AI chatbot development cost

What drives build cost

Scope

More topics mean more sources, rules and testing.

Source preparation

Scattered or outdated documents need sorting and rewriting.

Platform or custom

Configuring a product is usually quicker than building from scratch.

Integrations

Connecting to orders, bookings or a helpdesk adds development and testing.

Testing depth

Testing many real questions takes time but reduces risk.

Ask for running costs to be estimated at your expected monthly conversation volume.

Chatbot projects compared

How chatbot projects are priced

Running costs

AI model providers often charge according to usage, so the bill grows with the number and length of conversations. Platforms may charge a monthly fee, sometimes with conversation limits. Ask for an estimate at your current volume and at double it, and check whether fees are paid directly by you or passed through by the provider.

Source preparation

If your help information is current and well organised, preparation is quick. If it is spread across old documents, emails and staff knowledge, writing clear source material can be a substantial part of the project. That work also improves your website and staff guidance, so it has value beyond the chatbot.

Example: a pilot to test value

The cost of looking after it

Someone needs to review conversations, update sources when policies change and adjust instructions when answers drift. Whether that is your staff or a provider's monthly service, budget the hours. A chatbot nobody maintains becomes less accurate over time.

Integrations

Answering general questions needs no integration. Looking up a customer's order or booking means connecting securely to other systems, handling identity checks and testing failure cases, which adds cost and data protection considerations.

Comparing quotes

Check scope, source preparation, platform or custom build, integrations, testing and shared results, handover design, data protection support, running cost estimates, ongoing support and ownership of configuration.

Keeping the cost down

Start with a focused scope, tidy your help information first, use a platform if it fits, defer integrations until the basic chatbot proves useful and review conversations in-house.

Judging the value

Compare total cost with what it replaces or improves: emails answered, enquiries outside office hours, staff time freed. Count only what you can measure, and include the time spent maintaining it.

Data protection work

A data protection impact assessment, privacy notice updates and supplier terms take some time and, if you use an adviser, some cost. ICO guidance says AI use usually requires a DPIA, so include it in the budget.

Testing

Testing against a large set of real questions, including difficult and out-of-scope ones, takes time but is how problems are found before customers see them. A quote with little testing may be cheaper and riskier. Ask how many test questions are included and whether results are shared.