The Big Four's AI Trust Problem: What the Reporting Scandals Mean for UK Clients

Ask a management consultant today how much of their work involves artificial intelligence and the honest answer, for most large firms, is: nearly all of it. Research on AI credentials and adoption tends to converge on a similar figure — somewhere north of three-quarters of UK consulting firms now say AI is woven into their day-to-day delivery, not sitting in a pilot programme off to the side. That shift happened faster than most clients realised, and faster, it turns out, than some firms' own quality controls could keep up with.

The clearest illustration came out of Australia last year, when Deloitte's local arm found itself explaining to a government client why a report it had been paid to produce contained fabricated references and errors traceable to generative AI tools used somewhere in the drafting process. The firm eventually settled the matter. It was an uncomfortable moment for an industry that sells judgement and rigour as its core product, and it wasn't an isolated one — similar stories, with varying degrees of severity, have surfaced across the sector over the past year, enough that clients commissioning advisory work are starting to ask a question they rarely asked five years ago: who actually checked this?

Why AI errors are harder to catch

That question matters more than it might first appear, because the mechanics of AI-assisted consulting work make errors genuinely harder to catch than the mistakes a junior analyst might make. A hallucinated citation looks exactly like a real one until someone goes and checks it. A plausible-sounding market statistic, generated rather than sourced, reads no differently on the page than one pulled from a genuine industry report. Traditional quality-review processes at consulting firms were built around catching human errors — miscalculations, outdated data, sloppy reasoning — not around the specific failure mode of a system that produces confident, coherent, entirely invented information.

Insurers are starting to notice the gap too. Professional indemnity policies, the standard cover consultancies carry against claims of negligent advice, were written with human error in mind. Whether an AI-generated fabrication counts as the kind of error a PI policy is meant to cover is, in a lot of existing contracts, genuinely unclear. Firms that have thought this through are starting to update policy wording explicitly, and — more usefully for clients — starting to keep audit trails showing where and how AI was used in producing a given deliverable, so that if something does go wrong there's a record of what happened rather than a shrug.

The questions worth asking before you sign

None of this means AI has no place in consulting, or that firms using it are cutting corners as a rule. The research is fairly consistent that AI-assisted delivery, done properly, produces faster and often better work — the UK's Management Consultancies Association has found that consulting buyers are now roughly three times more likely to actively want generative AI involved in their engagements than they were a year ago. The issue isn't the tool. It's verification keeping pace with adoption.

For a UK business currently choosing between consultancies, that translates into a short, practical list of questions worth asking before signing anything. Ask how AI-generated content is checked before it reaches a client, and by whom. Ask whether AI use is disclosed in the engagement letter, and whether the firm can show an audit trail if asked. Ask, bluntly, who is accountable if something in the final deliverable turns out to be wrong — the individual partner, the firm, or an unsatisfying answer that lands somewhere in between. Firms with good answers to these questions tend to have them ready. Firms without good answers tend to change the subject.

Some firms are responding by creating something that didn't really exist as a job title two years ago: internal AI-assurance roles, sitting somewhere between quality control and compliance, whose entire remit is checking AI-assisted work before it leaves the building.

Trust as the actual product

There's a reasonable case that this moment ends up strengthening the firms that handle it well rather than damaging the industry broadly. Trust has always been the actual product a consultancy sells, more than the specific deliverable in any given engagement, and firms that can demonstrably show robust AI verification processes have a genuine differentiator to point to against competitors who can't. The KPMG-commissioned research into trust and AI, produced with the University of Melbourne, has found a consistent pattern across multiple markets: people are broadly willing to trust AI-assisted output, but that trust is conditional on visible human oversight, not blanket acceptance of AI involvement regardless of how it's checked. That's a useful finding for any UK consultancy thinking about how to talk about its own AI use with clients — the answer isn't hiding it, it's being specific about how it's supervised.

The regulatory backdrop is starting to catch up as well, and not just for the consultancies doing the work — the EU's AI Act now imposes formal governance obligations on high-risk AI use across sectors, a shift covered in more detail elsewhere on this site. Consulting sits in an odd position relative to that regulation: firms are simultaneously subject to it when they build AI systems for clients, and exposed to their own version of the same underlying problem when they use AI internally to produce the advice they're selling. Trust, in an industry whose entire value proposition rests on being right, is not a small thing to get wrong twice.