How to get your first ten AI agency clients

Ten clients takes roughly a thousand first-touch emails, about two hundred dials, and one demo you can put in a stranger's hands inside four minutes. That is the whole answer. The part nobody tells you is that the hard number is not the volume, it is the trade. Agencies that pick one trade and work it for ninety days get to ten. Agencies that sell "AI solutions" to whoever answers stall out around three.

I want to walk through the actual math, the actual objection you will hear, and the part of the first ten that has nothing to do with selling.

How many conversations does ten clients actually take?

Start with the only outbound numbers that come from a real dataset rather than a blog post. Instantly published a 2026 cold email benchmark report built from platform-wide sending between January and December 2025. The overall average reply rate across their users is 3.43%. The top 10% of senders clear 10.7%. The top quarter clear 5.5%.

So a clean list of a thousand owners in one trade, sent well, gets you somewhere near thirty-four replies. Some of those replies are out-of-office. Some are "take me off this list." Apollo argues in its own benchmark piece that total reply rate is a polluted metric and you should track positive replies instead, calling 2% to 3% the minimum viable rate and 4% to 6% the target. Those two framings do not reconcile cleanly, and Apollo's numbers read as a recommended target rather than a measurement. Take the pessimistic read. Assume a thousand emails produces ten to fifteen conversations worth having.

Two other findings from the Instantly data change how you send. First, 58% of all replies come from the first email. Second, the sequences that perform sit at four to seven touches. If you are sending a two-email sequence, you are leaving roughly 40% of your replies on the table for the sake of two more drafts.

Apollo adds one more thing that runs against everything the cold email tooling market sells you: campaigns sent to fifty or fewer recipients consistently beat campaigns sent to a thousand or more. That is the shape of the first ten. Not a blast. Fifty roofers in one metro, each email referencing something on their actual site.

Ten to fifteen conversations from a thousand emails, closing at a rate you should not assume is better than one in four in your first quarter, means one outbound engine alone gets you three or four clients. The other six come from the phone, from referrals off the first three, and from the local trade association nobody in your city bothered to show up to. Plan for all four channels. Do not plan for LinkedIn posts.

Why local service businesses are a buyable market right now

The gap you are selling into is measurable, and it has stayed open longer than most people expected.

The Census Bureau's Business Trends and Outlook Survey tracks AI use across roughly 200,000 businesses every two weeks. In its May 2026 writeup of six months of that data, covering collection from December 14, 2025 through May 3, 2026, overall AI use hovered between 17% and 20%, landing at 19.8% nationally. Firms with at least 250 employees ran at 37%. Firms with 100 to 249 ran at 32%. And the sentence that matters most for anyone selling to local businesses: AI use rose among firms with at least 20 employees over that period but "didn't change significantly among firms with fewer than 20." Under 20% of firms with four or fewer employees reported using AI at all.

Big companies are pulling away. The HVAC company with six trucks is standing still. That is your entire market thesis, and it comes from the Census Bureau rather than from a vendor deck.

Now the part that reframes the pitch. Simply Business surveyed 1,047 US small business owners between Q4 2025 and Q1 2026 and found that 62% use AI to run their business. Look at what they use it for: research and brainstorming at 51%, content creation at 44%, visuals at 37%. They use it privately, to draft a Facebook post or rough out an email. Almost none of it touches a customer.

So your prospect is not an AI skeptic. They have a ChatGPT tab open right now. What nobody has done is put AI anywhere near their phone, their inbox, or their booking calendar, which is where the money actually leaks. The distance between "I use it to write captions" and "it answers my phone at 7pm" is the distance you get paid to close.

Even the software they already pay for has not closed it. When Jobber announced its AI features in a September 2025 release, the headline product, Jobber Voice, let the pro speak commands to the app while on a job. Useful. It is not an agent that talks to a homeowner. That release contained no AI receptionist and no customer-facing call answering. The field service software category is building AI for the operator, not for the operator's customer.

Pick one trade and stay in it for ninety days

Every agency owner I talk to who is stuck at three clients is selling to five trades. Every one who got past ten picked one.

The reason is not focus as a virtue. It is that the second call in a trade is dramatically easier than the first, and the tenth is a different job entirely. By call ten you know that roofers get their leads from storm season and a canvasser, that their office manager is the actual buyer, and that the thing keeping them up is not missed calls but the estimator who quits every eighteen months. You stop discovering and start recognizing. Your demo stops being generic because you know which four objections to load into it.

Pick on three criteria. The ticket has to be big enough that one recovered job pays for a month, which rules out most retail and a lot of food. The business has to run on inbound phone calls rather than walk-ins. And you need to be able to name forty of them within a fifty-mile radius by Friday.

We wrote up what this looks like inside roofing specifically if you want the shape of a single-trade playbook. The vertical packs cover more than a hundred trades, and the point of them is that the demo speaks the buyer's language without you learning a new industry from scratch every week. But the pack does not replace the ninety days. It just means the demo is not the reason you lose.

What to say when they tell you AI is not for their business

Here is the objection you will actually hear, and it is not price.

The SBA Office of Advocacy pulled the BTOS data apart in a September 2025 research spotlight and found that nearly 82% of businesses under five employees gave relevance as the reason they were not planning to use AI. Not cost. Not privacy, which came in around 6.3%. Not lack of knowledge, at 6.7%. Relevance. Four out of five of the smallest businesses in America looked at AI and concluded it was for somebody else.

That number should change your opening line. If you lead with capability, you are answering a question they did not ask. "Our AI can handle scheduling, follow-up, and lead qualification" lands on someone who has already decided none of that applies to a four-person plumbing company.

What works is arithmetic about their business, not claims about your product. How many calls came in last Saturday. What happens to the ones that ring out. What a booked drain job is worth. Then you show the thing answering a phone in their trade's vocabulary, and the relevance objection dies without you arguing against it.

The second objection follows right behind, and the Simply Business survey has the number for it too: 86% of respondents said the ability to speak to a human is important when they deal with any AI-led service, with 66% calling it very important. Your prospect knows that. They have been on the other end of a bad phone tree.

Do not fight it. Sell the agent as the thing that catches what currently goes to voicemail and hands a real caller to a real person faster. Any pitch that implies the agent replaces the human fails on a stat that two thirds of the market feels strongly about. It also sets up a client who churns in month three when their customers complain.

Why the demo has to happen on the first call

The single biggest change in close rate is not a better script. It is moving proof from the second meeting to the first one.

A prospect who watches you talk about AI has to trust you. A prospect who calls your demo number themselves, hears it answer in their trade's language, asks it what a service call costs, and gets booked, has verified it. Nothing you say afterward carries the same weight as thirty seconds of them poking at it trying to break it.

That is the reason our whole product is built around a live demo you can hand over rather than a slide. Pick the trade, type in their business name, send one link. The demo states plainly that it is a sandbox and that bookings in it are simulated, which sounds like a weakness and is the opposite. Prospects relax when you tell them what is fake. They start testing instead of auditing you.

Practically, run the first call like this. Two minutes on their call volume and ticket size. Then stop talking and give them the link while you are still on the phone. Let them dial it. Say nothing for ninety seconds. The silence is the pitch.

What ten clients are worth at published wholesale prices

Run the numbers with figures you can check rather than a promise.

Lumina Sales is $49 a month, listed on the pricing section with the usage limits sitting right next to it: 100 live voice minutes, 500 chat messages, 50 website audits, one published demo at a time. Delivery is priced separately as wholesale, and those are prices you pay us, not prices your client sees. The top tier, which includes voice, is $500 a month. Tier two is $275. Tier one is $225. Single services run $40 to $350.

So ten clients on the top tier costs you $5,000 a month in wholesale plus the $49. Retail is yours to set. At $1,500 a month you are running $15,000 in revenue against $5,049 in cost. At $2,000 you are at $20,000. There is no build fee in that structure, which is the part that changes the sales cycle more than the margin does, because you are no longer asking for four thousand dollars up front from someone who met you last Tuesday.

If you want the full reasoning on where to land retail, we wrote a whole post on pricing an AI buildout. And if you are weighing this against assembling the stack yourself, the vendor rate cards are public. Retell lists voice infrastructure at $0.055 a minute with text to speech from $0.015 and telephony at $0.015, and its own calculator defaults to $0.11 a minute all in. The per-minute cost was never the expensive part. We broke down the rest of that comparison in build versus buy.

Who this does not work for

I would rather lose the signup than have you find this out in month two.

This does not work if you will not do outbound. A thousand emails and two hundred dials is a real amount of work spread over a quarter, and there is no version of the first ten clients where the demo does the prospecting for you. If your plan is content and inbound, your first ten will take a year, and that is a different business with different funding needs.

It does not work if you need revenue inside thirty days. The sequence that actually produces a signed client runs four to seven touches spaced a few days apart, then a call, then a demo, then a decision that involves a spouse or a business partner. Six to ten weeks from first email to first payment is normal. Budget for it.

It does not work if you are not willing to be technical about the boring parts. Onboarding is where deals die, not the pitch. Hours, service list, price bands, the calendar, who gets the lead. Someone has to collect all of that and it is usually you. We built onboarding assistance exactly because this is the step that stalls closed deals, and we wrote up what actually happens between signature and delivery, but nobody can make a client send you their logo.

And it does not work if you want to sell full replacement of a receptionist. The market told you no on that one. Two thirds of people call human access very important. Sell the overflow, the after-hours, the Saturday. That offer is easier to believe and it survives contact with the client's customers.

The first thirty days, in order

Week one. Pick the trade. Build a list of two hundred businesses in one metro with a phone number and a website you can look at. Build the demo once, in that trade, and call it yourself twenty times until you know how it breaks.

Week two. Send fifty emails, not a thousand. Four to seven touches, three or four days apart, first email doing the heavy lifting since that is where 58% of replies come from. Start dialing the same list in parallel. Aim for ten conversations, not ten closes.

Week three. Run demos. Send the link on the first call every time. Track which objection kills each one and rewrite the opening line to kill it earlier.

Week four. Close one. Then ask that client which two other owners in their trade you should call, and actually call them the same week, because a referral inside a trade you now understand converts at a rate no cold list ever will.

Repeat that for a quarter and ten is realistic. Skip the trade discipline and you will spend the same quarter getting to three.

Common questions

How long does it take to get the first client, honestly? Six to ten weeks from the first email is the normal range once you are actually sending. The sequence itself takes three weeks, then there is a call, a demo, and an internal decision at the client. The people who get there in two weeks usually had a warm relationship already.

Do I need to know how to build AI systems? No, but you need to know what you sold well enough to answer questions about it. The build is handled through the fulfillment platform, so you are not writing prompts at midnight. You do need to understand hours, routing, and how a call gets escalated to a person, because those are the questions an owner asks in month one.

Should I niche down if I already have clients in three trades? Keep them and pick one trade for new business. You are not abandoning revenue, you are choosing where the next forty calls go. The compounding only starts when the calls are in the same vertical.

What do I say when they ask why they can't just use ChatGPT? Point at the difference the Simply Business data shows. They already use AI to write things for themselves. ChatGPT does not pick up the phone at 7pm, does not know their price bands, and does not put a job on their calendar. That is the product, and it is a different thing from a chat window.

Is $49 a month really the whole cost to start? For the platform, yes, with the usage limits published next to the price. Delivery is wholesale on top, and you only pay that once you have a client to deliver to. There is no build fee in either direction.

Where to start this week

Do one thing before Monday. Pick the trade, then go talk to the live demo yourself and try to break it. Ask it something a real customer in that trade would ask, the awkward pricing question you would not want a human to fumble. If it holds up, you have the first call handled and the only thing left is the list.

Then start your workspace and build the demo in your own trade with your own logo on it. Ten minutes, most of which is choosing a subdomain.