Where should an agency start with AI?

With production work that repeats across clients — adaptation, versioning, transcreation, briefing structure and reporting. It is low risk, high volume and easy to measure.

Will AI reduce the quality of creative work?

Not if the human sign-off line is explicit. AI should handle generation, structuring and scaling; direction, craft and client judgment stay with people.

How do we handle client data and consent?

Document which clients have approved AI-assisted production, what data may be shared with which tools, and who owns the decision. Regulated clients will ask for this in writing.

Should agencies reprice retainers after AI adoption?

That is a deliberate commercial choice: keep the margin, lower the price, or reinvest the hours in better work. Decide it before the pilot, because it shapes how the team behaves.

Guide · 6 min read

How to Implement AI in an Advertising Agency

By Fredrika Frenkiel, Head of Studio & AI Creative Operations at Lunar, founder of Code of Alfred. About my work

Läs den här guiden på svenska

Short answer

An advertising agency implements AI by rebuilding its production line, not its creative pitch. Start with the work that is repeated across every client, adaptation, versioning, transcreation, reporting and status, put AI there under a named owner, and leave idea, craft and client judgment with people. Agencies that lead with AI in the pitch and leave delivery unchanged lose credibility twice: once with the client and once with their own creatives.

1. Understand what makes an agency different from a brand team

An agency sells hours, reputation and craft. That makes AI adoption a commercial question, not only an operational one. If AI cuts the hours on a retainer, the agency has to decide whether it takes the margin, reprices the deal or reinvests the time in better work. Decide that before the first pilot, because the answer changes what you measure and how the team behaves. Teams that suspect AI is a cost-cutting exercise aimed at them will quietly refuse to use it.

2. Start where the work repeats across clients

The highest-value first targets are almost never the big creative idea. They are versioning a campaign across formats and markets, transcreating copy, building competitive and category summaries, structuring incoming briefs, drafting status and performance reports, and preparing pitch research. This work is repeated on every account, has a known good outcome and rarely carries brand risk, which makes it the safest and fastest place to prove value.

4. Build shared systems, not individual habits

In most agencies AI adoption is a handful of people with private prompts. That produces no compounding benefit and no consistency. Turn the good private habits into shared assets: a brief-intake structure everyone uses, a per-client tone and brand context that models are given automatically, an adaptation workflow with defined inputs and outputs, and a library of prompt chains owned by a named person who keeps them current.

5. Protect craft with a hard human line

Write down the stages where a person must sign off: strategic direction, final creative judgment, anything client-facing, and anything with legal or regulatory exposure. This is not a compromise with AI, it is what allows fast adoption. When people can see exactly what remains theirs, resistance drops and speed goes up.

6. Price and prove it

Track a real before and after on the workflows you changed: hours per adaptation round, time from brief to first presentable work, number of revision loops, and on-time delivery. Agencies that can show these numbers can defend a retainer, win a pitch on delivery capability, and reinvest saved hours in the work clients actually remember.

How to do it, step by step

01
Decide the commercial model first
Agree whether saved hours become margin, lower price or better work, before the first pilot.
02
Target work repeated across clients
Start with adaptation, transcreation, briefing, research and reporting rather than the creative idea.
03
Write the client and data rules down
Document approved clients, permitted data and sanctioned tools on one page.
04
Turn private prompts into shared systems
Convert individual habits into owned, versioned workflows with defined inputs and outputs.
05
Set the human sign-off line
Name the stages where a person must decide: direction, craft, client-facing and legal exposure.
06
Measure and price the gain
Track hours, revision loops and time to first presentable work, then use it commercially.

I have run this from both sides — in-house and agency — and the sequence rarely changes: repeated work first, governance in writing, craft protected, results measured. See how I approach this in practice.

Common questions

Where should an agency start with AI?: With production work that repeats across clients — adaptation, versioning, transcreation, briefing structure and reporting. It is low risk, high volume and easy to measure.
Will AI reduce the quality of creative work?: Not if the human sign-off line is explicit. AI should handle generation, structuring and scaling; direction, craft and client judgment stay with people.
How do we handle client data and consent?: Document which clients have approved AI-assisted production, what data may be shared with which tools, and who owns the decision. Regulated clients will ask for this in writing.
Should agencies reprice retainers after AI adoption?: That is a deliberate commercial choice: keep the margin, lower the price, or reinvest the hours in better work. Decide it before the pilot, because it shapes how the team behaves.

Last updated: 2026-09-09