments you have already made. We do not replace your vendors' agents. We decide, with you, how they work together.
Technology- and software-agnostic
By design. The recommendation is never bent toward a product we need to sell.
Not backed by a software company
No vendor funds this work, so no vendor sets its conclusions.
Not tied to a particular LLM
The model is a procurement choice. We help you keep it one.
No orchestration platform to sell
We decide how your agents work together. We do not sell the thing that runs them.
Board Briefs
For the leadership table.
Nine briefs on the Retail AI Operating Model. Free, no sign-up.
One executive question per page: what a leadership team has to settle before AI agents touch the decisions that run the business. Who owns which call. What an agent may do alone. What proves it worked. Ready to take into the next meeting.
All Board Briefs in the Library
The enterprise intelligence layer
From pilot sprawl to coordinated enterprise intelligence
More pilots do not equal more maturity. Coordination determines whether value compounds or cancels. Four recognisable states, each earned by fixing what the previous one lacks.
Your enterprise strategy is missing from the prompt
The board approved the priorities. The merchandising, supply, pricing, space and marketing agents have never read them, because nobody translated the priorities into rules and trade-offs an agent can use.
When two AI agents disagree, who wins?
The forecasting agent recommends more inventory. The margin agent recommends less. Both are technically correct. Without a rule, the answer is whichever system ran last, and a planner reconciling by hand on Friday afternoon.
Frequently asked questions
Is this AI governance?
No. Governance is a chapter inside the operating model, not the title. AI governance asks whether an agent is permitted to act. The Retail AI Operating Model asks which decisions matter, how agents cooperate, what judgment they require and who remains accountable. Risk, policy and compliance sit inside that, alongside strategy, coordination and adoption.
Do we have to replace our vendors' agents or buy an orchestration platform?
No. The operating model is complementary to the SaaS, data and agent investments you have already made. It sets the objectives, decision rights, shared knowledge and conflict rules those agents run under. Whether a given vendor can take part, for example whether your own agent can call its system, becomes a question you can answer from published documentation before you buy.
What does the Sprint actually end with?
One instrumented cross-functional workflow moving toward production, with its baseline measures, decision rights and handoff rules in place. The seven deliverables are working assets you keep. The point of them is the workflow, not the paperwork.
How is the work scoped?
We scope each engagement to your situation and agree the details together before we start. The Executive Agent Coordination Review is the natural first meeting. The Sprint follows when there is a decision worth coordinating. Terms are discussed in conversation, not published.
When is the right time?
The strongest trigger is several pilots, vendors or functional initiatives running with unclear coordination between them. An organisation with no AI activity yet is usually better served by the free AI Maturity Assessment and a single decision mapped end to end.
Next step
Start with one decision your agents already disagree about.
Availability and margin. Assortment and space. Forecast and supply. Bring the one that is being reconciled by hand today, and leave the session with a view of who should decide it, what the agents need to share, and what to instrument first.
Book the Executive Agent Coordination Review
Take the free AI Maturity Assessment
No pitch. A working session with an independent adviser.
The EIG Library · Insights · AI Maturity Assessment · Contact