Chapter 9: The First 90 Days

By now, most agencies have already experimented with AI.

That is no longer the challenge.

The challenge in May 2026 is turning scattered experimentation into real operational capability without creating chaos, flattening quality or exhausting the team.

Because most agencies are now sitting in the messy middle.

Too many tools.
Too many disconnected experiments.
Random prompts everywhere.
Half-built workflows.
Three AI note takers fighting for custody of the same meeting.

Nobody completely sure what the “official” process is anymore.

This is where AI stops being exciting and starts becoming operational.

Good.

That is where the real value sits.

The first 90 days are not about “AI transformation”. That phrase already feels slightly tired.

The goal is simpler:

Move the agency from scattered AI usage to a controlled, working capability layer.

That is realistic.

And honestly, that is enough to create a serious advantage.

Step 1: Appoint an AI lead

AI cannot belong to “everyone”.

That means nobody owns it.

You need one accountable internal lead.

Not necessarily the most technical person. Usually the best AI leads are practical operators who understand:

• where the friction lives

• where quality breaks down

• where adoption will fail

• how the agency actually works under pressure

Their job is to:

• coordinate rollout

• keep standards aligned

• track adoption

• maintain workflows

• surface risks

• stop the agency disappearing into tool soup

They also need leadership backing and protected time.

Otherwise AI becomes another “important initiative” everyone quietly ignores while pretending they’re definitely getting to it next week.

Step 2: Simplify the stack

Most agencies already have too many tools.

This is now becoming an operational problem.

Different teams using different models, workflows and standards creates:

• duplicated work

• inconsistent quality

• governance risk

• workflow confusion

• rising costs

• terrible adoption

The answer is not more tools.

It is fewer, better-used tools.

A realistic first stack usually includes:

• one primary thinking model

• one creative generation layer

• one meeting/transcript tool

• one shared workflow/documentation space

Enough capability to move properly.

Not enough to create internal civil war.

Step 3: Find the real friction

Do not start with:

“What cool AI things could we build?”

Start with:

“What is slowing the agency down?”

Usually the answers are painfully obvious:

• briefing

• rollout

• versioning

• research synthesis

• meeting admin

• approvals

• QA

• handovers

This is where AI creates immediate value.

Not because it replaces people.

Because it removes repetitive operational drag.

A lot of burnout inside agencies is not caused by creativity.

It is caused by workflow stupidity repeated 400 times a week.

Step 4: Upgrade two or three workflows only

This is where agencies often lose the plot.

They try to AI-enable everything simultaneously.

Don’t.

Pick two or three workflows.

Map the current process properly. Remove the nonsense. Add AI where it genuinely helps. Add human judgement where it matters.

Then test it on real projects.

Good first candidates are:

• meeting → summary → brief

• research → synthesis → strategy draft

• creative exploration → review → refinement

• rollout → QA → delivery

This is where measurable gains start appearing.

Not from hype.

From operational clarity.

Step 5: Train by role

Generic AI training is mostly useless now.

People need workflows tied to the actual jobs they do every day.

Strategists need synthesis and interrogation systems.

Creatives need exploration and critique workflows.

Designers need rollout and consistency systems.

Producers need routing, QA and workflow management.

Account teams need briefing, expectation-setting and client communication.

Leadership needs to understand the commercial and cultural implications of all of it.

And learning needs to become social.

People should be sharing:

• prompts

• workflows

• failures

• useful discoveries

• before-and-after examples

• quality concerns

That is how capability spreads.

Not through one exhausted “AI person” quietly carrying the entire agency on their back.

Step 6: Add governance before problems appear

Governance gets ignored right up until somebody pastes confidential client strategy into the wrong system and everyone suddenly discovers the word “risk”.

Move earlier than that.

Keep it simple:

• approved tools

• data rules

• disclosure logic

• human sign-off

• red-line use cases

• workflow ownership

Nothing bloated.

Just clear enough that people know how to work safely and confidently.

Good governance creates confidence.

Bad governance creates panic and Slack messages with far too many exclamation marks.

Where external help actually matters

Most agencies do not need somebody to explain what ChatGPT is anymore.

They need:

• structure

• pace

• proven workflows

• governance patterns

• training

• operational perspective

• accountability

A good external partner compresses months of wandering into a few focused decisions.

The agency still needs to own adoption internally.

But outside perspective helps stop teams disappearing into endless experimentation without operational progress.

The real goal

By the end of 90 days, the agency should not be “fully transformed”.

That is fantasy deck behaviour.

It should be:

• more capable

• more aligned

• more confident

• less chaotic

• operationally sharper

• commercially clearer

The team should feel calmer, more supported and more excited about the work.

Leadership should finally have visibility into where AI is genuinely creating value.

That is a successful first 90 days.

Not hype.

Capability.