How to Use Customer History in AI Assistant Replies to Enquiries (and Stay GDPR-Safe)

BOTrento Team25 October 20267 min read

Why history makes a reply better

A generic reply to a returning customer reads like a form letter. A reply that remembers the last conversation reads like someone is paying attention. For a small sales team this matters most in three situations:

• Repeat customers. An events agency or equipment rental firm often hears from the same buyers every season. They expect you to know what they booked last time.
• Long threads. When an enquiry has moved across several emails and more than one person, the next reply must not ask for information the customer has already given.
• Open promises. If someone on your team wrote “I'll send revised numbers by Friday”, the next reply should deal with that, not ignore it.

None of this requires a complicated system. It requires the right text in front of the assistant at the moment it writes the draft.

What counts as useful history (and what does not)

Useful for a reply

• Earlier emails in the same thread and previous enquiries from the same contact.
• The last quote: what was included, what was excluded, and what the customer pushed back on.
• Stated requirements, such as venue limits, delivery windows or preferred contact person.
• Promises made by your team and their deadlines.
• Short, factual CRM notes, for example “prefers a single total price” or “needs a PO number on invoices”.

Better kept out

• Personal remarks about individuals, such as health, family or opinions unrelated to the deal.
• Internal comments about the customer's attitude or “difficult” behaviour.
• Payment or credit details that the reply does not need.
• Anything from years ago that no longer reflects the relationship.

A good test: would you be comfortable if the customer read the note you are feeding in? If not, it probably does not belong in the draft context.

Step by step: set up history-aware drafts

1. Pick one enquiry type. Start with returning customers asking for a repeat or similar job. It is the easiest place to see the value.
2. Clean the notes you will rely on. In Pipedrive, HubSpot or your spreadsheet, keep a short “customer facts” field: preferences, constraints, last quote reference. Move gossip and personal remarks elsewhere or delete them.
3. Give the assistant the thread, not the whole archive. The current conversation plus the most recent related quote is usually enough.
4. Ask for specific use of the history. For example: “Mention the previous quote only if the scope is similar, and answer any open promise first.”
5. Review each draft against a checklist: are the facts from history correct, is anything private included, is the price and date right?
6. Fix, then send yourself. The assistant drafts; a person approves and sends.
7. After two weeks, count the corrections. If you keep removing the same type of detail, adjust what you feed in.

An example of a history-aware draft

Suppose a returning customer writes: “We'd like the same setup as in spring, but for a larger group.” A history-aware draft might look like this:

Hello Anna,

Thanks for getting in touch. We'd be glad to repeat the spring setup. Last time we worked with the main stage, sound and one operator, and you asked for the operator to stay on site for the whole day. Shall I assume the same for this event?

For the larger group I'll need to confirm the guest numbers, the room layout and the load-in time. Could you send these? I'll then prepare an updated quote. You asked last time for a single total price, so I'll set it out that way.

Best regards,
Tom

The draft does not state a price or confirm availability. Those stay with a person, who checks the details before anything leaves the building.

GDPR guardrails for customer history

This is general guidance, not legal advice. For your own situation, talk to a lawyer or your data protection contact.

• Purpose. Use history to answer the customer's enquiry, not for unrelated profiling. If you want to do more, check it separately.
• Data minimisation. Send the assistant only the text needed for the reply. Less context means less risk and usually a cleaner draft.
• Accuracy. Old or wrong notes produce wrong drafts. Keep the “customer facts” short and up to date.
• Know where processing happens. Ask any vendor where data is stored and which provider processes the text. An EU server that calls an external AI API is not a closed computation, so treat it accordingly in your records.
• Human approval. A person should read every draft. This also protects you from AI inventing a detail that was never in the history.
• Access and deletion. You should be able to export and remove customer data if asked.

For a fuller checklist, see our answer on whether an AI sales tool is GDPR-compliant.

How BOTrento handles history in drafts

BOTrento Chief of Staff is a managed process for enquiries, quotes and follow-ups. It watches the enquiries in your mailbox and in the data you upload, and prepares draft replies, quotes and follow-ups in four languages (CS, EN, DE, FR). It works from the text of the emails and the data you import from your CRM or spreadsheet as CSV or XLSX. It does not read attachments.

It also picks up promises such as “I'll get back to you by Friday” and reminds you of them, and every working morning it shows a short “Today 1+3” overview. Nothing is sent without human approval, and it does not send anything by itself. It does not replace your CRM or accounting system.

On data: it is stored on servers in Germany (EU). The text needed for the task is processed by a commercial AI provider through its API, either through our account or, optionally, your own API key (currently Anthropic only; see the help page on API keys). Commercial API providers do not use the data for training. AI is included in the plan, with a monthly usage limit.

You can see how this looks in practice in the process demo. To judge it with your own threads, start the 14-day free trial: no card and no API key of your own are needed, and the 14 days begin when you first upload data.

Common mistakes to avoid

• Feeding in everything. More context does not mean a better draft. It means more to check.
• Over-personalising. Mentioning something the customer did not expect you to remember can feel intrusive. Stick to business facts from your own dealings.
• Trusting history over the new email. If the customer's new request contradicts an old note, the new email wins.
• Skipping review. Check names, dates, quantities and prices every time.

For related reading, see how an assistant categorises and prioritises enquiries and when templates or AI drafts work better. If you want to test this on your own enquiries, the 14-day trial plan shows what to measure.