On this page
What does an AI automation agency actually do?
The name covers a lot. Some firms sell you one chatbot. Others rebuild how the whole business runs. We're at that second end. We look at how enquiries come in, how they turn into clients and how the work gets delivered, and then we build the software and automation that joins those steps up.
In real terms that's some mix of the following:
- Front end: websites, funnels and landing pages that actually catch the enquiry.
- CRM: a new setup or a move from the CRM you've got now, so there's one record of every contact and deal.
- Automation: workflow automation, follow-up messages, reminders and alerts for your team.
- Software: client portals, internal tools, web apps and SaaS builds.
- Reporting: dashboards with the numbers you actually make decisions on, not charts that just look nice.
- AI: chat AI and chatbots, voice agents, and AI assistants that only use information you've signed off.
My advice is don't do all of it at once. Start with the one process that causes the most chasing, connect only what that process needs, and then add the rest once the first bit is being used every day. The full list of what we build is on our solutions page.
How much does an AI automation agency cost?
The scope decides the price. One automated process costs a lot less than a platform with lots of user roles, payments, approvals and moving data across. Our projects start at £4,500. Bigger builds are £10,000 to £20,000, and platforms with lots of roles or complex builds are £20,000 to £30,000 or more.
Client portals are a good way to see how the scope moves the price:
| Build | Typical price | What you get |
|---|---|---|
| Single client type | £4,500 to £10,000 | Login, onboarding, files, messages, progress |
| Several roles | £10,000 to £20,000 | Approvals, payments or integrations |
| Multi-role or white-label platform | £20,000 to £30,000+ | Complex workflows, AI, data migration |
Portals take three weeks to three months from a signed scope. You pay 50% upfront and 50% on launch. Running costs (hosting, software, phone minutes, AI usage) go straight to the providers from your own accounts, so we're not adding a margin on top of your tools. Ongoing support and improvements are optional, on a monthly retainer priced separately.
Before we quote, we write a scope. That covers the systems involved, the integrations, who can see what, and how we'll judge whether it worked. If you want to turn up with half of that already written, our guide on how to write a software brief tells you what to put in.
What usually goes wrong with AI automation projects?
Automating a broken process faster
Say enquiries sit for two days before anyone opens them. An AI agent that replies straight away is only fixing the surface. Harvard Business Review's "The Short Life of Online Sales Leads" (2011, US data) found firms that tried to contact a web lead within an hour were nearly seven times as likely to qualify it as firms that waited even an hour longer, and more than 60 times as likely as firms that waited 24 hours or more. Of the 2,241 US companies they checked, 37% replied within an hour and 23% never replied at all. So sort out who gets each lead and who owns it first. There's more on that in speed to lead and response time.
Thinking the AI's mistakes aren't yours
In Moffatt v Air Canada (2024), the British Columbia Civil Resolution Tribunal held the airline liable for wrong fare information its website chatbot gave out, and threw out the argument that the chatbot was responsible for its own actions. Anything an AI system says in your name is you saying it. That's why our AI systems only use approved information, say when they don't know, pass people to a person and tell people they're automated.
Ignoring the marketing rules
Automated marketing calls in the UK need the person's specific consent first, under PECR. General marketing consent isn't enough, per ICO guidance. A voice agent that answers calls coming in is a very different thing from one that rings people.
Per-task pricing that scales badly
No-code tools are a great place to start and we use them where they fit. But watch the meter. Zapier counts each successful action step as a task. Triggers, filters and paths don't count, and when you hit the plan limit new runs are held unless you pay for extra tasks. A five-step workflow running on every enquiry adds up. We compare the two routes in Zapier vs custom automation and custom software vs SaaS.
Taking the person out of decisions that need one
Keep a person on money, legal commitments, complaints, regulated advice and any message sent in your name. We plan where those sign-offs go at the scope stage, not after something's gone wrong. See which decisions to keep human.
When is hiring an AI automation agency not the right choice?
There are three times we'll tell you not to build:
- A tool you can buy already fits without workarounds. If the off-the-shelf product does 90% of the job and the other 10% is just preference and not cost, buy the tool. A custom build pays for itself when the workarounds are what's costing you. The cost of disconnected tools is a good way to test that.
- The process is still changing every week. Software locks decisions in. If you're still arguing about who owns a stage or what happens after a no-show, run it by hand for another month and write down what actually happens. Then scope it.
- You need it next week. Portals take three weeks to three months from a signed scope. If your deadline is sooner, the honest answer is a stopgap in a tool you already pay for.
Most of the time the cheaper option isn't a different agency. It's a smaller first build, so one process, one automation or one dashboard. And then you decide whether to go further based on what it actually did, not on hope. If you're not sure AI belongs in the picture at all, read is your business ready for AI agents.
How we run a project
- Audit. We map how enquiries come in and where work gets stuck. No build decisions yet.
- Design. A written scope with the systems, integrations, who sees what, how we'll measure it, and the price.
- Build. Inside your own accounts. We keep what already works and don't replace things just for the sake of it.
- Test. Real cases, odd cases and what happens when things fail, before anything reaches a customer. How we do it is in how to test an AI agent before customers see it.
- Launch. In stages. We train your team and hand over the documents.
- Improve. We look at the numbers and agree the next changes. Optional retainer, priced separately.
The rules behind that order are on our approach page: built in your accounts, clear permissions, a person checking where it matters and a proper handover. At the end of it you own the code and the data.
What this looks like in practice
MRM Group, a UK property development and investment company, needed their public funnel and the way they run things inside joined up. We connected the funnel to GoHighLevel and built a logged-in internal command centre for clients, deals and tasks. Then we built client and deal workspaces where people agree before they pay, with document handling, task workflows, reminders, a full history and reports on how things are running. A person signs off at the sensitive financial-promotion and AI decision points. Related: systems for property investment companies.
TMMB Academy, a TikTok Shop academy, came to us with a funnel in pieces, software subscriptions going up and no single system. We built one portal with its own workspace for each role, so admins, coaches, creators, sales, assistants and brands. It covers onboarding, creator performance, subscriptions and payments, course modules, calls, messages, community resources, brand deals, deliverables and alerts for the team. The build ran for more than 60 days. TMMB was able to take on more, grow more easily and keep people longer, and they report making six figures within roughly two to three months (client-reported, shared with permission). Related: systems for online academies.
We build for ourselves too. The Agentryx platform does branded client onboarding from one link, delivery boards shared between an agency and each client, ad reporting across Meta, Google Ads and GoHighLevel, a white-labelled client portal on each agency's brand, and AI that turns a call summary into tasks and a draft follow-up email for a person to check. We also moved our own CRM out of GoHighLevel into a system we built. That was more than 1,600 contacts, around 100 deals with values, over 11,000 emails, over 1,000 text messages, nearly 200 calls and several thousand activity records. The import can be run again because it's keyed on the original IDs, do-not-contact settings came across, and both systems ran side by side before we switched the old one off. That move is done now.
Two word-for-word Trustpilot reviews, both five stars: "When it comes to building systems to acquire clients, Quam is your guy." (Denzil Jones). "Quam introduced me to the idea of having a solid infrastructure and helped me build it." (Stefano Ezeokaka).
Frequently asked questions
How much does an AI automation agency charge?
Our projects start at £4,500. Bigger builds are £10,000 to £20,000, and platforms with lots of roles or complex builds are £20,000 to £30,000 or more. The scope decides the price, and you get a written scope and price before anything is built. You pay 50% upfront and 50% on launch.
Who owns the systems once they're built?
You do. We build everything inside your own accounts, and the code and the data are yours. Running costs like hosting, software, phone minutes and AI usage go straight to the providers from your accounts.
How long does a build take?
Client portals take three weeks to three months from a signed scope. It depends on the roles, the integrations and whether data needs moving across. The TMMB Academy build ran for more than 60 days. We don't promise a timeline outside the written scope.
Do I need ongoing support afterwards?
No. Ongoing support and improvements are optional, on a monthly retainer priced separately. Launch comes with team training and handover documents, so your team can run the system without us.
Can AI handle customer conversations safely?
Up to a point. Our AI systems only use approved information, say when they don't know, pass people to a person and tell people they're automated. Anything involving money, legal commitments, complaints, regulated advice or messages sent in your name keeps a person involved. In Moffatt v Air Canada (2024) the airline was held liable for wrong information from its own chatbot, so the business is still on the hook.
Should I just use Zapier instead?
A lot of the time, yes, to start with. What you want to watch is volume. Zapier counts each successful action step as a task, triggers and filters and paths don't count, and when you hit the plan limit new runs are held unless you pay for extra tasks. Once a workflow with a few steps fires on every enquiry, the sums change. Our Zapier vs custom automation comparison goes through it.
When should I not hire an agency to build this?
When a tool you can buy already fits without workarounds, when the process is still changing every week, or when you need something live next week. In the first two, custom software locks in decisions you haven't made yet.
Can you move us off our current CRM?
Yes. We did it for ourselves out of GoHighLevel. That was more than 1,600 contacts, around 100 deals with values, over 11,000 emails, over 1,000 text messages, nearly 200 calls and several thousand activity records. The import can be run again because it's keyed on the original IDs, do-not-contact settings came across, and both systems ran side by side before we switched the old one off.