You need a repeatable engagement, which is a narrower thing than a niche. 'AI for dentists' is a label. 'Patient recall automation for dental groups on Dentrix' is an engagement you can price, deliver twice, and sell a third time with a case study. Pick the process first; the industry follows from where that process hurts most.

Should I start with n8n, Make or Zapier?

Start with whichever your first client can keep running after you leave. Zapier is the easiest to hand over and the most expensive at volume. n8n is the most capable and can be self-hosted, which some clients will require and others cannot manage. Make sits between. The choice matters less than the handover.

How much can I charge for a first project?

Charge for the outcome the process returns, and check it against the hours. If an automation gives an operations manager ten hours a week back, a five-figure build is easy to justify and a four-figure one leaves money on the table. Published entry prices on this site's listings start around $2,000 for a scoped discovery.

Do I need a website before I have clients?

No. You need three clients, three write-ups and one page that shows them. The agencies on this site that rank well publish named clients with figures. The ones that publish a services list and a stock photo do not.

How to build an AI agency | AI Agency Radar

Build it the way any service business gets built. Find one process a specific kind of company pays to fix, fix it for three of them at a fixed price, write down what you did and what it returned, then sell that write-up to the fourth. Tooling is the last decision. Branding is the one after.

Pick a process, and let the industry follow

Every guide to this business tells you to pick a niche. That advice is right and usually applied wrong. A niche is an industry label; what you can sell is a process. “Quote-to-cash for a manufacturer with two salespeople and a spreadsheet” is a process. So is “inbound lead qualification for a firm that gets forty enquiries a week and answers twelve.” Each one has a before and an after that can be measured in hours, and each one exists in a dozen industries.

Pick the process by looking at where you already have proof. If you have run operations somewhere, you know which handoff cost the most. If you have built software for clients, you know which request kept coming back. Start there. The industry you specialize in is the one where that process hurts most and where the buyer can be reached.

Get three clients before you get a logo

The first three engagements are research that pays. Sell them at a fixed price, low enough to be an easy yes and high enough that you are taken seriously, and treat each one as the case study it will become. Agree up front on the figure you will measure: hours returned per month, response time, error rate, whatever the client cares about. Then measure it after the build, and get the number in writing from the client.

That number is your marketing. A named client, a described process and a figure that can be checked is worth more than any amount of copy about being AI-first. It is also the main thing this directory looks for when it ranks an agency.

Price the fix, then price the retainer

A build has a price. What happens after it has a different one. Every automation needs someone to answer when a vendor changes an API, a credential expires, or the client adds a product line. Sell that as a retainer from the first proposal, and describe what it covers in plain words: monitoring, fixes, a set number of hours for changes. Agencies that leave this out either do the work for free or lose the client to whoever picks up the phone.

For the build itself, the working rule is to price against the outcome and sanity-check against your hours. An automation that returns forty hours a month to a team is worth a five-figure build to that team. Charging by the hour for work that removes hours is how an agency talks itself into being cheap.

Deliver something the client can keep

The difference between an agency and a freelancer with an n8n account is what gets handed over. A good handover includes the workflow, the documentation to change the things the client will need to change, the credentials moved into the client’s own accounts, and a written answer to “what do we do when it breaks.” A client who can run the thing without you will keep paying you, because now you are the person who made that possible. A client who cannot will resent you, and then leave.

This is also where the tool choice actually gets made. Choose the platform the client can operate. The comparison of n8n agencies and Zapier and Make agencies on this site shows what each ecosystem’s specialists sell, which is a fair guide to what each platform is good for.

Skip these

A course. A community. A YouTube channel about building an agency. A services page listing eleven things you could do. A brand exercise. Each of them is a way of doing marketing instead of getting a client, and each one is more pleasant than a sales call, which is why so many new agencies do them first and fold before the second client.

Then get listed

Once you have three named engagements with figures, list the agency here. It costs nothing, it puts you in front of buyers who are choosing, and the review will tell you plainly what a buyer sees when they look you up. Before that, read why AI agencies fail. It is shorter than this page and more useful.