The Short Answer
A single-purpose AI chatbot or automation — one that answers from your documents, or automates one workflow with a clear rule — is typically live in one to three weeks. Connect it to a system you already run, so it books, updates records or takes an action rather than only answering, and the realistic range moves to three to six weeks. Roll it out across several departments, several systems that have to agree with each other, or a regulated process, and six to twelve weeks is normal.
That is not a hedge. It is what the same build actually costs in time once you add the parts a quick quote leaves out: the system it has to read from, the person who has to approve access to it, and the testing against real, messy input rather than a demo. This guide sets out what moves a project between those ranges, what to ask a supplier who quotes a single number with no tier attached, and the two things that genuinely sit outside anyone's control. If you are building a product rather than adopting a tool inside your business, our guide to [how long an MVP takes to build](How Long Does It Take to Build an MVP in the UK? The Clocks You Control and the Ones You Do Not) covers the different clocks that apply there — app store review, FCA authorisation, data protection consultation. The tiers below match what [AI chatbot development](AI Chatbot Development) and [AI automation](AI Automation) are actually scoped and quoted against.
What Decides the Tier
| Single Build | Connected Build | Multi-System Rollout | |
|---|---|---|---|
| What it does | Answers questions or runs one rule-based task, using its own information | Reads from and writes to a system you already run — a diary, a CRM, an accounting package | Coordinates across several systems that have to stay in agreement, or across several departments |
| Typical range | 1–3 weeks | 3–6 weeks | 6–12 weeks |
| Chatbots | FAQ and handoff bot | Integrated support bot, booking into a real diary or CRM | Custom agent taking actions across several systems |
| Voice agents | Call answering | Booking and CRM agent | Multi-site and outbound |
| Automation | One workflow | A bundle of related workflows | Practice-wide programme with staged rollout |
| Integration | AI added to one tool | A knowledge system over your documents with permissions | AI coordinated across CRM, ERP and support |
| What it needs from you | A named contact and sample material | Read and write access to one system, plus someone who owns it | Access to several systems and the people who own each one |
| What most often adds time | Changing the brief mid-build | The system permitting less than its own documentation claims | Getting every system owner in the same room before testing starts |
Two forces push a project up or down that table more than anything else.
How many systems it has to touch, and what each one actually permits. A chatbot that only answers from a document you give it needs nothing from anyone else, which is why that tier is fast. The moment it needs to check a real calendar, write to a real CRM record, or read a customer's order history, its timeline stops being about the chatbot and starts being about that system: whether it has a documented API, whether the API allows writing as well as reading, and whether someone has to grant access before any of that is possible. We check this before quoting rather than after, and any supplier who has not is guessing.
Whose approval the build is waiting on. This is the one that catches people out, because it has nothing to do with engineering. A private integration into your own CRM or helpdesk — which is what almost every UK business actually needs — does not go through a vendor review queue at all in the ordinary case; it goes through your own IT admin or system owner approving access for a new connected app, which several major platforms now require by default rather than letting any user grant it themselves. That approval can take an afternoon or three weeks, and it depends entirely on how quickly your business can produce a decision, not on us.
A concrete version of this: two businesses ask for what sounds like the identical build, an AI agent that reads support tickets and drafts a reply in the helpdesk. The first has one person who owns the helpdesk account and can grant API access the same day it is asked for; that project runs at the pace of the build itself, typically the three-to-six-week Connected Build range. The second has the same helpdesk, the same brief, but the account sits with an outsourced IT provider who needs a change request raised and approved through their own ticketing process. The engineering is identical. The second project runs weeks longer, and none of that time is spent building anything — it is spent waiting for a decision that was never ours to make. Knowing which business you are before the build starts is worth more than any estimate.
The One External Clock That Is Genuinely Outside Anyone's Control
If a build includes a WhatsApp Business channel — increasingly the default for UK customer-facing chatbots, and the usual choice for the [estate agents](AI for Estate Agents) and retail-style businesses that field enquiries outside opening hours — Meta requires business verification through Meta Business Manager before the number can send template messages at scale. Guidance from providers working directly with Meta's process puts a straightforward verification at two to five business days, extending to as long as fourteen days if the submitted documents are incomplete or need a second look. Using a Business Solution Provider can compress the visible setup to as little as a day or two once verification itself is done, but the verification step is Meta's and nobody can promise a shorter one. We submit this in week one of any build that needs it, in parallel with everything else, precisely because it is the one clock in a chatbot or automation project that neither we nor you control.
The equivalent for a business wanting to list a packaged integration publicly on a platform's own marketplace — Salesforce's AppExchange is the clearest example — is a formal security review that typically runs six to nine weeks and can extend into months if it triggers a remediation cycle. That scenario is rare for the businesses we build for: it only applies if you are shipping a product to other companies through that marketplace, not if you want AI connected to your own CRM. We mention it only so nobody confuses the two and expects a private integration to take as long as a public one.
What to Ask a Supplier Who Quotes One Number
A single figure with no tier attached is answering a different question than the one you asked. Four questions separate a real estimate from a guess.
Which of the three tiers is this? If the answer does not distinguish between a standalone build and one that reads and writes to a live system, the number has not accounted for the part that actually varies.
Have you checked what our system allows, or are you assuming? The gap between "the platform's documentation says it has an API" and "your specific account, on your specific plan, with your specific permissions, can use it" is where estimates go wrong.
What happens to the timeline if the brief changes? It will, at least once. The honest answer is that a change in week one costs a conversation and the same change in week four costs the rework already done around it — which is why we quote a fixed price against a written scope and re-quote when the scope moves, rather than after the work is done.
Whose approval does the fast end of your range depend on? If the answer is "yours," find out now what that approval actually requires, because it is usually the reason a three-week estimate becomes a six-week one.
Next Step
If you have a date in mind, the more useful question than "how long will it take" is "which tier is this, and whose approval does it wait on." That is a free discovery call, not a quote pulled from a table. [Book one](our contact page), message us on WhatsApp, or call if you would rather talk it through first.
Frequently Asked Questions
Can we pay more to make it go faster?
Sometimes, and we will tell you honestly when it will and will not help. Money buys more people working in parallel on our side, which shortens the build-week portion of the estimate. It does not shorten Meta’s WhatsApp verification, and it does not make your own IT team approve system access any faster — those clocks run on their own schedule regardless of budget. A supplier who promises to buy their way past either is promising something they do not control.
Why did another agency quote us a completely different number for what sounds like the same project?
Because "build a chatbot" is not one project until the tier is fixed. One quote may assume a standalone FAQ bot with no integration; another may have priced in a CRM connection because that is what you actually described in the brief. Ask both suppliers which tier they costed, what they assumed your system permits, and whether the number includes the access and approval step on your side. The quotes usually turn out to be answering different questions rather than genuinely disagreeing.
Does adding WhatsApp make the whole project slower?
It adds a step, not a multiplier. Meta’s business verification runs in parallel with the rest of the build rather than after it, so submitting it in week one usually means it clears before the chatbot itself is ready to test. It only extends the overall timeline when the verification documents are incomplete and need a second submission, which is why we check what Meta will ask for before the build starts rather than when the channel is meant to go live.
What is the single most common reason a chatbot or automation project overruns?
Waiting on access. Not the engineering, and rarely the AI itself — it is the week spent waiting for someone at the client to approve a connected app, hand over API credentials, or confirm which of two systems is actually the source of truth. We ask for the named person who can make that call in week one, because a project with no single owner on the client side is the one most likely to slip.
Do you offer a faster, cheaper self-serve option instead of a full build?
For a single, standalone chatbot with no system connections, a self-serve tool can be genuinely faster to launch and is a fair choice if that is all you need — we will say so on the call rather than quote a build you do not need. Where a bespoke build earns its place is the moment the chatbot has to read from or write to a system you already run, which most self-serve tools either cannot do or do shallowly. We check which situation you are actually in before recommending either.