Custom AI Agents & Chatbots
An assistant that answers from your pricing sheet, your policies and your past work — not from whatever it absorbed on the open internet.
Two kinds of assistant, two very different projects
Internal assistants
Built for your staff. A new hire asks “what is our policy on rush orders” and gets your actual answer with a link to the source document. Lower risk, faster to deploy, and usually where we recommend starting.
Customer-facing assistants
On your website or phone line. Higher value and considerably higher risk: everything it says is something your business said. These need tighter guardrails, escalation paths and human review of transcripts.
Grounded in your material, not guessing
The failure mode everyone worries about is a confident wrong answer. The fix is architectural: we build retrieval-grounded systems, meaning the assistant looks up relevant passages from your approved documents and answers from those, with citations back to the source. When it cannot find a grounded answer, it says so and hands off to a person rather than inventing something.
- Answers cite the source document so staff can verify
- Explicit “I don't know” behaviour with escalation to a human
- Scope boundaries — it declines topics outside its remit
- Full conversation logging so you can audit what it told people
- A content review process so answers stay current as policies change
What we build
Knowledge assistants
Trained on SOPs, policies, product specs, pricing and past proposals. Cuts the “who knows how we do this” tax on your senior staff.
Website assistants
Answer pre-sales questions, qualify enquiries, book consultations and hand warm leads to your team with context already gathered.
AI phone agents
Handle after-hours calls, take messages properly, book appointments and route emergencies. Effective for trades, practices and property management — and easy to do badly.
Document processing agents
Read incoming invoices, applications, insurance documents or plans and extract structured data into your systems.
An honest caution on phone agents. Voice agents are the most heavily marketed AI product for small business right now and the most frequently regretted. They work well for scheduling, message taking and simple routing. They work badly for anything emotionally charged or genuinely complex. We will build you one, and we will also tell you which of your calls should never reach it.
Security and data handling
Before a single document goes into an assistant we document where it will be processed, whether the provider can train on it, how long it is retained, and who can query it. For regulated data we use enterprise agreements with zero retention terms and, where HIPAA applies, a signed Business Associate Agreement. Consumer chatbot accounts never touch protected information.
Typical project shape
| Type | Timeline | Typical investment |
|---|---|---|
| Internal knowledge assistant | 4–6 weeks | [$X,XXX–$XX,XXX] |
| Customer-facing website assistant | 6–8 weeks | [$XX,XXX] |
| AI phone agent with booking integration | 4–8 weeks | [$X,XXX–$XX,XXX] |
| Document processing pipeline | 6–12 weeks | [$XX,XXX] |
Find out what AI is actually worth to your business
Book a free 30-minute working session. We map your three most repetitive workflows, estimate the hours and dollars on the table, and tell you plainly whether AI is the right tool — or whether it isn't.
Frequently asked questions
Can it be trained on our documents without leaking them?
Yes. Retrieval-grounded systems keep your documents in storage you control; the model reads relevant excerpts at query time rather than being retrained on your data. With enterprise API agreements your content is not used to train the provider's models and is not retained. We document this chain explicitly before we build.
What stops it from making things up?
Grounding, scope limits, and refusal behaviour — plus honest testing. We build an evaluation set of real questions from your business, including the hard and adversarial ones, and measure accuracy before launch. We share those results with you, including the failures.
Will customers know they are talking to AI?
Yes. We disclose it clearly. Beyond being the right thing to do, undisclosed AI in customer communication is a growing legal and reputational exposure, and customers are markedly less forgiving when they feel tricked.
What does it cost to run each month?
Depends on volume. A low-traffic internal assistant might cost $20-$80 a month in API usage; a busy customer-facing assistant or phone agent runs higher. We model this from your actual call and enquiry volumes in the proposal.
How do we keep it up to date?
We build a content refresh process into the handover — usually a shared folder or CMS where updating the source document updates the assistant's knowledge. Stale answers are the most common reason these projects decay, so we plan for it from day one.