Candor Limited
Return on AI, evidenced.
Izam Ryan MBA FCA helps UK mid-market businesses whose peers are pulling ahead get a return on AI they can evidence. Say it plainly. Show the working.
What we do
For businesses whose AI spend hasn't shown up in the numbers yet.
Before the spend, the assessment shows where the return is, in pounds, ahead of anything being built. Where the spend is already running, it finds what was actually realised and what was saving on paper only.
Where AI will pay back, or already has, in pounds. Four layers, read in order: what the business sells, who holds the power, how the work gets done, and what that does to the numbers. It is built from the work, task by task: the hours it takes, what a machine could take from it, and what that is worth.
It ends in one number in pounds, with a list of what may be given to a machine and what never should, and the steps in the order the number sets.
The work the assessment cleared, built with partners, with the evidence trail designed in from the first step: every output carries its source, every step is recorded, and someone who was not there can re-perform the result. If a machine cannot show its working to that standard, nobody can stand behind what it produced.
The saving measured against a baseline taken before the build: the hours and cost the work no longer takes, checked at set points rather than assumed on paper. Alongside it, the changes in how people work that make the saving stick.
How it is governed
Never delegate what you cannot reperform.
Six questions before anyone relies on an AI result. Their initials spell the name.
- 1
Consequence
Should AI touch this at all? What would a wrong answer cost, and could it be taken back?
- 2
Accountability
Who is answerable for it, across every supplier in the chain, and who can switch it off?
- 3
Non-discrimination
Is it fair to the people it affects, tested before it goes live and watched after?
- 4
Data and model provenance
Can every input, and the model itself, be traced to a source, an owner and a licence?
- 5
Operational integrity
Is the system tested, secure and watched for drift while it runs?
- 6
Reperformance
Can what it did be disclosed, explained and repeated by someone who was not there? Re-perform means a second person can repeat the work and get the same answer.
Work
The work, where you can see it
What is live, and what is on request.
The questions we would ask: where AI moves the margin, AI readiness, relying on an AI result, and where the work leaks. Each runs in your browser and ends in a scored read-out. Nothing is recorded unless you choose to send your answers.
What the filed accounts of 25,907 South East companies with 10 to 999 employees say when you add them up. Public registers only, aggregate figures only, and the same snapshot gives the same answer.
First edition, snapshot 1 September 2026. On request.
What a system has to do before anyone can rely on its figures: nothing overwritten, everything dated, and every number with a trail back to its source. A short explainer for a reader who is not technical.
On request.
Send me the question you cannot answer yet.
One address. I read and reply to every email myself.