Almost every agency uses AI now. Nearly 9 in 10 organisations use it regularly in at least 1 business function, according to McKinsey's State of AI 2026 survey. Yet only about 6% qualify as high performers, meaning they credit AI with at least 5% of their earnings and call its impact significant.

So why does AI for agencies so often stall? In our experience, it isn't the tools. It's that the knowledge those tools need sits in people's heads, and the repeat work has never been written down as a standard.

We know because we were the same. Most of our growth process sat with 1 person, very little was written down, and every answer meant asking the founder. So we fixed it, and our own sales and marketing now run on the result.

This post explains what that system is, why it works and how you can try it. We're also looking for a few partner agencies to run a pilot with, so read on if that sounds like you.

Why AI for agencies stalls after the first few prompts

AI stalls when each person uses it in their own way. One person has a good prompt for proposals. Another has a different one for reports. Little is shared, so little improves as a team.

McKinsey's figures show the same gap at scale. About 8 in 10 respondents say AI has improved their own productivity, but only 37% report any earnings impact for the organisation, about the same as last year. Personal speed gains aren't turning into business results.

Many agencies have a particular version of this problem. Pricing lives in old proposals. Positioning lives in the founder's head.

Every new proposal starts from a blank page, and every question ends up with the same person. AI can't fix that on its own, because it has nothing of yours to work from.

The risk isn't only wasted time. A drafting tool that doesn't know your pricing rules can invent an offer, quote a price you wouldn't give or write in a tone that isn't yours. That's speed without quality, and it creates more checking work, not less.

Speed is easy. Zero quality loss is the hard part

You keep quality by giving AI your standards, not by hoping it guesses them. Out of the box, an AI tool answers from the internet. It doesn't know your offers, your pricing rules, your tone of voice or what a good result looks like for your agency.

As technologists, we haven't asked what AI tools can do to save us an hour. We've asked how to customise them so they give us the most speed and the best quality at the same time. Giving up quality for speed isn't a trade we're willing to make.

That comes down to 2 rules. First, the AI works from your own documents. Second, a person checks the result before it goes out.

The AI does the heavy first draft. Your team keeps the judgement, the pricing sign-off and the final say.

In practice, every task has 3 things written down before AI touches it: the inputs it needs, the standard the output must meet, and who signs it off.

3 things that make AI workflows for agencies work

Here's what we set up inside an agency, and what we've already built for ourselves:

• An AI proposal workflow: your sales process turned into a repeatable routine.
• An agency knowledge base: one shared home for your commercial knowledge, connected to your AI tools.
• Skills: your most repeated tasks, written down so AI does them your way.

An AI proposal workflow

Start with proposals, because they cost senior time and vary the most from person to person. We build a single services and pricing reference, so nobody invents a new offer on the fly. Each offer has a name, a price or a way to calculate one, a minimum and a clear next step.

We then write the pricing rules down. That covers scoping before quoting, the formula for a build price, a floor test that stops a proposal going out below your minimum and the discounts you allow. These are the questions people usually ask the founder.

Next come a proposal template and a pre-send checklist. The last piece is the drafting flow. Call notes and your scoping document go in, and a first draft priced against your own rules comes out, ready for a person to check and send.

An agency knowledge base, or data room

Ours is called the data room. It's one shared folder that holds your strategy, positioning, messaging, pricing, briefs and decisions, plus what you learn from wins and losses. We connect it to your AI tools, so answers come from your documents rather than the open internet.

A typical structure covers strategy, positioning and messaging, services and pricing, briefs, a decision log, win and loss reviews and a scorecard. We also set clear rules for which document wins when 2 disagree, so the AI doesn't pick the wrong one.

Your team still uses their own judgement. They just don't have to ask the founder first, because the standard for each process is written down and easy to find. Training covers day to day use: finding answers, drafting from the data room and keeping it current.

Skills for your repeat work

A skill is a set of written instructions that an AI tool follows to do a recurring task your way. It carries your tone of voice, your quality checks and your approval steps, so the output looks the same whoever runs it.

Your most repeated work tells us where to start, and the best 2 or 3 tasks become skills. Our own examples include writing blog posts to a brief and SEO rules and producing recurring client reports.

Website audits follow a fixed scoring method too. Because each skill follows fixed steps and checks, the output stays consistent whoever runs it, and a person still reviews it.

A short guide to maintaining and adding skills stays with you, so the library keeps growing after handover.

2 examples from our own work

At Untapped, we've always worked in meticulous detail. What's changed is that working to those standards is quicker than it has ever been. Here are 2 examples, with real numbers.

Estimating project costs before discovery

We used to spend 4 to 8 hours estimating a project before discovery. That gives the client a range, with final costs confirmed after discovery.

Now we use a custom skill trained on 9 years of our own project data, built to copy our manual process. It creates the estimate spreadsheet in about 30 minutes. Then the right members of the team spend 30 to 60 minutes on a senior review.

That review matters. Normally about 10% of the estimate needs to change, and that 10% is where years of experience comes in.

Website audits

We're proud of the audit process we've built over the years. It used to take 3 days to cover UX, UI, brand, SEO, GEO, performance, accessibility, backlinks, GDPR, security and competitor research in enough detail that we could stand behind it. That averaged about 2 hours per category, sometimes more.

We've now automated most of it. The tool uses trusted third party services such as Google Lighthouse and Keywords Everywhere, combined with what we've learned from more than 300 audits over 9 years. It does most of the process based on our training, and then it takes about 2 hours to verify the key findings.

What both examples have in common

Both are faster, and both still get a human review. Speed is the benefit. The review is what lets us stand behind the result.

Both are also only as good as what we fed in. The estimating skill works because it learned from 9 years of real projects. The audits work because they draw on more than 300 past audits.

A thin or messy knowledge base gives you thin, messy answers, just faster. That's why we start by writing your knowledge down.

What changes when the knowledge isn't in one person's head

The founder stops being the bottleneck. When positioning, pricing and process are written down, a new starter can find the answer, and a proposal can start from a solid draft instead of a blank page.

It also matches what the research says about the few organisations that get real value. McKinsey found that nearly three quarters of high performers fundamentally redesign their workflows around AI, compared with about one quarter of everyone else. They don't just add a tool to the way they already work. They change the work.

Size matters here too. McKinsey reports that about one third of organisations under $1 billion in revenue, which covers nearly every agency, are scaling AI across the business, against 54% of larger ones. In our view, a smaller team has fewer processes to untangle, which gives it a real chance to close that gap.

Across all organisations, 44% now report scaling AI across the enterprise, up from 38% a year ago. Using AI is common. Fewer than half have scaled it across the business.

Be ready for the up front effort. Writing things down is the slow part, and it's the part most teams skip. It's also the part that pays back each time someone uses it.

Try a simple test this week. Pick the task your team repeats most. Write down the steps in order, the standard a good result meets and who signs it off.

Give that to your AI tool along with the relevant documents, then compare the result with your last attempt. You'll see quickly where the gaps in your written knowledge are.

What stays with your people

AI handles the first draft and the searching. People keep the decisions. Here's how we split it:

• AI does: finds answers in your documents, drafts proposals and reports, checks work against your checklist, keeps formatting and structure consistent.
• People do: set the price, make the creative call, own the client relationship, check the draft and press send.

‍

Take a proposal. The AI reads your call notes, pulls the right offer from your reference, applies your pricing rules and drafts the document. A person then checks the scope, confirms the price and adds the personal touches that win the work. Nothing goes to the client until someone has approved it.

That split is what protects quality. The system makes the routine work faster and more consistent, and it leaves your team more time for the parts clients actually pay you for.

Who we're looking for to run a pilot

We're looking for a few agencies, consultancies and fractional leaders to run a pilot with. It suits teams of roughly 5 to 50 people where proposals, pricing and know-how still depend on the founder. It's especially useful if you already use AI in an ad hoc way and want it to work as a system.

The pilot runs in 4 steps, after an intro call where we learn what you sell, how proposals get made today and where your team loses time:

1. Diagnostic workshop: we map your sales process, the knowledge that needs writing down and the repeat work worth turning into skills.
2. Setup sprint: we build the data room, the proposal workflow and the first skills with your team.
3. Handover: your team runs live proposals and tasks through the system, with us alongside.
4. Review: we look at what changed and write it up together.

‍

You finish with a working data room in your own tools, a pricing reference and proposal workflow, and 2 or 3 skills built for your agency. You also get training for your team and a written guide to running and extending it all.

If you'd like, regular reviews can follow to add skills and keep the data room current. Because every agency is different, we adapt the system to yours rather than hand over a template. Your offers, your tone and your tools shape the result.

It works for us. We haven't yet shown it across a range of agencies, and we'd rather learn that with a few partners than guess. The pilot is a paid engagement, and we'll talk through the details with you on a call. We'd also ask for honest feedback and, once you're happy with the results, a named case study.

Where to start

If you take 3 things from this, make them these. Write your knowledge down in one place. Turn your repeat work into a standard. Then connect your AI tools to both, so AI for agencies becomes a system your whole team uses, not a set of individual habits.

If that sounds like your agency, or you know someone it would suit, email us at hello@thisisuntapped.com. Tell us a little about your agency and where your team loses the most time, and we'll take it from there.

FAQs

How can agencies use AI without losing quality?

Give the AI your own standards to work from, and keep a person in charge of the final check. When your pricing rules, tone of voice and process live in documents the AI can read, its first drafts start much closer to your standard. Your team then spends their time reviewing instead of starting from scratch.

Is AI for agencies only worth it for large teams?

No. Smaller teams can feel the benefit quickly, because proposals, pricing and know-how often sit with the founder. McKinsey's data shows smaller organisations are scaling AI more slowly than large ones, which means there's room to catch up.

What is an AI proposal workflow?

It's a repeatable routine for turning a sales conversation into a proposal. It combines a services and pricing reference, written pricing rules, a template and a pre-send checklist. AI drafts from your call notes, and a person checks and sends.

What is an agency knowledge base?

It's a shared home for your strategy, positioning, pricing, briefs and decisions, connected to your AI tools. The AI answers from those documents, not from the open internet. We call ours a data room.

What are AI skills?

Skills are written instructions that an AI tool follows to do a recurring task your way, with your tone, quality checks and approval steps. They keep the output consistent whoever runs the task. Blog posts, website audits and client reports are good examples.

Does work done with AI still need a human review?

Yes. At Untapped, our estimates and our website audits both get a senior human review before they go anywhere. On our estimating skill, about 10% of the first draft normally changes after review, and that is where experience comes in.

Is AI only as good as the data you give it?

Yes. Our estimating skill learned from 9 years of project data, and our audits draw on more than 300 past audits. Without that, you'd get generic output, just faster. That's why we start by writing your knowledge down.

Do we need to be technical to set this up?

No. We build the data room, the proposal workflow and the first skills with your team, then train them to run and extend it. You leave with a written guide to refer back to.