I ran an AI audit this week with a senior investment consultant at a mid-sized financial consultancy operating across two markets.
Here’s what I found, what I recommended, and what it tells us about where most professional services firms actually are with AI right now.
Who I was talking to
My client leads investment consulting for institutional clients including pension funds and insurance companies.
His day spans investment strategy, quarterly performance reporting, board-level presentations, minutes production, regulatory compliance, and cross-office coordination with two other country offices.
He’s sharp, self-directed, and already using AI without anyone asking him to. That last part will be important later.
What his daily workflow looks like
Before making any recommendations, I needed to understand where his time went. I asked him to walk me through a typical day, and what came out was a detailed picture of a highly capable person doing a significant amount of work that doesn’t need him.
| Task | Frequency | Time cost |
| Investment committee meeting | Regular | 1.5 to 2 hours |
| Minutes production (post-meeting) | Per meeting | Half a day to a full day |
| Quarterly report writing | Per client | 60 to 90 minutes |
| Data entry | Monthly, batched quarterly | Up to half a day |
| Presentation formatting and copy-paste | Per client, quarterly | 1 to 2 hours |
| Strategy simulation | Periodic | Full day, sometimes past 8pm |
The strategy work he enjoys, and it’s clearly where he creates the most value. Minutes production, data entry, and copying graphs into PowerPoint decks consumed a substantial portion of his week.
When I asked which tasks he most wanted off his plate, he didn’t hesitate: data entry, minutes taking, and copy-paste formatting. All three are strong candidates for AI automation.
The shadow AI problem
This part didn’t surprise me at all but it matters the most.
He pays $20 a month out of his own pocket for ChatGPT Plus. His associate uses the free tier. Other colleagues dip in and out on free accounts.
Nobody has a company-sanctioned AI programme, nobody has a formal policy, and leadership actively discourages AI use on the grounds that it makes people intellectually dependent.
So what’s actually happening? People use AI anyway, on personal accounts with no data governance, no audit trail, and no organisational oversight. Client fund data, member figures, and performance numbers all go into a consumer ChatGPT account that the company knows nothing about.
The leadership’s skepticism isn’t preventing AI adoption—merely making it invisible, which is the worst possible outcome from a compliance and risk standpoint.
This pattern has a name: shadow AI. It’s far more common than most organisations want to admit. The consultancies, law firms, and financial services businesses that think their teams aren’t using AI are almost always wrong.
Their teams are using it, just not telling anyone.
What his AI use looked like
He used ChatGPT to polish minutes and report sections after drafting them himself, working section by section rather than submitting full documents at once.
He’d write a section, feed it in, ask for a language and tone pass, and move on. Careful, deliberate, and methodical.
The most striking example he gave me was a 5,000-row Excel dataset he’d been asked to reconcile. Numbers weren’t adding up across a pivot table, and the problem traced back to how XLOOKUP handles duplicate entries.
He’d been working on it for hours without resolution. He pasted the file into ChatGPT, described the problem, and had a diagnosis and a fix within minutes.
The tool identified the duplicate-key issue, explained why the lookup was only returning the first matching row, and recommended a unique composite key as the structural solution.
He implemented it and sent the file back the same day.
That’s a senior analyst using AI to solve a problem that would otherwise have required either a specialist or several more hours of trial and error.
What I recommended
The recommendations fell into four areas, ordered by how quickly he could act on them.
Meeting transcription
He was spending half a day to a full day producing minutes from meetings, listening back to recordings, pausing, writing summaries by hand, and then polishing in ChatGPT.
I recommended Fathom, the AI notetaker I use and have tested most thoroughly. It joins the meeting, transcribes in real time, and produces a structured summary with action items when the call ends.
Minutes production drops to 20 to 30 minutes of review and light editing.
Client context files
His hesitation about using AI for quarterly reports was legitimate: ChatGPT doesn’t know his clients’ circumstances.
The fix is a one-page context file per client covering the fund’s mandate, portfolio composition, manager mix, and notable recent events.
He builds it once, updates it quarterly, and feeds it into the model at report time alongside the quarter’s key data. The model drafts the client-specific sections; he reviews and adds the human detail it can’t know.
The compounding benefit: after each meeting, he can feed the Fathom transcript alongside the existing context file and ask the model to update it automatically.
Over time, that becomes a living client intelligence file that requires almost no manual maintenance.
Power Query for data entry
ChatGPT had already recommended this when he used it to debug the XLOOKUP problem, and he’d applied it successfully.
The same principle applies to his quarterly data consolidation: a one-time Power Query setup replaces the half-day manual batching exercise with a scheduled refresh.
Weekly AI news briefing
He relied on a single source for quarterly market context and supplemented it with memory.
A weekly ChatGPT prompt with web search enabled, structured around his five standard market categories, replaces the single-source dependency and gives him an organised log of the quarter’s events to draw from at report time.
All four recommendations were free or already covered by tools he paid for. None required organisational approval.
The bigger conversation
The recommendations above are individual-level fixes. They make his working week substantially better. But the more valuable outcome of the audit was giving him a structured case for two harder conversations.
The first is about shadow AI
A company whose leadership discourages AI use while its staff fund their own subscriptions and enter client data into consumer tools has a governance problem, not an AI problem. The answer isn’t prohibition (which almost never works), but a policy that brings usage into the open, sets clear boundaries around data handling, and gives the team the tools and guidance to work properly.
The second is about tool access
His office operated at a systematic disadvantage relative to the firm’s other offices, without direct access to key data platforms that colleagues in other locations used as a matter of course.
He was funding the productivity gap personally. That’s a retention risk and a quality risk.
An AI audit gives you the evidence to have both conversations. It turns a vague feeling that things could be better into a specific, documented account of where the time goes, what the risks are, and what a better state looks like.
| Recommendation | Cost | Timeline |
| Fathom for meeting transcription | Free to start | Today |
| Client context files and report prompts | £0 | This week |
| Weekly AI news briefing | £0 | This week |
| Power Query data consolidation | £0 | This month |
| Shadow AI policy conversation with leadership | Time only | When ready |
| Tool access business case | Time only | When ready |
What this AI audit confirmed
Most organisations are behind on AI because the people who see its value are working around a culture that hasn’t caught up yet.
My client wasn’t waiting for permission. He was paying for tools out of his own salary, solving real problems, and producing better work because of it.
The audit didn’t tell him anything he didn’t already know. But it gave him a clear picture of the full scope, a language for the risks, and a set of specific next steps with no ambiguity about where to start.
Get your own audit: If any of this sounds familiar, a free 30-minute AI audit will give you the same clarity. We’ll look at where your time actually goes, what tools your team already uses, and where AI could remove the most friction from your workflow. You leave with a clear picture of where you stand and a specific next step, with no obligation to go further. Book yours at mohammedshehu.com/audit.