AI Didn't Raise Your Agency's Margin. You Didn't Redesign the Work.

By Craig S. Cody, CPA, Certified Tax Coach.

Here's the answer up front. If your agency adopted AI and your margin looks about the same, the problem almost certainly isn't the tool. It's that you handed a faster tool to a process that was designed around what humans could do by hand, and never changed the process. The companies actually getting paid for AI aren't the ones with the best subscriptions. They're the ones who rebuilt how the work moves. And there's one number that tells you which of those two you did, in about ten minutes, using a report you already have.

I work with more than 70 marketing and advertising agency owners every month. The tools they're using are good. The gap I keep running into isn't a technology gap at all.

Why Isn't AI Showing Up in Your Profit?

Because adoption and return are two completely different things, and almost everyone is measuring the first one.

McKinsey surveyed 1,719 professionals and business leaders for its 2026 State of AI report. Nearly nine in ten said their organization regularly uses AI in at least one business function. Eighty percent said it improved their own productivity. Then the survey asked whether any of that reached earnings, and just 37% could attribute even some EBIT impact to AI.

Now read the part almost nobody quotes. That 37% is unchanged from a year ago.

Sit with that for a second. Between those two surveys, companies poured record money into this. Twenty-eight percent now say AI eats more than 10% of their IT budget. Sixty percent plan to spend more next year. And the share who can point at a profit line and say "AI did that" moved by zero.

That's not a technology story. Technology got dramatically better in that window. That's a story about what people did with it.

What Separates the Companies Getting Paid?

McKinsey put a name on the group that's winning, and the definition is worth reading carefully.

They call them AI high performers: organizations that attribute at least 5% of their EBIT to AI and describe the impact as significant. That's the bar. Five percent. And only 6% of respondents clear it, which is also unchanged from last year.

So what do those humans do differently? According to Fortune's reporting on the survey, they're far more likely to have fundamentally redesigned their workflows rather than adding AI to existing processes. Nearly three-quarters of them have, up from 55% the year before.

That's the whole finding, and it's a quiet one. The winners didn't buy better AI. They changed the shape of the work, and then put AI inside the new shape.

Here's what that distinction looks like on the ground. Adding AI to an existing process means your team still does the same eleven steps, and steps four and seven got faster. Redesigning the work means you looked at the eleven steps and asked which ones only existed because a human was doing all of them by hand. Usually two or three steps stop existing. That's a different exercise, and it isn't a software purchase. It's a decision somebody has to sit down and make.

Isn't a Small Agency at a Disadvantage Here?

This is where the survey gets misread, and where I think agency owners have been handed exactly the wrong lesson.

The size data looks brutal at first glance. Fifty-four percent of organizations above a billion dollars in revenue are scaling AI enterprise-wide, against roughly a third of smaller firms. On AI agents specifically, big companies went from 27% to 40% in a year while smaller companies sat flat at 22%.

Read that as "we're falling behind" and you'll go buy more tools, which is the exact move that produced the 37%.

Read it against the actual finding and it flips. The thing that separates the winners is workflow redesign. Now ask which organization can redesign a workflow faster: the one with a billion in revenue, four layers of approval, a change management office and a systems integrator on retainer, or a fourteen-person agency where the owner can pull the account team into a room on a Tuesday and rewrite how a project moves by Friday.

The enterprise has more money. You have less distance between deciding something and it being true. On this specific problem, that's the more valuable asset, and it's the one nobody's telling you that you have.

Big companies are buying their way toward the thing you could just go do.

What Number Tells You Which One You Did?

EBIT is the wrong instrument for you. It's a fine number for a public company with an investor relations department, and it isn't what a marketing or PR agency owner actually runs on day to day.

The agency translation of McKinsey's test is agency gross income per full-time employee, measured across two periods.

Quick definition, because the acronym collides. AGI here means agency gross income, what's left of your billings after the pass-through costs of delivering the work. Media, freelance, print, production. It's not the Adjusted Gross Income line on your 1040. If you've been running your ratios against total billings, that's its own expensive problem and I've written about why billings is the wrong denominator separately.

Now the test. Pull your AGI for the last twelve months and divide it by your full-time headcount. Do the same for the twelve months before that. Two numbers. That's the whole exercise.

If AI changed how the work gets done, that number went up. If AI just made the same work feel easier, it didn't.

Here's why this particular ratio is the right one, and it's the part worth actually understanding rather than just running. AGI per FTE is a fraction. There are only two honest ways to move a fraction up: raise the top or hold down the bottom. Which means freed capacity has exactly two legitimate destinations, and they're literally the numerator and the denominator of the same number.

Destination One: Billable Output You Actually Added

This is the numerator. More AGI, same people.

The saved hours became work a client paid for. A new retainer the existing team absorbed without anyone drowning. A service line you couldn't previously staff. Faster turnaround that let you take on a fifth account instead of a fourth.

The test is unforgiving on purpose: can you point at an invoice that exists because of the capacity AI freed up? Not a proposal. Not a plan. An invoice.

One warning here, because this is where the numerator quietly leaks. If AI let your team do more and you gave that "more" to existing clients without changing the fee, you didn't add billable output. You widened the scope at the old price, which is a discount you never decided to give. Faster work you don't collect on doesn't move the top of the fraction at all.

Destination Two: The Hire You Didn't Have to Make

This is the denominator, and in my experience it's the destination agency owners are most likely to have actually reached and least likely to have counted.

Nobody books a number for a hire that never happened. There's no line on the P&L called "the person we didn't need." So the return is real, it's sitting in your bank account, and it's invisible.

Put a figure on it. A $95,000 salary isn't a $95,000 decision. Add employer Social Security at 6.2%, Medicare at 1.45%, unemployment taxes and a realistic benefits load and you're at roughly $118,000 all in. And that still isn't the number that decides the hire. Drew McLellan and the Agency Management Institute put healthy agency people costs at or under 55% of AGI, so that $118,000 seat has to carry about $215,000 of new agency gross income to sit inside a healthy payroll envelope. I walk the full build through in what it actually costs to hire an employee.

So if AI absorbed the work that would have forced that hire, the return isn't "we saved some time." The return is a $118,000 annual cost you don't carry, and $215,000 of new business you didn't have to go win just to stand still.

Write that down somewhere. It's the single most under-claimed AI return in this industry, and it's the one that shows up in your AGI per FTE immediately, because you held the denominator flat while the numerator kept moving.

And If It Went Somewhere Else?

Then you have your answer, and it's the useful kind.

When an agency owner tells me AI has made their delivery faster, I ask the same thing back: did that contribute to a better bottom line for the agency, or did that margin evaporate?

Those are the only two places it can go. The speed is real either way. But if the capacity didn't become billable output and it didn't prevent a hire, it went into the third place, which isn't a destination at all. It became a slightly easier week. More internal revisions because generating another round is cheap now. Longer review cycles, because AI is confidently wrong sometimes and someone senior is catching it, which moves labor onto your most expensive humans where nobody writes it down.

None of that is a moral failure. It's just what happens when you speed up a process instead of redesigning it. The work got easier and the shape stayed the same, so the money stayed the same too.

And notice that evaporation still costs you. Twenty percent of the companies in that survey say AI operating costs are already constraining how much they can use it. The subscriptions are real whether or not the return showed up.

What This Article Is Not About

Three things sit next door to this one, and I'm deliberately leaving them alone.

What your AI stack costs you in total is a real question, and it comes before this one. What you should charge now that delivery costs less is a pricing question, and it's a separate sheet. What your AI is allowed to touch, spend or decide on its own is a controls question, and it's a third.

This piece covers one thing: whether you changed the work or just accelerated it, and the arithmetic that tells you which.

Who This Is For, and Who It Isn't

If you've rolled out AI across your agency, the team genuinely likes it, and you've quietly wondered why the margin looks about like it did last year, this is for you. That gap between obvious speed and unchanged profit is the most common thing I see right now, and it's fixable in a way that doesn't require buying anything.

If you're looking for someone to tell you the answer is a better model or three more subscriptions, I'm not your guy. And if you haven't got clean AGI figures for the last two years, start there instead. You can't run this test on numbers you don't trust, and the agencies that flourish are the ones that fixed the reporting before they went looking for the return.

Frequently Asked Questions

Why isn't AI improving our profit even though our team is faster?

Usually because the workflow never changed. Adding AI to an existing process makes individual steps faster while leaving the same number of steps, handoffs and review cycles in place, so the saved time gets reabsorbed internally instead of converting to money. McKinsey's 2026 survey found only 37% of organizations could attribute even some EBIT impact to AI, unchanged from the prior year, while the small group seeing real financial gains were far more likely to have fundamentally redesigned workflows rather than layering AI onto existing ones.

What does "redesigning the workflow" actually mean?

It means removing steps, not speeding them up. Walk through how a project currently moves from kickoff to client approval, list every step, and identify which ones exist only because a human was doing the work manually. Those steps get eliminated rather than accelerated. If your process still has the same number of stages it had before AI, you accelerated the work, you didn't redesign it.

How do I measure whether AI is actually making my agency money?

Divide your agency gross income by your full-time headcount for the last twelve months, then do the same for the twelve months before that. Agency gross income is billings minus pass-through costs like media, freelance and production, not the Adjusted Gross Income on your personal return. If AI changed how the work gets done, that per-employee figure rose. If it only made existing work feel easier, it stayed flat while your tool spend went up.

Where should the time AI saves us actually go?

There are only two destinations that show up in your numbers: billable output you added, or headcount you avoided adding. The first raises agency gross income with the same team. The second holds your headcount flat while the business grows. Anything else, more internal revision rounds, longer review cycles, a generally easier week, is capacity that evaporated, and it won't appear in your margin no matter how real the time savings felt.

Is a small agency at a disadvantage compared to large companies using AI?

Not on the thing that matters. Large organizations are scaling AI faster, with 54% of billion-dollar companies deploying it enterprise-wide against roughly a third of smaller firms, but the factor separating financial winners from everyone else is workflow redesign. A fourteen-person agency can redesign how a project moves in a week, where a large enterprise needs a change program to do the same thing. On this specific problem, being small is an advantage.

How do I count the value of a hire I didn't have to make?

Price the seat you avoided. A $95,000 salary carries roughly $118,000 in all-in annual cost once you add employer Social Security, Medicare, unemployment taxes and benefits, and that seat needs to carry about $215,000 of new agency gross income to stay inside a healthy payroll envelope of 55% of AGI, a benchmark from Drew McLellan and the Agency Management Institute. If AI absorbed the work that would have required that hire, the avoided cost and the revenue you didn't have to chase are both real returns, even though neither one appears as a line on your P&L.

The Bottom Line

The tools aren't the variable anymore. Nearly everybody has them, and a year of spending more on them moved the profit number for exactly nobody new.

What moved was the work, for about six percent of companies.

You don't need a bigger budget to join them. You need an afternoon, your last two years of agency gross income, your headcount for each year, and the willingness to find out that the answer is flat. Flat is useful. Flat tells you the capacity went somewhere, and now you get to go decide where it should have gone instead.

Then pick a destination on purpose. More billable output, or a seat you never have to fill.

If you'd like the tax and profit side of keeping more of what your agency makes in one place, I wrote a book for agency owners. You can request a free copy at the link below. And if you want a second set of eyes on the numbers themselves, let's talk.

Request a free copy of my book

Craig S. Cody is a CPA, Certified Tax Coach, and retired NYPD Lieutenant. His firm works with more than 70 marketing and advertising agency owners every month, helping them keep more of what they make through proactive tax planning.

This article is general education, not advice for your specific situation. Confirm your own facts with your advisor before acting.