The Barking Cat AI Diagnostic
How the AI Diagnostic is undertaken
For the boards of UK retail and consumer businesses from £20m to £1bn, founder-led, PE-backed and listed: an independent read of where AI would pay back in the operating model, where it would be theatre, and the binding constraint that decides which.
The Review walks a proposition outside in, the way a customer does. The Diagnostic walks the operating model inside out, the way an operator does. It separates the AI that pays back from the AI that is theatre by establishing what is true in the business rather than trusting the dashboard. Two to three days in the building, deliberately on-site, because the finished score comes from evidence, not a survey.
Before you commit to an AI programme, establish where the payback sits
92%
of retail and CPG executives plan to increase AI budgets
6%
are capturing real enterprise-level value from it
92%: NVIDIA, State of AI in Retail and CPG 2026. 6%: McKinsey, The State of AI, November 2025.
Leaders are under pressure from every direction at the moment: AI in every vendor deck, and an answer expected by the end of the quarter. Nearly everyone is spending and almost no one is winning, and the gap between the two numbers above is organisational rather than technical. There is rarely anyone in the room whose job is to separate the commercial problem from the technology being sold to solve it. That is where Steve’s job starts.
Diagnosis comes first because that is the house rule. Sometimes the finding is that you do not need what you are being sold, whether the seller is a software vendor, an agency, or an integrator with a replatform in mind. That answer is only available from an advisor who does not earn from the build.
Thirty minutes, no deck.
The method
01 · The AI Inventory
What the business has already bought, and whether it is integrated into a workflow, owned by someone against a commercial outcome, and creating value. This includes the purchases individual teams licensed that never reached a central register. Those below-the-line finds are often the most revealing evidence in the whole engagement and frequently reframe it before the scoring begins.
02 · The walk
The operating model, read across three lenses. Creation: the content, imagery, copy and studio pipeline. Agentic: the feeds, product data and machine discoverability. Process: service operations, merchandising workflow, reporting, reconciliation, and the manual work nobody has counted.
03 · The five foundations
Data, Process, People and Capability, Governance, Commercial Ownership. Each is scored Ready to Build, Contain and Prepare, or Foundations Absent. Most professionally run businesses sit at Contain and Prepare on most foundations. That is the ordinary resting place, and Ready to Build is earned rather than assumed.
01
The AI Inventory
What has already been bought, and whether any of it is genuinely used.
02
Creation
Where AI is used to make things, and what that output is worth.
03
Agentic
Where work has been handed to a system to run on its own.
04
Process
The workflow as it is actually walked, not as it is documented.
01
Data
Trusted, reachable, current enough to act on.
02
Process
Reliable enough to be automated at all.
03
People and Capability
Who can operate it once it is running.
04
Governance
Who decides, and on what basis.
05
Commercial Ownership
Who owns the payback, by name.
The finding
The binding constraint
The one foundation holding the operating model back. Everything else waits behind it.
Process appears in both rows and means two different things: the workflow observed on the walk, and the reliability of that workflow once it is scored
It narrows twice. Four evidence sources into five scored foundations, five scored foundations into one binding constraint. Neither stage is skipped. The middle is where the instrument does its work.
04 · The opportunity engine
Every AI opportunity the walk surfaced, named as a specific commercial move rather than a technology, then plotted for value against the effort to build it here. Each one is then tested against the foundation it depends on. Two businesses with an identical opportunity can receive opposite recommendations at this step, because their foundations differ. That is the mechanism, not an inconsistency.
05 · The binding constraint
Across the five foundations, the one whose absence blocks the most opportunity, named directly, with the alternatives rejected and the reasons given.
06 · The working session
Argued with the leadership team, the sequence agreed, the exclusions stated plainly, including where the answer is not a technology purchase at all. The written diagnosis stays as the record.
What the Diagnostic is not
It is not automated, and presence is the product. A remote or pre-visit read is a hypothesis layer, labelled as such throughout, never presented as a finished Diagnostic.
It is not an implementation plan, a vendor selection, or a build. The instrument diagnoses and sequences. It does not configure or supply.
It is not legal, regulatory or compliance advice. The AI Inventory records what has been bought and whether it is used. The legal and reputational risk of that use is a question for a specialist advisor.
How it connects
Every Diagnostic closes with a scope and boundary note stating what the read saw and did not see. Where a sequenced plan calls for ongoing counsel, that is raised once, in the working session, and nowhere on this page.
You can run the five foundations on yourself first: twenty-one checks, thirty minutes, nothing leaves your browser.
The same Diagnostic, set out for boards asking a simpler question first, where will AI pay back: the independent AI readiness assessment.
What happens next
Thirty minutes, no pitch. If there is work worth doing, a fixed-fee proposal in writing within two working days. Then the engagement, or a plain no.
Contact
Thirty minutes, no deck.
Tell Steve what is not moving. If there is a problem worth solving, the conversation will find it. If Barking Cat is not the right answer, you will hear that first.
steve@barkingcat.ai