Skip to main content

AI Services

AI Chatbot Development

The difference between a useful chatbot and an embarrassing one is grounding. A model answering from memory will invent things confidently.

An AI chatbot answers customer questions from your own content rather than from a model's memory. The Nexclick grounds it in your documents, cites the source so answers can be checked, and sets explicit rules for handing over to a person.

Book a 20-minute callProject, then retainer · AI Services from £3,500

Is this you?

What usually prompts the call

  • Your team answers the same twenty questions every day and has no time for the rest.
  • You installed an off-the-shelf chatbot and it answers everything wrongly and cheerfully.
  • Customers cannot find information that genuinely exists on your site.
  • Out-of-hours enquiries sit unanswered until the next morning.

What we do

The actual deliverables

Things that appear on an invoice, not adjectives.

Ground it in your own content
Retrieval over your documents, site and knowledge base, so answers come from source material rather than from what the model happens to recall.
Citations on every answer
Which document or page the answer came from, so a customer or a colleague can check it. This alone changes how much the output can be trusted.
Explicit scope boundaries
What it is allowed to discuss and what it must decline. A bot that answers questions about pricing it was never given is worse than one that says it does not know.
Handover rules
Written and tested escalation — complaints, anything urgent, anything sensitive, and repeated failure to help. The model never improvises on those.
Integration with your systems
Order status, account details or booking availability where useful, so it can answer specifics rather than generalities.
Transcript logging and review
Every conversation reviewable, so you can see what it got wrong and improve it. A bot with no transcript log cannot be improved.
Tuning on real conversations
The first month of live usage reviewed and the system adjusted. This is where most of the quality comes from, not from the initial build.

Comparison

Off-the-shelf chatbot, grounded custom bot, or neither?

Chatbots are sold as one category and are three quite different things. This is what each actually gives you, so you are not paying for the wrong one.

Rule-based / FAQ botOff-the-shelf AI botCustom grounded bot
Answers unanticipated questionsNoYes, sometimes wronglyYes, from your content
Risk of invented answersNoneRealLow — grounded and cited
Cites its sourceN/ARarelyYes
Knows your pricing and policiesIf you typed them inOnly what it scrapedYes, from source documents
Integrates with order or booking dataNoLimitedYes
Setup effortHoursDaysWeeks
CostLowLow to moderate subscriptionHigher build, modest running
Tuning from real transcriptsManualLimitedYes
Escalation controlBasicBasicExplicit, tested rules
Best forA dozen fixed questionsSimple deflection, low stakesSpecific policies, real integration, brand risk

How it works

Step by step, with timeframes

Timeframes are typical rather than guaranteed, and they assume we get account access and approvals when we ask.

  1. 01Week 1–2

    Establish the question set

    A month of real enquiries categorised, so the bot is built against what people actually ask rather than what you expect them to.

  2. 02Week 2–5

    Build retrieval and grounding

    Content ingested, chunked and indexed, with retrieval quality tested against the real question set before any interface work.

  3. 03Week 5–7

    Interface, boundaries and handover

    Chat interface, scope rules and escalation paths, tested including the questions it should refuse.

  4. 04Week 7–12

    Launch and tune

    Live with transcript review weekly for the first month, then monthly. Quality improves most in this period.

What you get

Reporting and ownership

  • A tested question set built from real enquiries, so the bot is judged against reality.
  • Citations on answers, so anything it says can be checked against source.
  • Written escalation rules, tested, covering complaints and anything urgent or sensitive.
  • Full transcript logs you own, reviewable and exportable.
  • Running costs modelled at your actual conversation volume before you commit.

Tools and platforms

  • Commercial LLM APIs (OpenAI, Anthropic)
  • Vector database for retrieval
  • Your CMS, helpdesk and knowledge base
  • Transcript logging and review tooling
  • Evaluation sets built from real questions

Timeline

How long this actually takes

Seven to twelve weeks including the tuning period. The system will be noticeably better at month three than at launch, because real conversations surface gaps no amount of testing predicts. Two honest points. A chatbot cannot answer questions your content does not address — if the information does not exist in writing, the project is partly a documentation project. And deflection rates depend on your question mix: where most enquiries are genuinely simple, deflection can be substantial; where they are varied and specific, it will be modest, and we will estimate it from your actual tickets rather than quoting a vendor figure.

Pricing model

Project, then retainer

Fixed price for the build, then a small monthly fee for hosting, monitoring and tuning. Model API costs are yours directly and shown at cost.

Full pricing

Questions

AI Chatbot Development questions

Will it make things up?

Grounding makes it far less likely and does not make it impossible. Answers come from retrieved documents rather than model memory, citations let anyone check, and scope boundaries stop it discussing things it was not given. What we will not tell you is that the risk is zero, because it is not.

What if the answer is not in our content?

It says so and offers to hand over, rather than guessing. That behaviour is configured deliberately. It also means the build surfaces gaps in your documentation — a chatbot project is frequently a documentation project wearing a different name.

How much support volume will it actually deflect?

It depends entirely on your question mix, so we estimate it from a month of your real tickets rather than quoting an industry figure. Where most enquiries are simple and repeated, deflection can be substantial. Where they are varied and account-specific, it will be modest.

What are the monthly running costs of a chatbot?

Model API costs scaling with conversation volume, plus hosting. For most small and mid-sized businesses that is tens of pounds a month rather than hundreds, though it depends on conversation length and model choice. We model it at your volume during scoping.

Can it access order or account information?

Yes, with an integration and appropriate authentication. It changes the bot from answering general questions to answering the customer’s specific question, which is a substantial difference in usefulness — and it raises data protection considerations that get scoped explicitly.

How is this different from a RAG knowledge base?

A chatbot is a customer-facing conversational interface. A RAG knowledge base is usually internal — staff asking questions of your own documents. The retrieval technology is largely the same; the audience, interface and risk profile are not, which is why they are separate services.

Tell us what you are trying to fix

A 20-minute call, no pitch deck. The Nexclick will tell you what we would do, roughly what it costs, and whether we are the right people for it.