How should an agency bill for AI when it makes the work faster but costs money? In short: the cost of the technology goes openly on the invoice, and the value of the work is measured by the outcome – not by the hours AI has saved. Mix the two up and you give away margin.
Just how live the question is shows in a guest commentary in the German trade title Horizont on 7 October. Under the headline ‘Das Märchen von der kostenlosen KI ist vorbei’ – the fairy tale of free AI is over – Niklas Karbowiak and Tobias Flosdorf of Mutabor argue that professional AI work ‘speeds up many processes enormously but consumes considerable computing power’. The agency draws a clear conclusion: from now on it will show AI compute as a separate line item in its proposals – the ‘M-Credit’.
Why isn't AI free for agencies?
Because it brings three blocks of cost that never feature in the classic hourly rate. First, licences and compute, which grow with usage. Second, build and upkeep: prompts, workflows, data connections, training. Third, quality assurance, because every AI output needs someone to check it and stand behind it.
At the same time, the visible effort shrinks. A monthly report that used to tie people up for days is done faster. A media plan with three scenarios instead of one takes less time. To the client, that looks ‘cheaper’. For the agency, it is a shift: less manual work, more technology and judgement costs.
This is where the hourly-rate trap snaps shut. If you bill by time, you earn nothing from efficiency – quite the opposite. Every hour saved is an hour off the invoice.
What happens when agencies pass all their AI efficiency on to clients?
Every leadership team knows the scene. Procurement asks: ‘If the AI does that now, why are we still paying the same?’ The agency doesn't want to put the account at risk and lowers the fee. In the short term, the client is happy. In the long term, you have taught them to see your work as a commodity whose price falls with every new piece of software.
Yet what the client wants is rarely a cheaper fee. A client who wants better media planning wants more from their ad budget: smarter allocation, less wastage, faster optimisation. A few per cent off the agency fee barely registers next to a better-spent media budget. If you don't make that case, you are negotiating over the wrong number.
Why value beats price: price elasticity in plain English
Economics has a term for this pattern: price elasticity of demand. It describes how strongly demand reacts when the price changes. When elasticity is high, customers walk at the slightest price difference. When it is low, demand holds steady even if the price goes up.
Agency services aren't elastic or inelastic by nature. It depends on what the client is buying. If they are buying hours of interchangeable execution – resizing, standard reports, campaign set-ups – they compare prices, and AI makes that comparison even easier. If they are buying judgement, category knowledge and a relationship in which someone truly understands their goals, price plays a much smaller role.
Put bluntly: value outshines price when connections thrive. Pass AI savings on through price by reflex and you push your own work into the elastic corner – where, in the end, only the cheapest supplier wins.
Which pricing models work with AI?
No model is right in itself. But each one handles AI efficiency differently:
| Model | What AI efficiency does | Risk for the agency | Best suited to |
|---|---|---|---|
| Hourly rate (time and materials) | Fewer hours, less revenue | Productivity is penalised | Unclear scopes, early consultancy phases |
| Retainer / monthly fee | Efficiency initially stays with the agency | Renegotiation at every renewal | Ongoing work with a stable scope |
| Project or fixed fee | Margin rises when delivery gets faster | Scope creep eats the advantage | Clearly defined deliverables |
| Value-based / performance-related | The outcome is measured, not the effort | Dependence on factors outside the agency's control | Campaigns with measurable goals and clean data |
| Hybrid with an AI line item (credits) | Technology costs become visible and predictable | Needs explaining to procurement | Agencies with a high share of AI in delivery |
For most independent agencies, the hybrid is the realistic route: routine work on fixed fees, technology shown transparently, strategy and creative priced on value. Mutabor's M-Credit is one example of that middle building block.
Build your AI pricing logic in six steps
- Capture costs honestly. Licences, usage-based compute, set-up effort and the time spent on quality control – per client, not just per agency.
- Sort services into three baskets. Routine (reporting, monitoring, resizing), judgement (strategy, media planning, consultancy) and ideas (concept, creative). Each basket gets its own pricing logic.
- Put routine work on fixed fees. This is where AI saves the most time. A fixed price per report or per month stops efficiency turning into lost revenue.
- Make the technology visible. A separate line item or a credit allowance shows procurement what it is paying for – and ends the debate about why ‘the AI’ costs anything at all.
- Share efficiency gains, don't give them away. Offer to put the time saved into more scenarios, more tests or more frequent optimisation. The client gets more impact; you keep the margin.
- Prove the impact. If you bill on value, you have to show value: better budget allocation, faster learnings, less wastage. A quarterly conversation about results replaces the timesheet.
How do you explain AI costs to procurement?
Procurement thinks in units. Give it some. Instead of ‘We use AI’, the line becomes: ‘This item covers the compute for 40 campaign variants a month.’ The number is purely illustrative – what matters is that the client sees a service, not a black box.
A hypothetical example: a 25-person media agency produces monthly reports for twelve clients. If it automates the reporting, it could lower the fee – or put the freed-up time into weekly optimisation rounds the client never had before. The second offer is worth more, and it is much harder to compare.
That such time savings are real is shown by influencer agency SAMY: with Defy Agents, it cut its performance reporting from three days to one. The real management question starts afterwards – what happens to the two days you have won back? That is exactly what the Defy platform is for: the machine takes the grunt work, the team takes the work clients are happy to pay for.
What stays human – and becomes more valuable?
Timesheets, monthly reports, status meetings, manual resizing, competitor audits: AI can and should take all of that. What is left is judgement, relationships and accountability. Exactly the things clients are willing to pay more for – as long as you don't hide them inside a bundle of hours.
The fairy tale of free AI is over. So, for that matter, is the fairy tale of the agency that gets rich on hours.
