How to choose a niche for your agency

You started an agency that can help almost any business, and that is why nobody can tell if it is for them. Here is how to choose a niche for your agency in 3 weeks: write 3 guesses from what you have already done, check each has enough businesses that pay for marketing, then test them side by side with 100 personal messages a day. Most new agency owners pick a niche from a list of "best niches" or argue with themselves for months, and keep sending messages to everyone in the meantime.

Do these three things

About 5 to 6 hours a day, every day, for 3 weeks. Decide after 2,100 messages, then keep that niche for the next 8,400 before you judge it again.

Show the three actions in fullHide the full list

1. 1Today: write 3 niche guesses, each one kind of business, its size and one problem you fix, starting from where you already got results. Keep the 2 or 3 where you can list 3,000 businesses by name that already pay for marketing.
2. 2Every day: send 100 personal messages, split evenly across your niches, each pointing out one fix and offering a free 5-minute video. Follow up with everyone 3 times.
3. 3Every Sunday: count conversations and calls per 100 messages for each niche. After 2,100 messages, keep the niche with the most calls and put it on your website and LinkedIn.

Get help doing these

ChatGPT or Claude reads this guide and works through it with you.

Do this with ChatGPT(opens in a new tab)Do this with Claude(opens in a new tab)

1. List where you already got results and write 3 niche guesses

Start your niche from the work you have already done, not from a list of "most profitable niches". Write down every business you got a result for, then turn the best ones into 3 one-sentence guesses: a kind of business, its size, and one problem you fix.

A niche is one kind of client with one problem you fix, and it sets your positioning. "B2B software companies of 2 to 15 people where the founder still writes every post" is a niche. "Small businesses" is not, and neither is "SEO" on its own. It is your ideal customer profile as an agency, and it decides who you message, what your offer says and what your price can be.

Pick the kind of client first, then the problem. In Promethean Research's 2026 guide to agency growth, 86% of agencies call themselves specialists, up from 30% ten years ago, and it says industry depth is the safer choice as AI makes single services cheaper to do. A service on its own ("we do SEO") competes with every freelancer and every AI tool. A kind of client plus a problem ("trial signups from Google for small SaaS teams") competes with far fewer.

Fill in a sheet with every business you have done marketing work for: current clients, past clients as a freelancer, employers, and your own projects. For each, write the kind of business, its size, what you did and the result, with a number if you have one. Add a column for how many people you know who work at businesses like it, because those people give you your first introductions.

Then write 3 guesses with this rule:

• Guess 1: the kind of business where you got the best result.
• Guesses 2 and 3: close neighbours of it: the same kind of business at another size, the same buyer in a nearby industry, or another kind where you know at least 20 people.
• If you have no results anywhere: pick 3 kinds of business you can find thousands of by name, and that you understand as a customer or an employee. Starting with no experience has its own guide.

[Kind of business] with [size: people or revenue] who [the problem you fix, in their words], and today they [what they do instead].

Ask ChatGPT or Claude to suggest neighbours from your sheet: paste it in and ask for 5 kinds of business close to your best result. Pick the 2 where you know the most people.

If you already sell websites or IT support: your current clients are guess 1. Write guess 1 as the kind of business that already pays you, and the marketing problem they ask you about most. Then turn each website project into monthly work for them.

• Tool: a Google Sheet, and ChatGPT or Claude. Both free to start.
• Time: 45 minutes.
• You will have: a sheet of past work, and 3 one-sentence niche guesses.
• It worked if: each guess names a kind of business, a size and one problem, and at least one guess comes from a real result.
• Common mistake: picking a niche because a video called it "high ticket", when you have never worked with that kind of business. The other mistake is writing a service ("Facebook ads") instead of a kind of client.

2. Check each niche has 3,000 businesses that already pay for marketing

Check each guess with three tests before you send a single message: you can find the businesses by name, they already pay for marketing, and they talk about the problem you fix. Keep the 2 or 3 guesses that pass, and drop the rest today.

Run each test on a timer, so the check takes one afternoon, not a month:

1. Can you find them? List 30 by name in 30 minutes. For each, find the business, the owner or decision maker and a way to reach them. Step 5 needs 100 messages a day, so a niche needs about 3,000 businesses you can reach. If 30 take you more than 30 minutes, the list will run dry in week 2. For local businesses, search Google Maps for "[kind of business] in [town]" and count the towns you could cover. For online businesses, search LinkedIn for the owner's job title and the kind of company. Ask ChatGPT or Claude for 30 that fit your sentence, then open each website yourself, because some will be wrong or closed.
2. Do they already pay for marketing? At least 3 of your 10 first names should. A business that already spends money on marketing can afford you, and does not need to be convinced that marketing matters. For 10 of your 30, check three things: search their website in Google's free Ads Transparency Center and their name in Meta's free Ad Library to see if they run ads; look on LinkedIn for a marketing job they posted; and look for an agency or freelancer they work with. Then check Clutch, where agencies list the industries they work with in their "Industry Focus". Other agencies already serving this kind of business are a good sign there is money in it. Note them for your competitor analysis.
3. Do they talk about the problem? Find 10 posts from the last 6 months. Search Google, LinkedIn posts and the communities where they ask for help, such as a subreddit or a Facebook group. Look for owners of this kind who complain about the problem you fix. Copy their exact words into the sheet. You will use them in step 3.

Score each guess pass or fail on each test. Drop any guess that fails test 1 or test 2. If all 3 pass, keep all 3. If only 1 passes, write a new neighbour from step 1 and test it too, so you always compare at least 2. One niche on its own tells you nothing, because you cannot see what "better" looks like.

• Tool: Google Maps, LinkedIn search, Google's Ads Transparency Center, Meta's Ad Library, Clutch, and ChatGPT or Claude. All free.
• Time: about 2 hours: 30 minutes for each list of 30, then 10 minutes for the other two tests on each guess.
• You will have: 2 or 3 guesses that passed, each with 30 named businesses and 10 posts in the owners' words.
• It worked if: every guess you keep has 30 names found in under 30 minutes, at least 3 of 10 that pay for marketing, and 10 posts about the problem.
• Common mistake: choosing a niche because "they have money", when nobody there spends it on marketing. A rich industry that never hires agencies is the slowest one to sell to.

3. Write one offer and one message for each niche, using the owners' words

Write one offer line and one first message for each niche you kept, using the words its owners used in the posts you found. The messages must be equally good, or the test measures your writing, not the niche.

The offer line is your value proposition for that niche: who it is for and the result they get, in their words. Take the phrases you copied in step 2. If accounting firm owners write "all our work comes from referrals and it's drying up", the offer says "new clients from Google, so you are not waiting on referrals", not "local SEO".

Keep everything else the same across niches: the same length, the same free 5-minute video offer and the same follow-ups. Only the first line, the offer line and the fix change. If one niche gets a better message, you will not know whether the niche or the message won.

[Your agency] gets [kind of business] with [size] [the result, in their words], so they [what changes for them].

Subject: [their business] on [Google / ChatGPT / LinkedIn / your homepage]Hi [first name],[The one fix you noticed for this business, in one true line.]I work with [kind of business] who [the problem, in their words]. I think [one change] would help.Want me to record a free 5-minute video showing how? No call needed.[Your name], [your agency], [phone]
[Postal address]
Not for you? Reply "no" and I won't write again.

Give ChatGPT or Claude your 10 posts for each niche and ask for the 5 phrases owners use most about the problem. Write the offer line yourself from those phrases. Then ask AI to rewrite the message for each niche, changing only the problem line, and check each version is the same length.

Write down the price you will say on calls: the Starter price from your three options. Use the same price for every niche, so step 6 can count who says it is fine.

• Tool: a Google Doc, and ChatGPT or Claude.
• Time: 1 hour 30 minutes, about 30 minutes for each niche.
• You will have: one offer line and one first message for each niche, with the same video offer and the same price.
• It worked if: you can read the three messages side by side and the only differences are the first line, the problem and the fix.
• Common mistake: writing your best message for the niche you already want and a quick one for the others. The test then tells you what you already believed.

4. Add 100 businesses a day to your list, split across your niches

Add 100 new businesses to your list every day, split evenly across your niches, each with an owner's name, a way to reach them and one fix. 100 messages a day for 3 weeks needs 2,100 names, so building the list is a daily job from day 1.

Use one sheet for all your niches, with a "Niche" column, so the Sunday count in step 6 is one filter. With 3 niches, add 34, 33 and 33 a day. With 2, add 50 and 50. Start with the 30 names you found in step 2.

If your niche is local businesses: search Google Maps for "[kind of business] in [town]", one town at a time. The owner's name is often on the website's About page or in their replies to Google reviews.

If your niche is online or software businesses: search LinkedIn for the founder or owner title plus the kind of company. A company that just posted its first marketing job may take an agency instead.

For each row of your lead list, find the owner or the person who decides, and find their work email if it is not on the site. Building a free B2B list has its own guide. Then note one fix in 3 minutes or less: something on their Google listing, the top of their homepage, their ads, or whether ChatGPT names them when you ask for their kind of business. That last one is the start of AI visibility work, and most owners have never checked it.

Let AI find the candidates: give ChatGPT or Claude your offer line and a town or a market, and ask for 30 businesses that fit, with their websites. Open each website yourself before a row goes in.

Niche | Business | Website | Owner name | Email | Phone | LinkedIn | One fix | Date contacted | Reply | Call | Price OK? | Follow-up due

• Tool: Google Sheets, Google Maps, LinkedIn search, and ChatGPT or Claude.
• Time: about 1 hour 30 minutes a day.
• You will have: 100 new named owners a day, split evenly by niche, 2,100 by the end of week 3.
• It worked if: every row has a person's name, a way to reach them and one true fix, and no niche falls more than 10 names behind the others.
• Common mistake: letting the easiest niche fill the list. If one niche is much harder to find, that is a result too: write it down, because a niche you cannot find 100 of a day will not feed your agency.

5. Send 100 personal messages a day to test your niches

Every day, send 100 personal messages, split evenly across your niches, and follow up with everyone 3 times. This is the test: the niche whose owners reply and book calls is the one to keep.

100 a day is the point. In Instantly's benchmark of cold email sent in 2025, the average reply rate was 3.43%, counting every reply, "no" included. So expect about 1 or 2 conversations for every 100 owners, after follow-ups. At 10 a day, each of 3 niches would get about one conversation every 3 weeks, and you would still be guessing in 3 months. At 100 a day, each niche gets 700 messages in 3 weeks, enough to see a difference.

Let AI draft. Paste 10 rows at a time into ChatGPT or Claude, each with its one fix, plus that niche's message from step 3. Ask for one message per owner, under 80 words, with no link. Read each draft, make the first line true, then send it. That is about a minute a message. Send each niche's share every day, so no niche gets a better day of the week.

Spread the 100 inside each platform's limits. Do not lower the number; spread it:

• Email. Gmail requires every sender to have SPF or DKIM set up and to keep spam reports under 0.3%, and since November 2025 it has been rejecting more mail that breaks these rules. Its guidelines also say to raise volume slowly. So send cold email from a second domain, and warm it up from 5 to 10 a day per mailbox. The full setup, 3 mailboxes rising to 25 a day each, is in the first-client guide.
• LinkedIn. A free account can add a note to only 3 to 5 connection requests a month (LinkedIn's help pages give both numbers). If you send too many requests, LinkedIn stops you for about a week. Send requests with no note to owners who fit, and message them once they accept.
• Phone and introductions. While your mailboxes warm up, fill the rest of the 100 with calls to local owners and introduction asks to people you know in each niche. Each counts as one message for that niche.

The law. In the US, the CAN-SPAM Act makes no exception for business email: every email needs your postal address and a clear way to say no, and you must honour a "no" within 10 business days. Check the rules in the UK and EU before you email owners there.

Follow up in the same thread on day 3, day 7 and day 14, each time with one new line. Stop the moment someone says no. In Instantly's data, 42% of replies came from a follow-up, not the first email. When someone says yes, record the video with Loom's free plan, which allows 5-minute recordings and up to 25 videos; delete old ones as you go. When someone watches it and asks a question, offer a 20-minute discovery call.

On every call, say your Starter price once: "Most clients start at $[Starter] a month." Write down if they say it is fine, too much, or nothing. That is the "Price OK?" column.

• Tool: Gmail from your warming mailboxes, LinkedIn, your phone, Loom (free), and ChatGPT or Claude to draft. Log everything in your step 4 sheet.
• Time: about 2 hours a day for 100 first messages, 1 hour 30 minutes for follow-ups once they start, and 30 minutes for videos and calls. Take the hours from your agency's marketing budget.
• You will have: 700 messages a week, 2,100 after 3 weeks, split evenly by niche, each logged with its replies and follow-ups.
• It worked if: at least one niche gets 1 conversation (a yes to the video, a question back, or a call) for every 100 messages, after follow-ups.
• Common mistake: sending the same text to everyone, or all 100 from your main address on day 1. Both end in spam folders, and your main address takes your client email with it. The other mistake is stopping a niche after 50 messages because nobody replied yet.

6. Count calls per niche every Sunday and pick one after 2,100 messages

Every Sunday, count messages, conversations, calls and price reactions for each niche. After 2,100 messages, 700 per niche if you test 3, keep the niche with the most calls per 100 messages and stop the rest.

Count per 100 messages, not in totals, so a niche that fell behind on its share is judged fairly. It is the same idea as tracking which channel brought each deal, with niches in place of channels. Use your sheet's filter on the "Niche" column.

Week | Niche | Messages sent | Conversations | Conversations per 100 | Calls | Calls per 100 | Price OK (of calls) | New clients

The decision rule after 2,100 messages:

1. Stop any niche under 1 conversation per 100. Owners there do not care enough about the problem, or you cannot reach the right person.
2. Keep the niche with the most calls per 100. A call means an owner gave you 20 minutes, which is the best sign you have before a client.
3. If two niches are within 0.5 calls per 100, keep the one where more people said your price was fine. If that is a tie too, keep the one where you found 30 names fastest in step 2, because it will feed 100 messages a day for longest.
4. If no niche reaches 1 conversation per 100, do not change niches yet. Change the one fix you point out and the first line, and send 700 more to your best niche. If it is still under 1 per 100, write a new neighbour from step 1 and test it against your best one.

Any client who signs during the test counts more than any number. If a niche brings a paying client, keep it, unless another niche brought two.

Write down what owners said on calls, in their words, the same way you would take notes in a customer interview. They are the words for your website, your LinkedIn profile and your next proposal.

• Tool: Google Sheets.
• Time: 30 minutes each Sunday.
• You will have: one row per niche each week, and after 2,100 messages, one niche marked keep.
• It worked if: you can say which niche won and give the numbers that made it win, in one sentence.
• Common mistake: keeping the niche you liked best when another one got twice the calls. The other mistake is judging on likes, profile views or replies that said "no".

7. Put your niche on your website and LinkedIn, and keep it for 8,400 messages

Once you pick a niche, say it everywhere a client looks: the top of your website, your LinkedIn headline and your offer. Then keep it for the next 8,400 messages, about 12 weeks at 100 a day, before you judge it again.

Rewrite three things today, using the offer line that won and the words owners said on calls:

1. Your homepage headline: the result and the kind of client, then one line of proof and your starting price. How to write the top of your site has its own guide.
2. Your LinkedIn headline: the same sentence, plus the services you sell, so your LinkedIn profile brings in clients from the niche.
3. Your offer: one package for this niche, priced from your own numbers.

Then move all 100 messages a day to the niche, test new channels only for that niche, and keep finding clients there with the same routine. Judge clients, not conversations, after 8,400 messages. A niche needs that long: conversations come in days, but a client takes a call, a proposal and a yes. Write back after 30 days to everyone who said "not now", and turn your first client in the niche into a case study you can point to in every later message. Until you have one, win clients without case studies with a short paid first month.

Keep your current clients, even the ones outside your niche. Do the work well and keep billing them. A niche decides who you market to, not who you serve. Agency consultant Kurt Schmidt (August 2026) warns against owners who "fire their misfit clients on day one", before the new niche brings work to replace them. Say yes to a new client outside your niche only if they pay at least your lowest price per hour, and do not market to their kind.

• Tool: your site editor and LinkedIn.
• Time: about 2 hours, once.
• You will have: one sentence that is the same on your website, your LinkedIn headline and your offer, and 100 messages a day going to one niche.
• It worked if: a stranger who sees your homepage or LinkedIn for 5 seconds can say what kind of business you help, and 4 of 5 people you ask get it right.
• Common mistake: changing niche after 2 weeks because the first replies are slow. The test already showed this niche replies more than the others. Give it 8,400 messages before you test again.

Your niche checklist for the next 3 weeks

0 of 13 done

Work through this list with an AI assistant

It drafts the work and keeps track of what is done.

Do this with ChatGPT(opens in a new tab)Do this with Claude(opens in a new tab)

Step 1 · List where you got results and write 3 niche guesses

• List every business you have done marketing work for, with the result you got each one.
• Write 3 one-sentence niche guesses, starting with the kind of business where you got the best result.

Step 2 · Check each niche has 3,000 businesses that pay for marketing

• Run the three tests on each guess and drop any that fails the first or second.

Step 3 · Write one offer and one message for each niche

• Write one offer line and one first message for each niche you kept.

Step 4 · Add 100 businesses a day, split across your niches

• Every dayAdd 100 named businesses to your list, split evenly across your niches.

Step 5 · Send 100 personal messages a day to test your niches

• Set up a second domain for cold email and start warming it up.
• Every daySend 100 personal messages split evenly across your niches.
• Every daySend the follow-ups due on day 3, 7 and 14, record the videos people asked for, and hold your calls.

Step 6 · Count calls per niche every Sunday and pick one

• Every SundayFill in one row per niche in the Sunday sheet.
• After 2,100 messagesStop any niche under 1 conversation per 100 and keep the one with the most calls per 100.

Step 7 · Put your niche on your website and LinkedIn

• Rewrite your homepage headline, LinkedIn headline and offer for the niche you kept.
• Move all 100 messages a day to your niche and keep your current clients.
• After 8,400 messagesCount new clients from your niche and decide whether to keep it.

Copy listDownloadReset