An owner I sat with this quarter pulled up her revenue by client for the last twelve months. The biggest client was 35 per cent. She had known that number for years, it had felt acceptable for years, the work was good and the relationship was strong. What was new on her desk was a second tab. Her AI tooling spend, broken down by which client’s work the tools were running on. The same client accounted for £600 a month of stack cost. Almost half of her total AI bill traced back to one relationship.
She said it more or less in those words. “It is the same number I had before, but now I cannot pretend it is not the number.” That is what has changed. The concentration ratio has not moved. The visibility has, and the visibility is what forces the conversation.
This post walks through what concentration risk actually means in the AI era, why the old thresholds still apply, and what to do when the biggest client is also the best client.
What is client concentration risk and what threshold actually matters?
Client concentration risk is the exposure that comes from one client representing a large share of total revenue. The thresholds in owner-managed services are 20 per cent as a yellow line worth tracking, and 30 per cent as a red line that needs an explicit decision. Sell-side advisers treat anything above 25 to 30 per cent as a material valuation discount in a sale.
What the thresholds describe is the basic dependency arithmetic. A client at 35 per cent of revenue is a client whose departure breaks the firm. If they leave next quarter, you do not have ninety days to replace the revenue, you have ninety days to cut costs by the same amount or close. That is the operational reality regardless of how warm the relationship feels.
Why does it matter for your business now in a way it did not before?
Two things have changed, and both sharpen the picture. The first is AI cost attribution. When £600 a month of tooling traces to one client’s work, the dependency is more concrete than the revenue view alone. The second is replacement capacity. AI productivity has lowered the cost of taking on a new significant client, which weakens the old defence that the revenue cannot be replaced.
The cost attribution side has come from the basic mechanics of how owner-managed firms have ended up paying for AI. Subscriptions are bought per seat or per workflow, and seats and workflows get assigned to specific client work. A senior advisor working two days a week on one large client is running their drafting stack, their research stack, and their meeting-summary stack against that client’s matters. The bill is no longer hidden inside salary cost. It sits on the credit card as a line item that traces back to a specific relationship.
The replacement capacity side is real but easy to overstate. AI does not magically generate new clients. What it changes is the cost of serving an additional client at acceptable quality, because the productivity gain on drafting, research, and routine analysis is higher than it was. Onboarding a new significant client used to require a partner to be available for a sustained period. The bar has come down, not to zero, but enough that the historical defence (we cannot replace this revenue, the firm cannot absorb a new client at that scale) is weaker than it was.
Where will you actually meet this in your firm?
You will meet it the first time you break your AI subscription costs down by client and realise the headline revenue percentage understates the concentration. A client at 35 per cent of revenue and 45 per cent of variable AI cost is more concentrated than the revenue number alone suggests. If they leave, the stack cost does not leave with them at the same speed.
You will also meet it in proposal conversations. The pipeline you used to need to bring on one significant new client a year now needs to bring on two, because the AI cost line is part of the baseline and the firm needs more revenue to carry it. The diversification cadence is not a marketing nicety, it is a basic load-bearing discipline. One to two new significant clients a year is the minimum for an owner-managed services firm that wants to keep concentration trending in the right direction.
The third place it shows up is in conversations with your accountant or your sell-side adviser, if you are within five years of an exit. Strategic Exit Advisors and others have written about customer concentration as a documented valuation drag. A firm with one client at 35 per cent of revenue sells for a meaningfully lower multiple than the same firm at 18 per cent, because the buyer is pricing in the risk that the client leaves post-acquisition. The discount is not a small number.
When to act on a concentrated client, and when to leave it alone?
Act when concentration is above 25 per cent, when you have deferred the conversation for over twelve months, or when the client operates in a sector under pressure. Leave it alone only when concentration is under 20 per cent and the firm already brings on one to two new significant clients a year. Unmonitored single-client risk is what gets firms into trouble.
The proportionate move at owner-managed scale is to commit to a named diversification cadence rather than fire the concentrated client. Write down the threshold no client is allowed to cross without an explicit decision, and run the conversation with yourself quarterly. A firm that holds 30 per cent of revenue with one client today, holding the same client at 22 per cent in eighteen months, has done the work. The relationship did not need to change, the denominator did.
The trap to watch for is the best client and the most concentrating client being the same one. They commonly are, that is how concentration builds in the first place, the relationship pays well and is easy to grow inside. The conversation that gets deferred is the diversification one, because the alternative feels like introducing risk into something that is working. The long-term cost of doing nothing is the same cost as any unmanaged concentration risk, it compounds quietly until the client’s circumstances change and the firm has no buffer.
Related concepts and what to do next
The adjacent posts worth holding in mind are the growth and profit dashboard, recurring versus project revenue with AI, the valuation discount for founder-dependent firms, and firing the 30 per cent client. The diagnostic I run fits on the back of an envelope. Top client as a percentage of revenue, same client as a percentage of variable AI cost, new significant clients added this year.
None of this is a recommendation to fire concentrated clients on principle. The biggest client is often the best client, and good relationships are worth holding. The point is that AI has made the concentration visible in a way it was not before, and the historical defence for sitting on it is weaker than it used to be. Run the diversification cadence regardless of how good the big client is. The firm gets more resilient, the valuation gets healthier, and the conversation you have been deferring stops sitting on your desk.
If this is where you are, Book a conversation.
