What is an AI audit and what does it involve?

An AI audit is a structured assessment of where AI would pay back in a specific business. It maps the workflows where hours and revenue leak, scores each candidate on the hours saved or revenue recovered against the cost to build, and names the first build with a number attached. At Blash the audit starts free: a 30-minute call in which you describe the business and the workflow that hurts, followed within three working days by a written audit of your highest-payback workflows, yours to keep even if you walk away.

How does an AI audit start?

The starting point is a 30-minute call, and it carries no obligation. You describe the business, the tools your team uses and the workflow that hurts most, and you get a plain answer on where AI pays back and where it does not. Nothing on the call is a pitch; its purpose is to gather enough detail for the written audit that follows.

Within three working days of the call you receive that written audit, free whether or not you go any further. If an engagement then opens, the audit deepens into two weeks spent inside your real workflows, data sources and bottlenecks, producing a full prioritised shortlist with the same scoring applied to live evidence rather than to a description.

What does the written audit contain?

The free written audit is short by design. It lists the automation candidates drawn from what you described, attaches a score to each one built from the monthly cost of doing the work by hand and the revenue the workflow currently loses, estimates the build effort each would take, and closes with a recommendation: the workflow to fix first, why it comes first, and the result you should expect to be able to measure.

It also says what should stay manual. Some workflows depend on judgement at every step, some lack the volume to repay any build, and some are covered adequately by an off-the-shelf subscription at a fraction of the cost. Naming those is part of the job, because an audit is only worth having if its no can be trusted as firmly as its yes.

How are the candidates scored?

The scoring rule is payback divided by build cost. The numerator is the monthly price of leaving the workflow alone: the loaded value of the hours it absorbs plus the revenue it leaks, such as leads that go cold while they wait for a reply. The denominator is the effort to automate it, driven by the systems involved, the state of the documents and the governance required. Our guide to costing a manual process shows how to assemble the numerator yourself.

Ranking by that ratio produces a different list from ranking by excitement. The candidate that scores highest is usually mundane: an inbox, a document pile or a re-keying loop that burns hours every week and follows rules a build can capture. Impressive-sounding candidates with vague payback sink down the list, which is where they belong.

What access does the auditor need?

For the free written audit, the call is enough: a clear description of the workflow, the tools it crosses, roughly how many items move through it and how many people touch it. The two-week engagement audit needs to see the work as it really runs, which means time with the people who do it, a sample of the documents and messages involved, and a view of the systems the workflow passes through.

What the auditor does not need is administrator credentials, a copy of your customer database or standing access to production systems. Observation and samples answer the audit questions, sensitive material can be redacted, and the work sits under NDA and your own data-handling policy. An auditor who asks for broad access before scoping anything has the sequence backwards.

How is an audit different from a strategy deck?

A strategy deck describes what AI could do for a business like yours; an audit commits to numbers for the business you actually run. The test is whether every recommendation carries a figure you can check, and a good audit refuses to recommend any build it cannot attach a number to. Where a candidate has no measurable baseline, the honest recommendation is to measure first.

A Midlands manufacturing group shows the difference in practice. The business had candidate use cases across procurement, finance, sales and operations and no way to choose between them. Sixteen workflows were scored on payback and risk, £18k of monthly manual-process cost was identified across the priority group, and three priority builds made the first roadmap, all inside a written audit delivered in 2 weeks. The Operations Director's verdict: it separated the useful workflows from the interesting distractions.

If you take one thing from this guide, take the scoring rule: payback divided by build cost, applied to every candidate. Any supplier can run that arithmetic for you, and any audit that arrives without it is a sales document under a different name.