Straight answers
What does an AI agent actually cost?
We are not going to print a price range on a webpage — the honest answer depends on things we would have to ask you about. What we can do is show you exactly what moves the number, so you can judge any quote you receive, including ours.
Build cost
What drives the one-time cost
How many jobs it does
One job — answering and qualifying enquiries — is the cheapest useful agent. Cost rises with each additional job, not linearly but noticeably, because each brings its own edge cases.
The state of your data
This is the swing factor most quotes hide. If your answers live in a clean catalogue, preparation is quick. If they live in five contradictory documents and someone's memory, deciding what is authoritative can outweigh the build itself.
How many systems it must touch
An agent that only talks is cheap. One that reads your stock, writes to your CRM and triggers your accounting costs more — the integration work is real engineering, and it is where timelines slip.
Channels
One channel is one build. WhatsApp plus website plus voice is closer to three, because each has its own constraints — WhatsApp templates, voice latency, web session handling.
How wrong it's allowed to be
An internal assistant can be occasionally wrong. An agent quoting prices to customers cannot. Higher stakes mean more grounding, more testing and more review — and that is cost.
Running cost
What you pay every month afterwards
Build cost gets all the attention; running cost is what decides whether the agent survives its first year.
Model usage
Charged per volume of text processed. Predictable once you know your conversation volume, and reducible through caching and using smaller models where they suffice.
Channel fees
WhatsApp Business API conversations and telephony minutes are billed by the platform, not by us. These are often the larger line at high volume.
Hosting
Ordinary application hosting for the agent and its retrieval layer.
Ongoing improvement
The honest one most vendors leave out. Agents drift as your business changes; someone has to watch traces and tighten it. Budget for it or it degrades.
A rule we hold ourselves to: if the running cost of a feature exceeds the value of the work it replaces, the feature has failed, however impressive it is.
FAQ
Common questions
Why aren't there prices on this page?
Because a number without your specifics would be marketing, not information. The same feature list can differ several times over depending on data readiness and integration count. Tell us your case and we will give you a real figure rather than a range designed to get a meeting.
What's the cheapest useful starting point?
One channel, one job, one system to write into — typically an agent that answers and qualifies incoming enquiries on WhatsApp and pushes them into your CRM. It is also the fastest to prove, which matters more than the price.
Is it cheaper to buy an off-the-shelf tool?
Frequently, yes, and we will say so. Off-the-shelf wins when your need is standard. Custom wins when the value is in your own data and processes, which no product ships with.
How do we avoid spending money on a failed project?
Scope the first agent so small it can prove or disprove itself quickly, keep a human in the loop, and agree up front what 'working' means in numbers. Most AI budget is wasted on projects that were too big to fail fast.
Want a real number?
Tell us the job, the channel and what systems it would touch. That's enough for us to come back with a figure rather than a range.