Diagnostic · Before the AI spend, not after
What AI is already being used in your business, what it is worth applying to next, what it would cost, and which of it you should not do — ranked, costed, and yours to execute with anyone.
Most owner-led businesses arrive at AI from the wrong end: a tool gets bought, someone is asked to find a use for it, and a year later nobody can say whether it saved anything. The question worth answering first is narrower and duller. Where in this specific business does probabilistic software do work a person is currently doing by hand, and what does it cost to get it wrong there?
That question has a real answer, and it is usually shorter than people expect. Some of the repeated work is a process problem that automation would only make faster. Some of it is genuinely well suited to a language model. Some of it should not be touched, because the cost of a confident wrong answer lands on a client, a patient, or a regulator.
This audit produces that ranking, the evidence behind it, and the number attached to each item. It is deliberately decoupled from the build: the plan is written so you can hand it to an internal team or another firm.
Fit
Symptoms
Scope
What tools people are actually using, what information goes into them, and under whose account. This is almost never nothing, and it is almost never what the owner expects. It is the single fastest way to find the risks worth closing this month.
Every candidate task ranked by frequency multiplied by time multiplied by the cost of getting it wrong. The same ranking the automation work uses, because the honest answer for a given task is often a deterministic script rather than a model, and the ranking is what makes that visible.
Whether the output can tolerate being probabilistic, whether a wrong answer is recoverable, whether a person is reviewing it anyway, and whether the input is even available in a form a model can read. Tasks that fail these tests are named as such rather than quietly dropped.
Build effort, licence and usage cost at your actual volume, and the review time the design implies. An automation that needs a person checking every output has a labour cost, and it belongs in the business case rather than in a footnote.
Which information may go to which vendor, what has to stay inside systems you control, and what your existing obligations already require — professional conduct rules, privacy law, and the client agreements you have signed. Written as decisions your team can follow, not as a policy nobody reads.
Two or three items worth doing first, in order, with what each requires and what it should produce. Plus the list of things not worth doing, which is usually the more valuable half of the document.
Process
What tools are in use, by whom, with what information. Interviews with the people doing the work rather than a survey of what the owner believes is happening.
Repeated tasks ranked by frequency, time, and the cost of an error. This is the same ranking the automation work uses; AI is one of several possible answers to it.
Each candidate assessed for tolerance of probabilistic output, recoverability of a wrong answer, review burden, and whether the input exists in a usable form.
Ranked recommendations, costs including running cost, the data boundary, and an explicit not-worth-doing list. Delivered as a document, walked through on a call.
Outcomes
These are the kinds of change this work produces. No numbers appear here, because no client result has been measured and published yet. When one has, it will appear in Work with its baseline, method, and time window.
Tooling
Categories of use, not endorsements. Where a deterministic rule is cheaper and more testable than a model, the plan says so.
Where this sits in the system
After this: the items that survived the ranking get built, with review points and a fallback path.
AI automationAlongside this: most of what looks like an AI opportunity is a process problem, and process work is the cheaper fix.
Business automationBefore this: if inquiries are being lost between the website and the CRM, that is the more urgent diagnosis.
Growth system auditStraight answers
If yours is more specific, put it in the audit form.
support@ariadne.fyiThe growth system audit diagnoses the path from a stranger finding you to revenue being recorded, across nine checkpoints. This one looks at the repeated work inside the business and asks where probabilistic software belongs. They overlap at automation and measurement. If you are unsure which you need, start with the growth system audit — it is the broader diagnosis and it will tell you whether AI is anywhere near your actual constraint.
Canada has no AI-specific statute in force for private businesses, so what applies is what already applied: privacy law, your professional obligations, and the confidentiality terms in your client agreements. A short written decision about what information may go where usually satisfies more of that than a long policy document. If you want a recognised structure to grow into, NIST’s AI Risk Management Framework is voluntary and free to use, and ISO/IEC 42001 is the certifiable equivalent.
Then that is the finding, and it is worth knowing before a project starts rather than after. It is rarely the answer for a whole business, though. More often two or three narrow tasks are clearly suitable, a larger number are marginal, and the remainder are either process problems or genuinely require judgement.
Only where a specific tool is the answer, and the reasoning will be in the document. Ariadne holds no reseller relationships and takes no vendor commissions, which is the only reason a tool recommendation from anyone is worth reading.
It depends on the size of the business, how many people touch the repeated work, and how much of it is already documented. Scope and price are confirmed in writing before anything starts, and if it is not a fit you will be told at that point rather than after an invoice.
These first-party sources support the platform, policy, and regulatory statements on this page. Ariadne’s recommendations and planning heuristics are identified separately.
Your practical starting point
The audit inventories what is already in use, ranks the opportunities by cost and risk, and names the ones not worth doing. You keep the plan whether or not you build it here.