Get Started
Start an AI agency

Mission 1 of 8

Stop calling yourself an AI agency. Sell one fix to one kind of business.

Most 'AI agencies' have no clients because nobody can picture what they'd be buying. In an hour you'll pick a niche, match an offer to it and write the one sentence that describes it.

60 minBeginnerFree toolsFree mission

What you'll do

Score your niche candidates

Test each one for reach, a repeated painful process and an existing budget line, so you pick with evidence.

Match one of five offers

See what each offer replaces inside a business, then choose the one your niche's painful process needs.

Write your positioning statement

Turn the decision into one page and one line for your bio, signature and headline.

  • Check your idea against ten real businesses before you build anything
  • Know what you will turn down
Before you start
  • An hour with no tabs open except this one and a blank document
  • Any experience you have of an industry: a job you had, a family business, a client you've done odd work for. Even a summer job counts
  • Honesty about what you can build today. The technical missions linked below fill the gaps, but your first offer should be something you can deliver this month
Your progress0/6
01/06Why 'AI agency' is not a positioning0 done
STEP 01 of 06~8 min

Why 'AI agency' is not a positioning

Understand what a business owner is actually buying, so you stop selling the wrong thing.

Before you pick anything, you need to see why the usual pitch fails. Start from the buyer's side of the table.

Two messages, one inbox

Put yourself in the chair of someone who runs a 12-person accountancy practice. They've got a backlog of client onboarding forms, two people off sick and a self-assessment deadline.

A message arrives: 'We're an AI automation agency helping businesses harness AI to work smarter.' What do they do with that? Nothing. It doesn't name their problem, it doesn't name a price, and it asks them to do the work of imagining what you might do for them.

Now a different message: 'We set up a system for accountancy practices that reads the onboarding forms clients send in, pulls the details into your practice software and flags the ones with missing ID. Usually saves a practice your size about a day a week in the busy season. Want me to show you on one of your actual forms?'

Same tools underneath. One of these gets a reply.

What changed

Three things changed. The buyer is named (accountancy practices). The process is named (onboarding forms into practice software). The outcome is in their units (a day a week, in the busy season).

That's a positioning. 'AI agency' is a category, and nobody has a budget line called 'AI agency'.

Why the generic version hurts you too

There's a second reason the generic version fails, and it's on your side of the table. If you'll do anything for anyone, every project is a first project. You rebuild the same thing in a different shape, you can't reuse your proposal, your documentation or your pricing, and your margin disappears into learning curves.

An agency makes money when the fifth client gets roughly the same thing as the first, faster. That only happens if you've picked a lane.

Three ideas to keep in mind

  • Niche first, offer second. The niche tells you which process hurts. The offer is just the fix for that process.
  • Specific beats impressive. 'Enquiry responder for letting agents' will outsell 'AI transformation partner' every week of the year.
  • You can widen later. Nobody is locking you in. You're picking where to start so that the first five clients look like each other.
Tip
If the word 'AI' isn't needed to describe what the client gets, leave it out of the offer. 'Replies to every enquiry in two minutes, day or night' is the product. AI is how you build it.
Checkpoint
You can say, in one sentence, why a plumber with six vans wouldn't reply to 'we're an AI agency', and what they'd reply to instead.
Stuck? Ask the community
STEP 02 of 06~12 min

The niche method: reach, a repeated painful process, an existing budget line

Score three to five niche candidates so you pick with evidence instead of vibes.

Now you know a named buyer beats a category. This step gives you a way to choose that buyer without guessing.

The three tests

A niche is worth pitching when it passes three tests at once. Fail any one and you'll spend months finding out the hard way.

  • Reach. Can you get in front of a hundred of these businesses in the next thirty days? You need a list (there's a mission for that: find businesses to pitch), a place they gather (a trade body, a Facebook group, a local association, a trade show) and ideally one warm connection. An industry you've worked in scores highest because you speak the language and you know people.
  • A repeated painful process. Something tedious that happens every day or every week, is done by a human, and costs money when it's slow or missed. Enquiries unanswered. Invoices retyped. Calls going to voicemail. Quotes written from a blank page. If the pain only happens quarterly, there's no urgency and no retainer.
  • An existing budget line. They're already paying for this, in staff hours, in a tool, or in an outsourced firm. A letting agent paying a part-time admin £1,400 a month to handle enquiries has a budget line. You're not asking them to find new money, you're asking them to spend some of the existing money better. That's a much easier sale than 'invest in AI'.

Where candidates come from

You need three to five candidates to score. Here's where they usually come from, in order of how well they tend to work.

  • An industry you've worked in
  • An industry a parent, partner or close friend works in (you get honest answers about the pain)
  • A local industry you can walk into (estate agents, dental practices, garages, solicitors, recruiters, trade businesses, private clinics)
  • An industry that's hiring for the exact admin role you'd automate (search job boards for 'sales administrator' plus a sector and read what the job involves)

Industries that typically score well in 2026 for a solo UK agency: letting and estate agents, recruitment firms, trade businesses with an office (electrical, plumbing, roofing, HVAC), private dental and cosmetic clinics, independent accountancy practices, small law firms (conveyancing especially), vehicle dealers and garages, and B2B distributors with order admin.

All have reachable owners, daily repeated admin, and someone already paid to do it.

Score each candidate

  1. Write three to five candidates and score each one with this card. Be harsh. A niche you find interesting but can't reach scores low on the test that matters most.

    Niche scorecard (copy into a doc, one per candidate)
    Niche candidate: ________________________
    
    1. REACH (can I get in front of 100 of them in 30 days?)
       Where they gather (trade body, Facebook group, local association): ____
       Do I know anyone in it, or have I worked in it? (yes = +2)      ____
       Score 0-5: ____
    
    2. REPEATED PAINFUL PROCESS (does the same tedious thing happen daily?)
       The process, in one line: ____________________________________
       How many times a day/week: ____   Who does it now: ____________
       What goes wrong when it's slow or missed: ____________________
       Score 0-5: ____
    
    3. EXISTING BUDGET LINE (are they already paying for this?)
       What they pay today (staff hours, a tool, an outsourced firm): ____
       Rough monthly cost of the current way: £____
       Score 0-5: ____
    
    TOTAL /15: ____     (below 10 = move on, 12+ = pitch it)

    Add up the three scores. Below 10 means move on, and 12 or more means it's worth pitching.

    Warning
    Avoid niches where the buyer is a committee (NHS trusts, councils, large corporates). The sales cycle is six to eighteen months and you need cash flow in six to eighteen weeks. Government work is its own game: win government contracts covers it, later.
Checkpoint
Three to five scorecards filled in, each with a named process and a rough monthly cost of the current way of doing it, and one candidate scoring 12 or more.
If it breaksEverything scores about 9. Nothing stands out + 2 more
Everything scores about 9. Nothing stands out
Your reach scores are probably honest and your pain scores are guesses. Spend twenty minutes on one candidate: read ten job ads for admin roles in that sector and note what the job actually involves. The pain score will move, up or down, and that's the answer.
My best-scoring niche is one I find boring
Good. Boring industries have fewer agencies pitching them, owners who value reliability over novelty, and problems that are well understood. Interesting industries are crowded. Take the boring one and be the person who understands it.
I have no industry experience at all
Pick by reach. Which of these businesses could you physically walk into this week in your town? Estate agents, dentists, garages, recruiters. Local reach beats remote expertise when you have neither, because a face at the door gets conversations LinkedIn messages don't.
Still stuck? Ask the community
STEP 03 of 06~15 min

The five offers that sell in 2026

Know what each offer replaces in the business, what's typically in scope, and which one your niche needs.

You have a niche and the painful process that goes with it. Now you pick the fix. Agency offers that get bought share one trait: the client can see what it replaces.

Here are the five that reliably sell to small and medium businesses right now, with what each one replaces and what a typical scope looks like. You will sell one of them to begin with. Not five.

  1. Read down this table first. It shows each offer, what it replaces in the business, and a typical scope.

    The five offers
    OFFER                         REPLACES                           TYPICAL SCOPE
    1 Lead response + follow-up   Enquiries sitting unanswered       Form/email/WhatsApp in -> reply in
                                  for hours; manual chasing          under 2 min -> booked call -> 5-touch
                                                                     follow-up -> CRM updated
    2 AI receptionist (phone/chat) Missed calls, voicemail,          Answer, qualify, book, take messages,
                                  a part-time receptionist           hand off to a human, transcript to inbox
    3 Document + back-office      Retyping invoices, POs, forms,     Inbox/drive watcher -> extract fields ->
      automation                  reconciling by hand                push to accounting/CRM -> exceptions
                                                                     queue for a human
    4 Sales + CRM automation      Stale pipeline, no follow-ups,     Enrichment, stage reminders, quote
                                  quotes written from scratch        drafts, proposal docs, win/loss notes
    5 Internal knowledge          "Ask Dave" when Dave is off;       Policies, SOPs, product sheets -> a
      assistant                   new staff taking months to ramp    chat assistant with citations, in
                                                                     Slack/Teams/the intranet

    The notes below take the offers one at a time.

Lead response and follow-up automation

The best first offer for most niches, because the cost of the problem is easy to show. A business gets an enquiry at 7pm, nobody replies until 10am, and by then the prospect has booked with someone else.

The system takes enquiries from the web form, email, WhatsApp or a portal, replies in under two minutes with a relevant message (not 'thanks, we'll be in touch'), offers a booking link, follows up over five touches if there's no reply, and writes everything to the CRM.

The build is n8n plus an LLM for drafting replies plus their calendar and CRM. Mission 6 builds it end to end.

AI receptionist (phone and chat)

Replaces missed calls and voicemail. Answers, asks qualifying questions, books into the diary, takes a message for anything it can't handle, and hands off to a human mid-call when it should. Strong for clinics, trades, garages and anywhere the phone rings while everyone's busy.

The technical build is in AI phone receptionist and the website side in FAQ chatbot for clients.

Sell this second, once you've got a lead-response client who'll vouch for you, because the client's phone is a more sensitive thing to hand over than their web form.

Document and back-office automation

Replaces retyping. Invoices and purchase orders arriving by email get read, the fields extracted, the record created in Xero or Sage or the CRM, and anything uncertain goes to a human queue. Also onboarding forms, applications, timesheets, delivery notes.

Scope is one document type per package; the second type is the first expansion. Automate small business admin covers the building blocks.

Sales and CRM automation

Replaces the stale pipeline. Every new contact gets enriched (company size, website, LinkedIn), every deal stage has a reminder, quotes get drafted from a template with the client's details already filled, and lost deals get a one-line reason logged.

Sells well to recruiters, B2B services and anyone with a sales team of two or more who 'has a CRM' that nobody updates.

Internal knowledge assistant

Replaces 'ask Dave'. The company's policies, SOPs, product sheets and past proposals go into a search index, and staff ask questions in Slack, Teams or a web page and get answers with the source cited. Strong for businesses with twenty-plus staff or high turnover (hospitality groups, care providers, franchises).

The build is a RAG app: build a RAG app. The sale is harder than 1 to 4 because the saving is diffuse, so make it an expansion, not a first offer.

How to choose

For each niche on your shortlist, which offer maps to the painful process you wrote on the scorecard? Enquiries unanswered means offer 1. Phone ringing out means offer 2. Retyping means offer 3. Deals going cold means offer 4.

If it's not obvious, it's offer 1, because nearly every business with a website loses leads to slow replies, and you can prove it in twenty minutes by filling in their form and timing the reply.

Tip
Each offer also has a natural retainer. Lead response needs the sequences tuned and the CRM mapped as the business changes. The receptionist needs its script updated. Document automation needs new document types. This is why these sell as a setup fee plus monthly, which is mission 2.
Checkpoint
You've written one line per shortlisted niche: '[niche] needs offer [n] because [the process on the scorecard]'.
If it breaksMy niche needs something that isn't one of the five + 1 more
My niche needs something that isn't one of the five
It might, but check first. Most 'custom' requests are one of the five wearing a different hat. 'We need a system to handle tenant maintenance requests' is lead response and follow-up with a different form. If it really is new, it's a worse first offer because you can't reuse anything.
I can only build offer 1 today, but my niche needs offer 2
Sell offer 1 to that niche anyway. Clinics with phone problems also have web forms nobody answers. Get a client, deliver, then learn the receptionist build while you've got a paying reference.
Still stuck? Ask the community
STEP 04 of 06~12 min

Check it against ten real businesses before you build anything

Replace your assumption about the pain with what ten businesses in the niche actually do.

You've picked a niche and an offer on paper. Now spend twelve minutes making sure the pain is real, using evidence that's already public. You don't need to send anyone a fake enquiry to find out how they handle enquiries; the signs are on their website and in their reviews.

If your offer is lead response

  1. Open ten business websites in the niche. Use Google Maps for your area, or the list from find businesses to pitch.

    For each one, note these things:

    • Is there a contact form, a phone number, live chat, or a WhatsApp button?
    • Are the opening hours shown, and do they stop at 5pm?
    • Does the form say when to expect a reply?
    • In the Google reviews, search for 'reply', 'respond', 'called back' and 'never heard' and count the complaints.
  2. Then, honestly, as yourself, send one real enquiry to the two or three you'd most like as clients. Say who you are and that you help firms in their sector reply to leads faster, and ask if they'd be open to a 20 minute audit.

    The time it takes them to reply to a real message is a fair data point, and it's also your first outreach.

If your offer is something else

If your offer is the receptionist: ring two or three at 5:45pm as yourself, ask who handles calls after hours, and say why you're asking.

If it's document automation, look at the job ads in the sector for 'administrator' and count how many mention data entry, invoice processing or 'inputting'.

Put the evidence in a table

  1. Build a small table: business, channels offered, hours shown, reply promise, review complaints about responsiveness, and (for the ones you contacted) time to first reply.

    This table is the first thing you'll show a prospect, because it's evidence about their industry they've never seen. Don't name the businesses when you show it; 'seven of ten agents in your town have no way to reach them after 5pm' is the point.

Read the result

What you're looking for: if six or more of the ten show the gap (no after-hours channel, no reply promise, review complaints), the pain is real and common.

If nine of ten have chat, a promise and clean reviews, either the niche has solved it or you've picked the wrong process. Move to your second-scored niche and repeat. Twelve minutes to find that out now beats three months finding it out through a cold outreach campaign nobody answers.

Warning
Keep it honest. Every message you send says who you are and why you're writing. No pretend customers, no throwaway inboxes, no pretext calls. Apart from being the right thing to do, the businesses you test are the businesses you'll be pitching next week, and the first thing they'll check is whether you're straight with people.
Checkpoint
A table with ten rows and a count: how many show the gap your offer fixes, and two or three real conversations started. That count tells you whether to keep this niche or move to the next one.
If it breaksOne of the forms I used to introduce myself didn't work or sent no confirmation + 1 more
One of the forms I used to introduce myself didn't work or sent no confirmation
That's a finding, and a good one. 'Your contact form didn't send me a confirmation on Tuesday' is an opener that gets replies. Note it per business for mission 4.
I can't test my offer this way
Then talk to three people in the niche instead. Not a pitch, a conversation: 'What's the most annoying repeated job in your office?' You'll find the pain in three calls or find out it doesn't exist.
Still stuck? Ask the community
STEP 05 of 06~10 min

Write the one-page positioning statement

Turn the decision into the sentence that goes on everything you send.

You've checked your niche and offer against real businesses. This step puts them on one page so you can say them the same way every time.

A positioning statement is five short blocks. Who you serve, the pain in their words, what you set up, the one true piece of proof you have today, and what you don't do. Then it collapses into one line.

  1. Copy this template into your document and fill it in:

    Positioning statement
    WHO:    [industry] businesses in [region or size band]
            e.g. independent letting agents in the South East with 3 to 15 staff
    
    PAIN:   who [the repeated painful process, with its cost]
            e.g. lose viewing enquiries because nobody replies to portal leads
            after 6pm or at weekends
    
    OFFER:  I set up [the offer, in their words]
            e.g. an enquiry responder that replies in under two minutes, books
            the viewing into your diary and chases the no-shows
    
    PROOF:  [the one true thing you can say today]
            e.g. built on the same tools I used to automate my own admin;
            first three clients at a founding rate in exchange for a case study
    
    NOT:    [what you deliberately don't do]
            e.g. not websites, not "AI strategy", not generic chatbots
    
    ONE LINE (bio, email signature, LinkedIn headline):
    "I help [WHO] stop [PAIN] with [OFFER]."

Rules for filling it in

  • Who has an industry and a size or region. 'Businesses' is not an answer.
  • Pain uses words the owner would use; 'lose viewings because nobody replies to Rightmove leads after six' not 'suboptimal lead conversion'.
  • Offer describes what the client gets, not the technology.
  • Proof must be true today. If you have no clients, your proof is a founding offer (first three at a reduced rate in exchange for a case study) plus whatever you've genuinely built, even for yourself. Never invent a client.
  • Not is the line you'll be most tempted to delete. Keep it. It's what stops you saying yes to a website build in week three.

Use the one-liner everywhere

Put it in your LinkedIn headline, your email signature, the first line of your site and the first sentence on calls. If you can't say it without stumbling, shorten it until you can.

Tip
Read it to one person in the niche and watch their face. If they ask 'how does that work?', good, it's specific enough to be curious about. If they nod politely, it's still too vague.
Checkpoint
The one-page statement is written, the one-liner is in your LinkedIn headline and email signature, and the NOT line names at least two things you'll turn down.
If it breaksMy one-liner is forty words long + 1 more
My one-liner is forty words long
Cut the offer to its outcome. 'I help letting agents reply to every enquiry in two minutes, day or night.' The how goes in the conversation, not the headline.
I want to serve two niches
Write two statements and pick the one with the higher scorecard. Run it for ninety days. If it isn't working by then, switch to the second. Running both at once halves your reach in each and you'll learn nothing from either.
Still stuck? Ask the community
STEP 06 of 06~3 min

Lock it in and move to pricing

Decide, write it down, and don't reopen it for ninety days.

  1. Write at the top of your working document: niche, offer, the date, and 'review on' ninety days from now.

    Decisions reopened weekly never get tested. You'll be tempted to switch after the first ten emails get no reply. Ten emails is not a test. A hundred, with the audit approach in mission 4, is.

    You now have the thing most 'AI agencies' never get: a named buyer, a named process, and a sentence that describes the fix in the buyer's own words. Next, turn it into a package with a price: productise the offer and price it.

Checkpoint
Niche, offer, date and a 90-day review date are written at the top of your document.
Stuck? Ask the community

Your first wins

It works. Now do something with it.

Pressure-test the niche

Give the model your scorecard and let it argue with you:

Prompt
I'm starting a small AI automation agency in the UK, solo. My shortlisted niche is [niche], the repeated painful process is [process], they currently handle it by [current method] at roughly £[cost] a month, and my offer is [offer]. Argue against this choice as a sceptical business owner in that niche would: what would make them not buy, what they've already tried, and what the real objection would be. Then tell me the single piece of evidence I should gather this week that would settle it.

Write the one-liner ten ways

Get options, then pick the one you can say out loud:

Prompt
Write ten versions of a one-line positioning statement for a UK solo agency. Who: [niche and size]. Pain: [pain in the owner's words]. Offer: [what the client gets, no technology words]. Each under 18 words, plain English, no jargon, no buzzwords like 'harness', 'empower' or 'transform', no exclamation marks. Then tell me which three a sceptical owner would find most believable and why.

Undo it cleanly

How to remove everything this mission installed.

Nothing installed, nothing published. The only thing you sent was a few honest introductions.

Next mission · Start an AI agencyProductise the offer: one fixed-scope package, three tiers and a price you can defend
→