Three kinds of assistant
| Type | Who uses it | What it does | Typical sources |
|---|---|---|---|
| Internal knowledge assistant | Your staff | Answers how-to, policy and product questions, with links to the source document | Handbooks, SOPs, price sheets, past proposals, training docs |
| Customer-facing assistant | Website visitors, customers | Answers common questions, collects details, books appointments, hands off to a person | Service pages, FAQs, service area, hours, booking rules |
| Sales follow-up agent | Your sales team (works in the background) | Drafts follow-ups on open estimates, re-engages old leads, summarizes each lead's history | CRM records, past emails, estimate history |
Internal knowledge assistant
The most common first build and usually the fastest payback. New hires stop interrupting senior staff. Senior staff stop answering the same question for the fifth time this week. Every answer cites the document it came from, so people can check it.
Customer-facing assistant
Answers the questions your front desk answers 30 times a week, collects the details needed for a quote and books the appointment. When a question falls outside what it knows, it hands off to a person instead of guessing. Customers always know they're talking to an AI assistant.
Sales follow-up agent
Works through your CRM, drafts follow-up messages for open estimates and stale leads, and queues them for a salesperson to approve. Over time, low-risk messages can send automatically if you decide they should. This one overlaps with workflow automation, and we often build them together.
Where the assistant lives
We put it where people already work, so nobody has to open another app.
- Microsoft Teams or Slack for internal assistants
- Your website as a chat window for customer-facing assistants
- Text messaging for customers who prefer it, through your existing messaging platform
- Inside Copilot, ChatGPT or Gemini as a custom agent, GPT or Gem, if your team already uses one of those daily
If you're already licensed for Microsoft 365 Copilot or Gemini, a custom agent inside that tool may be all you need. We'll tell you when that's the cheaper answer. See Copilot and Gemini implementation.
Guardrails we build in
An assistant that makes things up is worse than no assistant. Every build includes:
- Answers from your documents only. The assistant is instructed to answer from approved sources and to say "I don't know" when the answer isn't there.
- Source citations. Internal answers link to the document they came from.
- Scope limits. Customer-facing assistants don't quote prices outside your published ranges, give legal or medical advice, or promise dates.
- Access controls. Staff see only answers from documents they're allowed to see. HR and finance documents stay out unless you decide otherwise.
- Human handoff. Defined triggers send the conversation to a person: an upset customer, a complex question, a request to talk to someone.
- Logs and review. You can see what people asked and what the assistant said. We review a sample each month during the support period.
- Business-grade tools only. We build on platforms whose business terms don't use your data for model training by default, and we confirm the retention settings.
What we need from you
- The source documents. We help sort what's current, what's outdated and what should stay out.
- A subject-matter owner. One person who reviews answers during testing and owns updates afterward.
- A list of real questions. The 30 to 50 questions people ask most. These become the test set.
Messy documents are normal. Part of every build is cleaning up the sources the assistant will rely on, and that cleanup often helps your team as much as the assistant does.
How a build runs
- Scope (week 1). Pick the assistant type, sources, where it lives and what it should never do.
- Prepare sources (weeks 1–2). Gather, clean and organize documents. Remove outdated versions.
- Build and test (weeks 2–4). Test against your real questions. The subject-matter owner grades the answers. We fix gaps until accuracy meets the bar we agreed on.
- Pilot (1–2 weeks). A small group uses it on real work. We tune it.
- Launch and train. Roll out to everyone, with a short training session on how to use it and when to double-check it.
What it costs
Build: most Kansas City projects land between $4,000 and $20,000. Three to six weeks. An internal assistant on a clean set of documents sits at the low end. A customer-facing assistant that books appointments and connects to your CRM sits at the high end.
Running cost: $200–$800 per month, covering hosting, AI usage and routine updates to the sources. High-volume customer-facing assistants can run higher; we'll estimate yours during scoping.
Full details on our pricing page.
Book a Free AI Assessment Call
Common first builds by industry: a missed-lead responder for contractors, a policy-and-procedure assistant for professional firms, and a tenant FAQ assistant for property managers.
Frequently asked questions
Will the assistant make things up?
It can, which is why we build against it. The assistant is restricted to your approved documents, instructed to say when it doesn't know and tested against your real questions before launch. Internal answers cite their source so people can check. No assistant is perfect, so we also set rules for which answers always need a human check.
Is our data used to train someone else's AI?
Not on the platforms we use. We build on business plans and APIs whose terms state that customer data isn't used for model training by default, and we check retention settings during setup. We never build on free consumer accounts. Your AI policy should also spell out which data can go into the assistant.
How do we keep the answers current?
Each assistant has a subject-matter owner on your team and a simple process: when a document changes, the new version replaces the old one in the source folder. We set that up so it doesn't need us. During the support period we review answer logs monthly and flag questions the assistant couldn't answer, which usually point to a missing document.
Can it connect to our CRM or job software?
Usually, if the software has an API or a standard integration. Common CRMs, help desks and field-service platforms do. Older or custom systems sometimes need a workaround, such as a scheduled export. We confirm what's possible during scoping, before you commit to a build.
What's the difference between this and just using ChatGPT?
ChatGPT on its own knows the internet, not your business. A custom assistant knows your policies, products, prices and past work, follows your rules and lives where your team works. If your needs are simple, a custom GPT inside ChatGPT Business or a Gem in Gemini may be enough, and we'll say so. We'll walk through the trade-offs on the free call.
Can customers tell they're talking to AI?
They should, and ours say so. Every customer-facing assistant we build identifies itself as an AI assistant and offers a way to reach a person. Hiding it damages trust when customers find out, and a customer who knows they're talking to an assistant asks clearer questions. Clear disclosure is the simpler path.
See what an assistant could answer for your team
Bring a list of the questions your team gets asked most. On a free 30-minute call, we'll tell you which kind of assistant fits and what it would take.