Forty pitches a year. That's roughly what a mid-market management consultancy runs through - sometimes more, sometimes fewer, but somewhere in that ballpark. I was sitting with a managing partner a few months back, and we did the maths together on a napkin. He'd never actually added it up before. When we got to the number - somewhere north of 700 consultant hours a year, just on proposals - he went quiet for a moment. Then he said, "That's basically two full-time people."
And that's before you add the research hours at the start of every engagement, the time people spend hunting for "that framework we used for the logistics client in 2022," and the late nights reformatting slides that look suspiciously similar to slides you've already formatted three times this year.
That's not consulting. That's administration dressed up as consulting. And it's eating your margins alive.
Our value is in the judgement and the relationships. We can't automate that. And I'm not sure our clients would be comfortable knowing we used AI on their work.
I hear this a lot. The first half is absolutely right - your value is in the judgement. Which is exactly why your best people shouldn't be spending a third of their week on tasks that don't require any. The second half, the client comfort question, deserves a proper answer, and I'll come back to it. But let me be direct: the three use cases I'm about to walk through don't touch the advice. They don't touch the thinking. They touch the stuff that sits around the advice - the writing, the finding, the assembling. The stuff that currently burns junior consultant hours like kindling.
If you're a managing partner watching margin per engagement drift slowly downward while your competitors seem to be doing more with less, this is the most straightforward productivity opportunity available to you this quarter. And none of it requires a platform investment.
Start here. This is the one with the clearest return, the most mature tooling, and the lowest client-facing risk - provided you put a proper quality control step in place.
Here's what it looks like in practice. You take your firm's best past proposals, your case study library, and the specific client brief, and you feed them into Claude, Microsoft Copilot, or a similar tool alongside a well-structured prompt template. The AI produces a structured first draft - not a finished proposal, a draft - that pulls from your existing language, references relevant past work, and follows the structure your firm typically uses.
I want to be really clear about something: the AI should never be the final author of anything that goes to a client. Ever. A senior consultant reviews the draft, rewrites the sections that require original thinking or client-specific nuance, and applies the kind of judgement that - rightly - your clients are paying for. The AI handles the scaffolding. The human handles the substance.
I ran a version of this with a professional services firm last year. The first AI-assisted proposal came back and the partner leading the pitch was... sceptical, let's say. He read the draft, crossed out about 40% of it, and rewrote the rest. Fair enough - it was the first attempt, the prompt library was rough, and the firm's past proposals weren't well-organised. But the second one was better. By the fifth, he was spending about four hours on a proposal he'd previously have spent fifteen on. He still crossed things out. He still rewrote sections. But the scaffolding was doing its job.
The margin impact is meaningful. If proposal writing currently takes 15 to 20 hours per pitch and AI reduces the drafting stage to 4 to 6 hours, you're saving roughly 10 to 14 hours per proposal. For a firm pitching 40 times a year, that's somewhere between 400 and 560 hours of consultant time freed up annually. At a blended rate of, say, £180 an hour, you're looking at £72,000 to £100,000 in recovered capacity. Per year. From one workflow change.
What does it require to get started? A structured prompt library (which you can build iteratively - it doesn't need to be perfect on day one), a reasonably well-organised archive of past proposals, and a quality assurance protocol that everyone agrees to before the first AI-assisted proposal goes anywhere near a client.
And what should you measure? Time from brief to first draft submission. Senior review time as a proportion of total proposal time. And - this is the one that addresses the quality question directly - win rate in AI-assisted proposals versus non-assisted proposals over a six-month period. If the win rate drops, something's wrong with the process. If it holds or improves, you've got your answer.
Treat the first five AI-assisted proposals as a deliberate calibration exercise. What worked? What needed heavy revision? What would a better prompt library look like? That learning feeds everything else.
This one's trickier. Not because the tools aren't good - they are - but because the risk profile is different.
A typical engagement research phase chews through 20 to 30 consultant hours: background analysis, market context, literature reviews, competitor mapping. Junior consultants do most of it, and honestly, a lot of that time goes into finding and reading things rather than synthesising them. AI tools like Perplexity, Claude, and ChatGPT with web search can compress the initial research pass dramatically - we're talking a reduction from 20 to 30 hours down to 8 to 12 hours for the first sweep. For a firm running 20 engagements a year, that's potentially 240 to 360 hours recovered.
But, and this is a big but, you cannot skip source verification. This is not optional. It is the single most important safeguard in the entire workflow.
AI research tools are brilliant at surfacing relevant information quickly. They're also perfectly capable of confabulating - presenting something that sounds authoritative but is either inaccurate, out of context, or occasionally just made up. I've seen it happen. A junior consultant used an AI tool to pull together background for a market entry analysis, and one of the "sources" turned out to be a misattributed quote from a report that didn't exist. It was caught in review, thankfully, but it could easily have ended up in a client deliverable.
So build the verification step explicitly into the workflow. Not as a nice-to-have. Not as "yeah, obviously check the sources." As an actual, documented step with its own time allocation and its own sign-off. The consultant reviews every source, evaluates credibility, and flags where the AI has summarised accurately versus where it's paraphrased loosely or filled in gaps.
What to measure: research phase hours per engagement, source accuracy rate across the first ten AI-assisted research outputs (track it properly - you need to know your baseline), and consultant confidence scores. That last one matters more than you'd think, because if your people don't trust the output, they'll quietly stop using it and go back to doing everything manually.
Right. This is the one that every consultancy needs and almost no consultancy does well.
You know the conversation. Someone's staffing a new engagement and says, "We did something similar for a financial services client about eighteen months ago - who was on that?" And then three people spend twenty minutes trying to remember, someone checks their email, someone else pings a WhatsApp group, and eventually someone digs out a deliverable from a shared drive folder that hasn't been organised since 2019.
AI-powered search and synthesis across your internal knowledge base - past deliverables, methodologies, sector research, expert profiles - can make that conversation obsolete. A consultant types a query, and the system surfaces relevant prior work, identifies who in the firm has relevant expertise, and pulls together a synthesis of what the firm already knows about this type of problem.
If you're in the Microsoft ecosystem, Copilot for SharePoint and Teams is the natural starting point. If your document management system isn't well-integrated with Microsoft, there are specialist KM tools worth evaluating. For firms with more technical capability, a retrieval-augmented generation (RAG) setup over your existing document store gives you the most flexibility.
But here's the honest bit: this use case has the worst adoption track record of the three. And it's not a technology problem. It's a people problem.
Consulting firms have strong informal knowledge networks. Partners know who knows what. The "ask around" method works - not perfectly, not efficiently, but well enough that people don't feel the pain of not having a system. So when you introduce a shiny new knowledge management tool, the response is often polite indifference followed by quiet abandonment. I've watched it happen at two firms in the last three years. Both had decent tools. Both had champions. Both had launch events with slightly too much enthusiasm. Neither got meaningful adoption for the first six months.
The short version of what actually works: design for adoption from day one. Understand the existing behaviour you're competing with, make the new behaviour easier than the old one, and build visible early wins that give people a reason to change their habits. If your document hygiene is a disaster - and at most firms it is - sort that first. A RAG system over a chaotic shared drive is just a faster way to find the wrong thing.
What to measure: search query volume (is anyone actually using it?), time from query to relevant content found, reuse rate of past methodologies and frameworks in new engagements, and - my favourite - a reduction in those "who was on that project?" conversations.
Here's the framing that matters, because I think it's the one that determines whether your partners get behind this or quietly resist it.
These three use cases save consultant time on the activities that sit around the advice - writing, researching, finding - not on the activities that constitute the advice: thinking, synthesising, recommending, facilitating. A consultant who spends four fewer hours on a proposal has four hours to think more carefully about the client's problem. A consultant who finds the relevant precedent in ten minutes instead of an afternoon has an afternoon for analysis.
The margin improvement comes from applying consultant time to higher-value activities. If your consultants think AI is a prelude to redundancy, adoption will be glacial. If they experience it as something that takes the tedious bits off their plate, they'll be your biggest advocates.
I said I'd come back to this, so here's my honest position.
If you're using AI to assist with proposal drafting and research, should your clients know?
Yes. Not because you're legally obligated in most cases, but because hiding it is a worse strategy than owning it. Your clients are not naive. Many of them are using the same tools themselves. What they care about is whether the quality of your advice is maintained, whether the judgement is still human, and whether you've thought carefully about where AI is and isn't appropriate.
I'd go further. Firms that proactively explain their approach - "We use AI to accelerate our research and drafting processes, with senior consultant review at every stage" - will find that clients view it as a sign of sophistication rather than a shortcut. The firms that get caught using AI without disclosing it are the ones who'll have the awkward conversations.
Be upfront. Have a position. Write it down. Make sure every consultant in the firm knows what it is.
The sequence matters. Start with proposal generation - clearest margin impact, most mature tooling, lowest client-facing risk with proper quality control. Use the first five AI-assisted proposals to calibrate your approach. Then move to research acceleration, which builds on similar tool foundations but requires a more disciplined workflow around source verification. Knowledge management comes last, not because it's less valuable, but because the adoption challenge means it takes longer to show results.
None of this requires a six-figure platform investment. You can pilot all three with tools your firm likely already has access to. The real question isn't whether to start - your consultants are almost certainly using these tools already, informally, without governance. The question is whether you start deliberately, with defined success criteria and quality controls, or whether you let the ad-hoc adoption that's already happening proceed without guardrails.
If you want to understand which of these three use cases fits your firm's current capability - and where the margin impact will be highest - we've built a consulting AI use case scorecard that benchmarks your readiness across knowledge base quality, proposal library organisation, research tool access, and fee earner AI literacy. It takes about ten minutes. Worth doing before you start.