How AI changes what agencies should charge for

The honest version of the AI story for agencies isn't that you can do magical new things. It's that implementation got cheaper, and that is not automatically good news for a business that sells implementation by the hour.

Cheaper implementation compresses timelines. Compressed timelines compress prices, because clients can see the work took less time and eventually ask why the invoice didn't move. What's left is a smaller pool of hours to earn your margin on.

The arithmetic

A hypothetical agency, loaded cost $95 an hour, billing time and materials at $150.

A project that used to take 300 hours now takes 180 — a 40% reduction, which is roughly what happens when first drafts, variants, transcription, research summaries and repetitive production get faster.

	Before	After
Hours	300	180
Revenue at $150/h	$45,000	$27,000
Cost at $95/h	$28,500	$17,100
Gross margin	$16,500	$9,900
Margin rate	36.7%	36.7%

The margin rate is untouched, which is exactly why this is easy to miss on a dashboard. The margin dollars fell 40%, in lockstep with the hours.

To earn the same $16,500 on 180 hours, you'd need:

Required revenue = $17,100 + $16,500 = $33,600
$33,600 / 180 hours = $186.67 per hour

That's a 24.4% rate increase to stand still — announced to clients in the same year they can see the work getting faster. Good luck.

The same project priced on output

Now hold the price and change the model. Same 180 hours, but the work is sold as a fixed price for a defined outcome at the old $45,000.

Revenue $45,000 − cost $17,100 = $27,900 margin (62.0%)

That's not a trick. It's the direct consequence of pricing the deliverable rather than the input. Efficiency gains accrue to whoever owns the risk, and on hourly billing that's the client.

The catch, and it's a real one: at $45,000 for work the market increasingly knows is faster, you are relying on the client valuing the outcome rather than benchmarking the hours. Which means the price has to be attached to something other than effort — and defending that is now the core commercial skill.

What gets scarce

If implementation is the cheap part, value moves to the parts that didn't get automated.

Knowing what to charge. When production hours stop being a reliable proxy for value, pricing becomes a judgement call every time. Agencies with a real cost base, a measured estimate-error rate and a considered view of what an outcome is worth will price confidently. Agencies that price at "hours × rate" will price themselves down.

Proving you delivered. A client who suspects the work took a fraction of the time will want evidence of what they bought. Not timesheets — outcomes, decisions, options considered and rejected, the reasoning behind the thing. The deliverable is now easy to produce and the judgement behind it isn't, so the judgement is what you have to make visible.

Protecting scope. This is the one that bites hardest. When each change looks cheap to make, requests multiply. Fourteen revisions that each take twenty minutes instead of two hours still consume attention, review, coordination and approval — and they still eat a project priced on the assumption of three. Faster iteration without a change control process is just faster margin loss.

Selling the freed capacity. In the example above, 120 hours came back. If they're sold, the agency is better off. If they're not, the agency has 120 hours of paid people with nothing to bill, and utilisation — the other lever on your real hourly cost — falls. This is the part most AI-efficiency conversations skip.

What to actually change

Move away from hourly billing where you credibly can. Fixed price and retainers let efficiency gains stay with you. This was always true; it's now urgent, because the gap between hours worked and value delivered is widening every quarter.

Reprice the work that got dramatically faster, deliberately. Some of your services genuinely should cost less now, and pretending otherwise invites a competitor to do it for you. Choose which ones, drop the price openly, and use them as an entry point rather than defending an old number until someone undercuts it.

Charge for judgement separately from production. Strategy, direction, selection, quality control — price these as their own line rather than bundling them into an hourly rate for making things. The making is what got cheap.

Raise the floor on small work. Tasks that used to be two-hour jobs and are now twenty minutes still carry a full load of briefing, review and admin. If you bill them by the minute you'll lose money on every one. Minimum engagement sizes and banded pricing exist for exactly this.

Track utilisation more closely than you used to. Faster delivery means capacity appears without warning. Whether that's a margin gain or a margin loss is decided entirely by your sales pipeline.

What doesn't change

Your loaded hourly cost still determines the floor under every price. Scope still has to be written down and defended. Estimates are still optimistic, and now they're optimistic about a smaller number, so the same 30% error costs the same proportion of a thinner project. Change requests still need to be raised the day they happen.

None of that got easier. If anything, tighter projects with smaller absolute margins make the fundamentals matter more, because there's less room between the quote and the loss.

One thing to work out this week

Take your three most repeatable services and estimate honestly how many hours each takes now compared with two years ago. Then calculate what your revenue on those services would be at today's rates and today's hours.

If the number is materially lower and your prices haven't moved, you already have the problem — you just haven't seen it yet, because the margin percentage on your dashboard still looks fine.