Internal knowledge assistant for staff questions

What does an internal knowledge assistant actually do?

It sits where your team already works, usually a chat window or an internal tool, and answers questions like "what is the refund window on this plan", "which form do we send a new supplier", "what did we agree with this client in March". Instead of guessing, it reads from a defined set of sources you approve.

The useful version is narrow. It is tied to specific content, with specific permissions, and it refuses to answer outside that. The version that causes trouble is the one pointed at every file in the business with no rules about who can see what.

Four things separate a working assistant from a demo:

• Approved sources only. It answers from documents you have signed off, not from the open internet and not from its own training data.
• It admits ignorance. "I do not have that" beats a confident wrong answer every time.
• Clear permissions. A new starter and a director should not get the same answers about pay, contracts or client finances.
• A human handover. When the question touches money, legal commitments, complaints, regulated advice or anything sent out in your name, it routes to a person.

Those are the same principles we apply across our AI knowledge assistant work and the rest of our AI and automation services.

How much does an internal knowledge assistant cost?

We do not have a single product price, because every build is scoped around what the business needs and already has. Here is how the money actually works with us.

Payment default is 50% upfront and 50% on launch for projects, audit paid in full, retainer monthly in advance. The signed scope controls. Running costs such as hosting, software and AI usage are paid directly to the providers from your own accounts, so you see exactly what the assistant consumes and there is no mark-up in between. More detail on the paid audit and on AI chatbot pricing.

Confirm scope, tax treatment, currency and service allowances in writing before anything starts.

How do you build one without it going wrong?

Our process is the same six steps on every project, and it exists because the failures are predictable.

1. Audit. We map how the business runs, which tools hold the answers, and where staff currently go when they are stuck. Most businesses find the real source of truth is two people's heads, not the shared drive.
2. Design. A written scope: which sources are in, which are excluded, who sees what, which questions must escalate to a person, and how we will measure success.
3. Build. Inside your accounts, keeping what already works. You own the code and the data.
4. Test. Real questions, edge cases and failure paths before anything reaches staff. We deliberately ask it things it should refuse.
5. Launch. In stages, with the team trained and documentation handed over. Start with one team or one topic.
6. Improve. Review the numbers, see which questions it failed, agree the next changes.

Our advice on sequencing: start with the one journey that causes the most chasing, connect only the sources that journey needs, and add the rest once staff are actually using it. A pilot on expenses and onboarding questions teaches you more than a six-month project to index everything. See how we work and our wider workflow automation work.

Why a wrong answer is your problem, not the model's

In Moffatt v Air Canada (2024), the British Columbia Civil Resolution Tribunal held the airline liable for incorrect fare information its website chatbot gave a customer, and rejected the argument that the chatbot was responsible for its own actions. That was a customer-facing bot, but the lesson carries: if an internal assistant tells staff the wrong policy and staff repeat it to a client, the business carries that. Which is why we keep a person on decisions about money, legal commitments, complaints and messages sent in your name.

Custom assistant, off-the-shelf tool, or a better wiki?

If an existing tool fits without workarounds, use it. We say that to people who ask us to build. There is a fuller breakdown in custom AI chatbot vs off-the-shelf, and a related comparison in custom software vs SaaS.

When an internal knowledge assistant is not the right choice

Three situations where we would tell you not to build one:

• An existing tool already fits without workarounds. If your team's questions are all answerable from one platform you already pay for, buy the add-on rather than commissioning a build.
• The process is still changing week to week. An assistant trained on this month's policy becomes a liability next month. Settle the process first, then automate the answers.
• You need it next week. We never promise a delivery time before a scope is signed. If the deadline is immediate, write a one-page crib sheet and revisit the build later.

Two more honest caveats. If the underlying documents are out of date or contradictory, the assistant will return contradictions faster than a human would, so document clean-up is part of the job and part of the cost. And if the real problem is that work falls through the cracks rather than that staff cannot find answers, you may want process automation or an internal tool instead. A cheaper first step is often a no-code workflow. Zapier's own help centre explains that each successful action step counts as a task, with triggers, filters and paths excluded, and that at the plan limit new runs are held unless extra tasks are paid for. That ceiling is usually what pushes people towards a custom build. See Zapier vs custom automation.

What we have built that is closest to this

We will not claim knowledge-assistant case studies we do not have. These are the relevant builds, and they show the permissions, approval and draft-then-review patterns an internal assistant needs.

• TMMB Academy came to us with a fragmented funnel, rising software subscriptions and no single system. We built one portal with role-specific workspaces for admins, coaches, creators, sales, assistants and brands, covering onboarding, creator performance, subscriptions and payments, course modules, calls, messages, community resources, brand deals, deliverables and operational alerts. The implementation ran for more than 60 days. TMMB was able to take on greater capacity, improve scalability and retention, and reports generating six figures within roughly two to three months (client-reported, shared with permission).
• MRM Group, a UK property development and investment company, has an authenticated internal command centre for clients, deals and tasks, with document handling, task workflows, reminders and audit history, plus human approval at sensitive financial-promotion and AI decision points.
• The Agentryx platform, our own product, includes AI that turns a call summary into tasks and a draft follow-up email for a person to check. That draft-then-review pattern is the one we use for anything an assistant produces on your behalf.

On Trustpilot, Denzil Jones wrote: "When it comes to building systems to acquire clients, Quam is your guy." Stefano Ezeokaka wrote: "Quam introduced me to the idea of having a solid infrastructure and helped me build it." Agentryx was founded in Manchester by Quam Balogun and works with established businesses in the UK and US, including clients in Stockport, Altrincham and Hale and Warrington.