A guide to getting your first AI client.

Think about it from the owner's side for a second. If you say "I do AI automation," they still have to figure out why they should care. Something like "I help your team reply to new inquiries while you're busy" gives them a much clearer idea of what you're offering.

At the end of the day, they want customers and a business that can take care of them. So look at what happens from the moment someone finds the business. Maybe there aren't enough inquiries coming in. Maybe people do inquire, but nobody replies, or they book and forget to show up. Pick one of those problems to work on.

DM appointment setting

For a DM setter, I'd look at coaches or service businesses that already get people asking about working with them on Instagram. Ask the owner if you can look through some of those conversations together. What do people keep asking? How many seem interested but never get as far as a booking link?

You can start with something fairly small. It answers questions using information the owner has approved, asks what's needed to see if the person is a fit, and sends the right booking link. If it's unsure, it hands the conversation to a person. Make that easy, and let the owner pause it and take over. Nobody should have to argue with a bot to get help.

If their inbox is empty, I wouldn't start here. Same if the offer changes every week or the owner can't explain who it's for. A big follower count doesn't tell you how many people are asking to buy.

Here's a made-up example to work through, so you can see why the offer matters. Say a coach sells a $3,000 program. Four extra collected sales would be $12,000 in revenue, before delivery costs, refunds, and your fee. Ask for their actual numbers before using that reasoning in a pitch.

You could explore a $2,000–$5,000/month offer when the scope, inbound volume, and demonstrated value support it. I'd treat that as a price range to justify with the business, rather than a minimum that every DM setter can charge. A booking link sent is still a long way from a paid coaching client.

The download includes a conversation blueprint, not a connected Instagram automation. Before promising follow-ups, check the account's integration eligibility and current messaging rules. Use an official integration, respect the permitted reply window, and test the handoff. Don't build the offer around sending unlimited automated DMs.

AI receptionist

With a receptionist, you're looking for a business that gets customer calls and can't always pick up. Ask what happens after hours and who calls people back. Just seeing a phone number on their site doesn't mean you've found a problem.

The starter in this kit collects a non-emergency request and submits it to an operator mailbox for a callback. It gives you a complete small build to demonstrate. Calendar booking, live transfers, and more complex call handling need additional work and testing.

The part I'd check early is who's going to read the requests and call people back. You could build something that captures every detail correctly, and the customer still gets ignored because nobody checks the inbox. Sort that out with the owner first.

Ads plus speed-to-lead

Another option is running ads for a business and building the system that replies when someone fills out the form. So you're working on the offer and ads, and then what happens to the lead afterward: the first reply, a few questions, and getting them to the right person.

I think this is worth looking at, but there's more to learn here. You have to understand the ads too, including targeting, creative, tracking, and whether the leads are any good. The owner needs to approve the spend and be able to take on more customers. A fast reply won't make up for an offer people don't want.

If the business already runs ads, you can investigate the response process first. Track a test submission through the form with the owner's permission. Check whether it reaches the right person and whether the reply gives a useful next step.

And then, if they're getting calls they can't answer, you can talk about adding a receptionist. That's a possible upsell. I'd let what you find decide the order though. Some businesses need help answering calls first, and some already have leads sitting in a form inbox.

A website that makes booking easier

Look for a service business whose site makes it hard to understand the offer, inquire, or book on a phone. Walk through the process yourself. Point to the confusing step rather than telling the owner their website looks old.

A focused first project could improve one service page and its inquiry form. Agree where submissions go, what the visitor sees afterward, and who answers. Check the page on a real phone and send a test inquiry all the way through.

A prettier page is not proof of more customers. Measure the relevant action and keep claims about the result separate from claims about the design.

Reminders and rescheduling

Look at businesses with booked appointments and a documented attendance problem. Ask how they confirm appointments, how customers cancel, and whether available slots get offered to someone else.

The first version can send the agreed details and give someone a clear way to reschedule. Agree on the timing, channel, permission to contact the customer, and what happens after a cancellation. Avoid a setup that keeps messaging someone who has already opted out or moved their appointment.

Choose this when the business already has bookings. It won't solve a shortage of inquiries.

Google Maps review collection

This is one I'd look at as an upsell to ads plus speed-to-lead, a receptionist, or reminders and rescheduling. Once the business has served someone, you can help it consistently ask for an honest Google review.

The logic is straightforward: more genuine reviews and positive ratings can help local visibility. If more people find the business and trust what they see, there's more opportunity for inquiries and customers. That's the route toward more money, but each step still depends on the business. A new review doesn't guarantee a ranking change or a sale. Google explains how local ranking works here.

Build around a completed job or attended appointment, then send the customer the owner's Google review link through an agreed channel. An unanswered call or a new booking isn't the same as a completed service. Keep track of requests sent and opt-outs, and avoid sending duplicates.

Ask customers fairly for honest feedback, whatever you think they'll say. Don't offer rewards, write reviews for them, or send only happy customers to Google while hiding the link from everyone else. The google-review-collector.md file includes a request message, the trigger rules, and checks to run before turning it on.

Pick a type of business and write down the problem you want to help with, plus how you'd reach the person who deals with it. You can change your niche later. For now, I'd keep it focused enough that you can look at a few businesses and start noticing the same things.

Use three columns: what I observed, what I am assuming, and what I need to ask. "The website lists a phone number and closes at 5" is an observation. "They lose calls every night" is an assumption. Keep them separate when you contact the owner.

For example, imagine a plumbing business called Example Plumbing. Its site lists weekday hours and asks customers to call for service. You haven't seen its call records. Your question is: "What happens when someone calls after you close?" If the owner already has reliable coverage, move on or investigate a different problem.

If the owner says requests sit in voicemail until morning, ask who checks it, how often, and which details are usually missing. Now you can discuss a small intake system without pretending you know how much money they're losing.

Start by looking into ten businesses. I'm giving you ten so you have somewhere to start, and you shouldn't read it as the number that gets you a client or proves an idea works. Keep the ones that look suitable and write down why the others don't. By the end, you should be able to describe your buyer more clearly than "any local business."

Your offer sentence can be plain: "I help [type of business] handle [specific situation] by [specific process]." For the example: "I help plumbing businesses collect after-hours job details and callback numbers so the team has a usable request in the morning."

Here's what we're building. Someone calls after hours about a non-emergency job, the assistant asks for their details and reads them back, and then it submits a callback request. That request goes to the operator by email, so you'll need to check that their mail system actually delivers it.

The flow is: caller, ElevenLabs agent, authenticated n8n webhook, validation, operator email, result returned to the agent. The result confirms whether the mail server accepted the message. It doesn't confirm a callback time or a booked appointment.

I'd start in the ElevenLabs browser preview so you can test this before dealing with a phone number. Once the preview and email tests pass, you can connect the phone side. You'll need your own ElevenLabs and n8n access, an SMTP account to send email, and a mailbox to receive it. Check what those services cost before signing up. I'm giving you the files for free, but running them can cost money.

Set up the business information

Read sample-business.md and replace the fictional business facts with approved information before using a real business. Keep the scope small: service area, hours, job types, who handles callbacks, and what the assistant should do when it can't help.

Paste receptionist-system-prompt.txt into the agent's system prompt and add the business information to its knowledge. Use the included greeting. Read the prompt yourself. You should know which questions the agent asks and which promises it must not make.

Connect the handoff

Import receptionist-intake.n8n.json into n8n. It starts inactive. Select your Header Auth credential on the webhook and your SMTP credential on the email node. Set the sender and operator recipient in the email node. Those addresses come from your configuration, never from what a caller says.

Follow START-HERE.md to test the workflow with the sample request. Hosted ElevenLabs needs a reachable HTTPS webhook. A localhost address won't work. n8n has a temporary test URL and a production URL; use the production URL after configuring and publishing or activating your workflow.

Create the submit_callback webhook tool in ElevenLabs using elevenlabs-tool-setup.md. Use POST, copy the field definitions, and add the matching X-Build-Key header as a secret. Keep the secret out of the prompt. The setup file spells out which fields the tool requires and what each result means.

A detailed prompt still needs working tools behind it. The prompt in this kit includes call types, natural conversation, corrections, and failure handling, but the connected action is a callback request. If you want it to book appointments or send confirmations, you need those integrations too. START-HERE.md includes written practice conversations and a scorecard for testing the actual voice agent.

Try it when something goes wrong too

Send a valid synthetic request and check the operator inbox. Then try a missing phone number and a refused callback. An invalid request should not reach the email step. Disconnect SMTP in a test copy and make sure the agent doesn't say the request was submitted.

Use receptionist-test-cases.csv for the voice tests. Ask for a price the business never supplied. Ask to book a specific time. Change your number halfway through. Ask it to ignore its instructions. The agent should keep the agreed limits and correct the details before submission.

If a tool times out, you don't know whether the mail server accepted the request. The assistant should explain that it could not confirm submission. Automatic retries can create duplicate emails, so the starter doesn't retry an ambiguous handoff.

Before a real pilot, test actual voice behavior and inbox delivery on the configured accounts. Add the monitoring and duplicate handling your use case needs. A passing demo is the point where you can show your work and discuss its limits.

Connect a real number with Twilio

The browser preview lets you practice without a phone number. For someone to call this receptionist from their phone, you need an inbound-capable number connected to your agent.

In Twilio, get a voice-capable number and complete the account and country requirements. In ElevenLabs, open Phone Numbers, import that Twilio number with the credentials requested by the native integration, then assign your receptionist agent to it. Use the dashboard's secret fields, and keep credentials out of the prompt and your downloads. A verified caller ID by itself is outbound-only and won't receive these calls.

Call the number from another phone and check the full route: call answered → details confirmed → request submitted → email received. Also test a failure. Keep your demo number separate from a client's existing line while you're learning. Before changing a live business line, agree on forwarding hours, voicemail behavior, and how to restore the previous setup.

The twilio-phone-setup.md checklist walks through the connection and acceptance checks. Buying a number is only one part of the setup; the assistant, webhook, and operator inbox all still need to work together.

Start with the businesses you researched. Choose a contact route they use for business inquiries. Write to the person responsible for the process when you can identify them. A working email address or an active Instagram profile still doesn't tell you whether your offer matters to that person.

Mention something you actually found, explain why you're reaching out, and ask a question they can easily answer. I wouldn't open with "I noticed you're missing calls" when you haven't seen their call records. Just ask how they handle calls after closing.

Adapt the message to what you found. If you already have a relevant demo, describe it accurately. If you haven't built it, don't write that you have. The downloadable scripts include a DM version, a follow-up, and replies to common questions.

A quick thing about templates

I know this sounds a bit ironic because I just gave you a template, but be yourself when you reach out. People buy from people. The person reading your message should feel like there's a real human on the other side who has a reason to talk to them.

You can write the message yourself. A template can help when you're staring at a blank screen, but copying a set of words doesn't mean your outreach will suddenly work ten times better. Be natural, be straightforward, and say why you're reaching out. If a line sounds weird when you say it out loud, change it. I'd use the example to get started, then put it in your own words.

Record a short demo

Open with the situation you're handling. Show a caller making a request, then show the result in a test inbox. Explain what an operator would do next. Finish with the limits and a question about their current process.

Use fictional information or records you have permission to show. Keep passwords, real customer details, unrelated tabs, and private inbox messages out of the recording. Someone should understand the result without watching you configure nodes for twenty minutes.

Keep going, and keep track of what happens

Set aside time you can keep coming back to for finding businesses, reaching out, following up, and fixing the demo. Count how many different businesses you've contacted as well as how many messages you've sent. If you follow up three times with the same person, you're still talking to one prospect.

Track delivery failures, replies, relevant conversations, demos, pilot proposals, and paid pilots separately. If messages bounce, work on contact quality. If people reply but don't have the problem, work on the buyer criteria. If the problem is real but they don't understand your demo, improve the demonstration.

Try changing one thing at a time and write down when you changed it. If you change the niche, offer, message, and channel all at once, how are you going to know what helped? Small numbers are hard to read. A quiet week is a reason to look closer at what you're doing, but it doesn't tell you an entire service is dead.

Follow up when you have a relevant reason or a useful clarification. Respect a no and stop contacting people who ask you to stop. The tracker includes a do-not-contact column so that decision survives your next work session.

When someone replies, slow down enough to understand what they do now. Ask them to walk you through a recent inquiry. Who saw it, what did they need to know, and where did the details end up? You might find that something in your demo would be really inconvenient for their team. It's better to find that out before quoting them.

For a receptionist, ask which calls should reach a person, which requests can wait, and what the team is allowed to promise. Get the owner to define a useful handoff. A tidy email nobody reads is a bad result.

Write a pilot scope that both of you can check: which requests it handles, the accounts involved, where the handoff goes, who monitors it, and how you'll decide whether it's working. Name the things the pilot doesn't do when they affect the owner's decision.

For pricing, work out what you're taking on. How long will setup take, what does it cost to run, and what are you agreeing to maintain? Keep your fee separate from phone, voice, hosting, email, and ad spend. Work through a normal month and a busy one in the cost worksheet, so you can see whether you have enough left to support it.

Agree the price, payment schedule, delivery inputs, and acceptance criteria before starting paid delivery. The scope worksheet helps you describe the work; use an appropriate agreement for the actual engagement. Avoid attaching a revenue guarantee to a system whose sales process you don't control.

Ask for the payment while you're on the call

Would pay-per-result make more sense?

Sometimes you can earn more by charging for an agreed result. Say the alternative is $600/month, and you agree $60 per qualified appointment that actually takes place. Fifteen billable appointments would be $900, ten would be $600, and five would be $300, before your costs. These are example terms, not expected results.

You take on more uncertainty with this model. If the client doesn't answer leads, closes badly, or won't share records, you could do the work and struggle to get paid. A setup fee plus a smaller base fee and a result bonus is another option to discuss when there are fixed costs to cover. Work through the examples in pilot-and-handover.md before choosing.

Use accounts the client owns wherever practical. Have them grant the access needed for the work. Record how they can remove that access and what they keep if the engagement ends. Don't make their phone number or business records dependent on your personal login.

Go through the test checklist with the client. Show them a normal request, what happens when something fails, and where their team receives it. Then let them try it. If they can't find a request without you pointing at the screen, you've still got something to fix before launch.

Agree who checks the system, how problems get reported, and how it can be paused. Record a short handover showing the inbox or destination, the editable business information, and the failure path. Recheck after changing credentials, business hours, or workflow settings.

Measure what you can observe. Requests captured, successful handoffs, callback attempts, and booked jobs are different events. Ask the client for the downstream result rather than counting every submitted request as a new customer.

Then discuss the next problem you have evidence for. If new ad inquiries wait unanswered, improve the response. If calls go unanswered, investigate call coverage. If people book and fail to attend, look at reminders. Explain the additional work and agree a new scope.

So that's where I'd start: choose one of the opportunities, fill out the niche worksheet, and look into your first ten businesses. When you get to the build, the files are below. Keep a note of what you've tested and what still needs work as you go.