Repeat questions start from a draft you check
Sounds familiar? We answer the same questions over and over, from customers and from our own staff.
If your team types out the same answers about orders, policies or procedures every day, an AI assistant can draft them from content you approve. In the first stage only your staff use it, and a person checks every draft before it is sent. A pilot measures how often the drafts are wrong against an error rate you agree on up front. A customer-facing version needs the pilot to meet that rate first, and it would always say it is an AI and offer a route to a person. Because the setup depends on your questions, documents and data, a specialist agrees with you on the use case and its limits on a short call before scoping.

What you get
- Agreed use case, listing the questions the assistant may and may not answer
- Written data rules on what content and information may go to the AI provider
- Staff-facing assistant built on your approved sources
- Review step so a person checks and sends every draft
- Pilot review with the measured error rate against the agreed target
- Record of questions, drafts and corrections from the pilot
- Running-cost estimate for AI model and platform usage at your volumes
- Plan for AI disclosure, handoff to a person and answer logging, required before any customer-facing stage
Is this right for you?
- You answer the same customer questions by email or chat many times a week
- You field the same internal questions about policies, price lists or procedures
- You keep your policies, FAQs and product information written down and current
- You can name a person to check drafts and keep the source content up to date
- You want to try AI on one real task before it goes in front of customers
What it changes
Repeat questions start from a draft
Your staff open a drafted answer, check it against your sources, correct it and send it. The person who sends it stays responsible for the answer.
A measured error rate
The pilot records how often drafts were wrong on your real questions. Whether to go further is decided on that figure.
Written rules on data and cost
You have a record of what the assistant may use, what goes to the AI provider and what it costs to run each month at your volumes.
Scope and delivery
How we deliver it, what we need from you, what is not included. Timeline: Agreed after a short call; plan for 4–8 weeks for the staff assistant and its pilot
How we deliver it
-
Agree on the use case and limits
On a short call, we confirm which questions the assistant handles, which sources it may use and what data must stay out.
-
Prepare the sources
Your approved documents are gathered and tidied, and the assistant is set up to draw only on content you have signed off.
-
Build the staff assistant
The assistant is set up in the agreed tool with a review step. Nothing reaches a customer without a person checking it.
-
Run the pilot
Your team uses it on real questions for an agreed period while drafts and corrections are logged. The measured error rate is then compared with the target.
-
Decide the next stage
You choose to keep it staff-facing, adjust it or plan a customer-facing version with AI disclosure and a route to a person.
What we need from you
- Approved source content, such as policies, FAQs and product information
- A named person to check drafts during the pilot
- Real example questions and the answers you consider correct
- Agreement on what data may be sent to an AI provider
- An account with the AI provider or platform, with usage paid by you
What's not included
A specialist looks at your situation first, so the scope is agreed after a short call.
Not included
- Answers sent to customers without a person checking them, during the first stage
- Voice or phone bots
- AI model and platform usage fees, paid by you for as long as the assistant runs
- Legal sign-off on your use of AI, which is work for your legal adviser
Limits to know
- The assistant starts staff-facing: it drafts, and a person checks and sends. A customer-facing version comes only after the pilot meets an agreed error rate, with AI disclosure and a route to a person at all times.
- Answers can be wrong even when they come from your approved sources, so a person stays responsible for what is sent.
- Questions and source content go to the AI provider you choose, under its terms. What data may be used is agreed before anything is built, and that agreement is not legal advice.
- AI model and platform usage fees are ongoing and paid by you. The source content needs updating whenever your information changes.
Where it fits
What usually follows, what may already be in place, and what is planned alongside, with the reason for each.
This service: AI assistants with human review
4 connected services
You may already have this
Skip it if it's already in place.
-
Automation & AI assessment
Confirms the use case and data sensitivity first.
Keeps it running
Keeps it working after launch.
-
Automation & CRM care
Sampled answer reviews and content updates.
Part of these packages
-
Practical AI implementation
Pilot one workflow: One time-boxed pilot, staff-facing first, with a person reviewing each output and a success measure agreed up front.
Read before you decide
Common questions
Why not put a chatbot on our website straight away?
A bot that answers customers directly speaks for your business, and customers can hold you to what it says about prices, policies or refunds. Starting with staff lets you measure the error rate on real questions first. If it meets the agreed level, a customer-facing stage can be planned with AI disclosure and a route to a person.
Which AI provider do you use, and what happens to our data?
We choose the provider and platform with you, based on your data rules and volumes, and name them in the scope. Your questions and source content go to that provider under its terms. Those terms on storage and model training differ by provider and plan, so they are checked before anything is built.
What does our team have to do during the pilot?
A named person checks every draft, corrects it where needed and sends it. Those corrections are logged and used to measure the error rate. Someone also needs to keep the source content current.
Who looks after the assistant once it runs?
Your sources need updating as your information changes, and answers should be sampled for mistakes. Automation & CRM care can review a sample of answers and update the content each month. Handover covers logins, prompts (the written instructions the assistant follows) and documentation, with accounts set up for transfer to you where each provider allows.