There's a conversation happening in agency land right now that nobody wants to have out loud. Clients are starting to ask for "AI discounts." Not theoretical — one-third of agencies have already received explicit requests. Another half expect them soon.

And here's the thing: the clients aren't wrong to ask.

The Received Wisdom

The standard agency defence goes something like this: "AI is our tool, just like a designer uses Figma. You don't charge less because the designer is faster."

Plausible. Except it collapses under pressure.

Because when a client has paid you for 20 hours of work at £150/hour — and they find out that AI compressed it to five hours — the invoice still says £3,000. And it looks like a lie. Not because you did bad work. Because the unit you sold them was time, and time disappeared.

The Actual Problem

AI compresses delivery 3-4×. What is disputed is who captures that efficiency — the agency or the client?

Under hourly billing, the answer is: nobody. The agency cuts their own revenue 75% for equal or better output. The client gets a vague unease that they're being ripped off even when they're not.

The breakpoint math: a 20hr deliverable at £150/hr = £3,000. AI compresses to 5hrs. Under hourly billing: £750 invoice. That's a 75% revenue cut for the same strategic value. Nobody runs a business like that for long.

The rational exit is not to hide AI usage (agencies are trying this — it ends badly). It's to stop billing for effort entirely.

What the Data Says

Agencies with 60%+ of revenue on retainer show ~8 point net margin advantage over project-based peers. That's structural, not marginal.

The mechanics: retainer clients pay for access to an outcome-generating capability, not a deliverable count. The agency can apply AI to compress delivery without it hitting the invoice. Efficiency becomes a margin event, not a pricing conversation.

Second factor: retention. Retainer clients don't bid out every engagement. Embedded teams that genuinely understand a client's business are not easily replaced by a cheaper quote.

The embedded AI department model — positioning your agency not as a vendor you hire per project but as an AI capability you subscribe to — is the structure that matches the technology. AI agents work continuously, not per hour. The pricing model should reflect that.

What This Means This Week

If you're an SMB buying agency services on a project or hourly basis: you're leaving AI leverage on the table, and you have no visibility into what's actually being done on your behalf.

If you're an agency still defending hourly rates: the pressure will only increase. The only durable answer is moving the value anchor from time to result — before the client figures out they should be asking for 75% off.

The market is already pricing this: AI advisory retainers run £2,000-£20,000/month. Hourly AI consulting runs £100-£450/hour. The retainer isn't cheaper for the client in most cases — but it's cleaner, stickier, and produces better outcomes than a string of projects ever will.

The agencies that survive the AI pricing conversation aren't the ones with the best answer to "why aren't you charging less?" They're the ones who stopped selling hours before anyone thought to ask.