Have you ever sat through an “AI agency” pitch and come away with absolutely no idea what they’re actually going to build for you?
You’re not alone, and it’s not because you’re bad with technology. It’s because a fair number of these pitches are deliberately vague, or at least deliberately impressive-sounding, which isn’t quite the same thing as clear.
We’ve had several conversations recently with business owners holding a proposal for a “bespoke AI system”, trying to work out what’s genuinely custom and what’s essentially somebody else’s product with a new badge on the front. It’s a fair question. It’s also one a lot of agencies would rather you didn’t ask too directly.
This isn’t an article about whether you should buy from an AI agency. Plenty of them do genuinely good work. It’s about understanding what’s actually sitting behind the phrase “AI system” before you agree to pay for one.
Before We Go Any Further
This is simply our view, based on what we’ve seen while helping businesses make sense of these proposals.
We’re not against AI agencies as a category. Some are excellent, genuinely skilled at stitching together the right tools for a business and making them work properly. Others, frankly, are a website, a sales deck and an OpenAI account.
Have a look around, ask a few pointed questions of anyone pitching you, and make your own judgement. What follows is simply how we currently see the market.
What Is an “AI System” Actually Made Of?
Nine times out of ten, it’s built from three basic ingredients, combined in different proportions.
The first is an existing AI model, usually accessed through an API from a company like OpenAI, Anthropic or Google. The agency isn’t training their own AI from scratch. Almost nobody does. They’re sending your enquiry to somebody else’s model and passing the answer back to you, usually with some instructions bolted on to make it behave a certain way.
The second is an automation platform, something like Zapier, Make or n8n, or a custom version of the same idea. This is the bit that moves information from one place to another. It’s what takes a form submission, sends it to the AI model, and then pushes the result into an email, a spreadsheet or a CRM.
The third is an integration into your existing systems, most commonly your CRM, your booking software or your website’s contact form. This is usually the part that takes the longest and is genuinely the most valuable, because it’s specific to your business rather than off the shelf.
Put those three together, wrap them in a nice interface, and you’ve got what most agencies are calling an “AI system.”
“There’s nothing wrong with using existing tools well. There is something wrong with charging like you built them from scratch and hoping nobody asks the difference.”
Is That a Bad Thing?
Not necessarily. Using established building blocks well is often the sensible approach, not a shortcut.
A model trained by a company spending billions on it is almost certainly better than anything a small agency could build in-house. The value an agency adds usually isn’t the AI model itself. It’s understanding your business well enough to configure that model properly, connect it to the right systems and make sure it behaves sensibly when your actual customers start using it.
That configuration work is genuinely skilled. Working out what your AI assistant should and shouldn’t say, testing it against real customer questions, connecting it cleanly to your booking system without breaking anything else, that’s not nothing. It takes real time and real expertise.
The issue isn’t that agencies use existing tools. The issue is when “existing tools, configured well” gets marketed as “bespoke AI system, built specifically for you”, and priced as though the second description is true.
What’s the Difference Between “Off-the-Shelf” and “Custom”?
Off-the-shelf usually means the core engine, the AI model itself and often the automation platform underneath it, is a product anyone can access. Custom usually refers to how that engine is configured, trained on your specific information and connected to your specific systems.
Here’s a rough way to think about it:
- Off-the-shelf: The AI model doing the actual “thinking”. Almost always somebody else’s product.
- Off-the-shelf, lightly configured: A chatbot widget or automation template that’s been given your logo, your tone of voice and a basic script.
- Genuinely custom: Your specific business logic, pricing rules, booking availability and edge cases built into how the system responds, tested against your real customer questions.
- Genuinely custom: A proper integration into your CRM or booking system that didn’t exist before and had to be built to work with your particular setup.
Most proposals sit somewhere in the middle of this. The honest question to ask isn’t “is this custom or off-the-shelf”, because the answer is almost always both. It’s “which parts are custom, and how much of what I’m paying for is that custom work, versus the underlying product everyone else is also using?”
Imagine a Business Quoted £8,000 for a “Bespoke AI Booking Assistant”
We don’t have a real case file to share here, so this is a hypothetical, but it reflects a fairly typical shape of what we’ve seen described to us.
Say a hair and beauty business is quoted £8,000 for a “bespoke AI booking assistant” that answers website enquiries and books appointments automatically. The pitch mentions machine learning, natural language processing and a system that “learns your business over time.”
Underneath, it turns out to be a popular chatbot widget connected to an off-the-shelf AI model, with a basic automation linking it to the business’s existing booking calendar. The “learning your business” part is a document containing the business’s opening hours and price list, fed to the AI as background information. The actual build time, once you strip out the sales process, is closer to two or three days than the weeks implied in the pitch.
None of that necessarily makes it a bad system. If it answers enquiries accurately and books appointments correctly, it might do exactly what the business needs. The problem is the business paid £8,000 believing they were getting something built uniquely for them from the ground up, when in reality most of the heavy lifting was done by tools available to anyone with a subscription.
Had the business asked upfront what was custom versus off-the-shelf, they might still have bought it. They’d just have gone in with realistic expectations, and probably negotiated a fairer price.
Does That Mean Every AI Agency Is Ripping People Off?
No, and it would be unfair to suggest that. Plenty of agencies are entirely upfront about which tools they’re using and charge sensibly for the configuration, integration and support work they genuinely provide.
Some businesses also have needs complicated enough that the integration work really is substantial. Connecting an AI system cleanly into a business with several disjointed pieces of software, multiple teams and messy existing data can take real time, and a five-figure quote for that kind of project can be entirely fair.
Our point isn’t that using existing AI tools is dishonest. It’s that dressing up existing tools as something built entirely from scratch, and pricing accordingly, is where things start to feel misleading.
“We’re not saying don’t use the tools everyone else uses. We’re saying be honest about which parts are yours and which parts you’re renting from somebody bigger.”
So, What Should You Actually Do?
Before agreeing to any AI proposal, ask the agency to walk you through the build in plain English, piece by piece.
- Ask which AI model is being used underneath the system, and whether it’s accessed via a standard API.
- Ask whether an automation platform like Zapier, Make or n8n sits behind the scenes, and if so, which one.
- Ask specifically what’s being custom-built for your business versus configured on top of an existing product.
- Ask how long the actual build takes, separate from the sales and onboarding process.
- Ask what ongoing subscription costs exist underneath the agency’s own fee, and whether you’re paying for those directly or through a markup.
- Ask what happens to the system, and your data, if you stop paying the agency.
- Get the answer in writing, not just in conversation.
None of these questions should make a genuinely competent agency uncomfortable. A good one will usually be pleased you asked, because it means you understand what you’re buying.
Questions Worth Asking Any AI Agency Before You Sign
- What AI model or provider sits behind this system?
- Is there an automation platform underneath it, and which one?
- What exactly is custom-built for us, and what’s a configured version of an existing product?
- How long does the actual build take?
- What are we paying monthly on top of the initial cost, and what’s included in that?
- Who owns the account the AI model runs through, us or you?
- What happens to our data and our system if we stop working with you?
- Can you show us a similar system you’ve built for another client?
Frequently Asked Questions
Is it bad if an AI agency uses tools like OpenAI or Zapier instead of building everything from scratch?
Not necessarily. Building on established tools is often the sensible choice, since they’re typically more reliable than anything a small agency could build independently. What matters is whether you’re told this clearly and priced fairly for it.
How can I tell if an AI system is genuinely custom?
Ask specifically which parts reflect your business’s own logic, pricing, availability and edge cases, tested against your actual customer questions, versus which parts are a standard product with your branding added. A genuinely custom system should be able to answer questions specific to how your business actually operates.
Why do some AI agencies avoid explaining exactly what they’re building?
Sometimes it’s because the explanation is technical and they’re trying to keep things simple for the customer. Sometimes it’s because a clear explanation would reveal how little custom work is actually involved. Ask direct questions and see which one you’re dealing with.
Should I avoid agencies that use off-the-shelf tools?
No. Nearly all of them do, to some degree, because rebuilding an AI model from scratch would be pointless and enormously expensive. The question isn’t whether they use existing tools. It’s whether they’re honest about it and price accordingly.
What’s a fair price for an AI system built mostly from existing tools?
There’s no fixed figure, because it depends entirely on how much genuine configuration and integration work is involved. A simple chatbot connected to one system should cost considerably less than a system integrated across several complex pieces of software. Ask for the build to be broken down so the price makes sense against the actual work.
Is it reasonable to ask an agency to name the specific AI model they’re using?
Completely reasonable. A competent agency should be able to tell you straightforwardly, whether it’s a model from OpenAI, Anthropic, Google or elsewhere, and explain why they chose it.
Where Do We Stand?
We think there are genuinely good AI agencies out there doing honest, skilled work.
We also think a fair number of proposals currently landing in business owners’ inboxes describe an off-the-shelf chatbot and a basic automation as though it were built entirely from nothing, at a price that assumes the buyer won’t ask too many questions.
Our view is simple. Ask what’s actually underneath the system. Ask what’s custom and what isn’t. Ask what you’re paying for each part. A good agency will answer clearly. Anyone who won’t is telling you something too.
Trying to Work Out What You’re Actually Being Sold?
If you’ve got an AI proposal in front of you and you’re struggling to work out what’s genuinely bespoke and what isn’t, give us a call and talk us through it.
We’re not trying to sell you our own version instead. Sometimes it just helps to have somebody who’s seen a few of these run through it with you before you sign anything.
Pick up the phone. We’re always happy to talk it through.