AI in professional services is leaking margin. Here's how to plug it.

Every services firm I talk to is measured on three things: delivering on time, keeping billable utilization healthy, and delivering projects profitably without over-servicing. AI should help with all three. So far, it's helping firms deliver faster. It isn't yet making them more profitable.

SPI Research's 2026 Impact of AI on Professional Services study shows the gap. AI now touches 43.6% of projects, nearly double a year ago, and firms using it on most of their projects deliver far more of them on time. Yet while 84.8% of firms say AI is saving them time, only 29.9% say it's generating revenue. Utilization is flat, and the firms saving the most time are further from their revenue targets than firms that saw no change at all.

Unless AI adds real business value, it's just more noise. We learned that ourselves. Early on at Teamwork.com, like most software companies, we were quick to add AI to features. Getting clear on how it would make firms more profitable took longer.

Plenty of services firms are in that same phase with their own delivery. The time savings are real, and turning them into margin takes a few deliberate moves.

The hours don't vanish. They leak.

Most of the time AI saves goes to the client, or nowhere. SPI Research calls this the Leak. Very few firms choose it. It happens by default, because of how the work is priced.

Flat utilization is that third row showing up in the numbers. The hours were saved, and then nobody did anything with them.

We hear it from customers directly, and it's no longer just agencies. Their clients are asking, "How much of this work is being done by your people versus by AI?" That question lands straight on your margin. If you can't answer it with numbers, the client will answer it for you, usually with a request for a discount.

So the first move is to find your own leak. Go through your client list by pricing model and ask which row each one sits in. Hourly clients are where AI savings shrink the invoice fastest. Retainers and fixed fees are where they disappear into scope nobody bills.

Once you know where the hours are going, you can decide where they should go instead.

Decide what each saved hour is for

There are three good options for the hours AI frees up:

1. Keep your price where it is. If the deliverable is just as good, delivered faster, the client is paying for the outcome, not the hours. As a local business near us puts it, price is often long forgotten, but the memory of quality never fades.

2. Use the lower cost to win more work. A lower cost to deliver lets you compete on price without giving away margin.

3. Grow without adding headcount. This only works if the new work is profitable. After COVID, a lot of firms chased top-line growth without checking they could deliver it profitably, and their margins on existing clients paid for it when growth got harder.

The mistake is making that call once, for the whole firm. Make it per client or per service line instead.

A long-standing client who values quality is a case for holding the fee. A competitive bid is a case for pricing more sharply. Spare capacity in a team with a strong pipeline is a case for growth, as long as that growth is profitable.

Not choosing is still a choice. It just means the Leak decides for you. Each of those calls, though, depends on a number most firms don't have.

You can only decide if you know what the work costs

Every one of those choices depends on knowing what a deliverable costs to produce. That's where the research finds AI helping least. Demand forecasting and rate analysis rank lowest, because both need clean operational data from connected systems.

This isn't a failure of AI tools. It's a plumbing problem. A chat assistant can't price a proposal from your actuals if your actuals are spread across three spreadsheets and a time tracker. SPI Research expects the next big gains to come once AI can draw on the data in professional services automation (PSA), CRM and financial systems.

That's the space we're building Teamwork.com to fill, between your CRM and your finance system. Projects, resourcing and costs live in one place, so AI works from real numbers from real delivery.

Some firms are already moving. Consulting firms are tracking tokens and AI spend per project, and they want their PSA software to manage it. That's the right instinct: AI cost has to sit in the same place as people cost, or the margin numbers don't mean anything.

Getting there starts with making three things visible.

1. Log AI work the way you log people's work. Record the time AI saves and what it costs (tool spend, tokens or credits) against the project and the task, not in a separate AI budget. Then compare it against what comparable work cost before. When an agent does the foundational research and competitive analysis on a client engagement, the hours your consultant didn't spend are hours you can price, sell or redeploy.

2. Track where the freed hours go. Every saved hour ends up in one of three places: billable work, unbilled scope, or the bench. Look at billable and non-billable time by person and by client, so you can see where capacity is drifting into over-servicing.

In my conversations with customers, resourcing is the capability they most often underestimate. It isn't just scheduling. It's forecasting the people you need to win more work, and balancing workloads so the freed capacity doesn't land on the same few people.

3. Watch margin while the work is still running. A month-end report tells you margin slipped after you can do anything about it. Compare actual cost against the fee as the project runs, so you can have the scope conversation, rebalance the team or change the pricing while there's still time.

Teamwork.com captures resourcing and financials as delivery happens, not bolted on after, so you can track project profitability live, not after it's finished. And once AI's cost and output are visible alongside your people's, you can start managing it the same way.

Then plan AI like part of the team

Manage AI the way you manage your people: plan it, cost it and supervise it.

In an agency, that might be an agent doing the SEO or paid media work on a client account. In a consulting firm, it might be one compiling the research, competitive analysis and reports for a client engagement. In an IT services firm, it might be one drafting status reports across a transformation program. Either way, it's delivering client projects, and it needs to be planned and costed like anyone else on them.

When people and agents sit in the same cost of delivery, leaders can see who delivered each piece of work, what it cost, and whether it was up to standard.

This is where the payoff is. SPI Research calls it the Readiness Premium: the profit advantage of firms that built their AI foundations first. That means ready data, a way to measure what AI delivers, trained people, and agents doing real work. One firm in five already has agents running multi-step processes across its systems.

Data point: Firms running complex agents or an AI-first model report 20.5% EBITDA. Firms using AI only as a chat assistant, or not at all, report 13.3% (SPI Research, 2026 Impact of AI on Professional Services).

That gap doesn't prove agents created the extra margin. It does show which way the most profitable firms are moving. Customers increasingly expect software to deliver some of the work, not just plan and coordinate it. The firms that use AI to deliver work, rather than just talk about it, will have the best margins and the best ability to win and keep clients.

You don't need a firm-wide rollout to get started. Put one agent on one deliverable, cost and measure it like a new hire, and keep what earns its place. It's why the delivery agents we're building at Teamwork.com show up like the rest of your team: each one a costed, supervised line item with an owner.

Where to start this quarter

The time savings are already here. Whether they become margin depends on decisions most firms haven't made yet.

If you do one thing this quarter, take your three largest clients and answer these for each:

1. What is AI saving us on this client, hour by hour?

2. Where is that time going: billable work, unbilled scope or the bench?

3. Would we know if margin started to slip while the work was still running?

Wherever the answer is "we don't know", you've found your leak. That's the place to start.

Frequently asked questions