AI for the back office: drafting quotes from your own data

The bit you're already doing, slowly

A quote leaves your business. Before it does, somebody has:

• Looked the customer up in the CRM
• Checked what they've ordered before
• Pulled the right pricing tier from a spreadsheet (or, if you're lucky, the ERP)
• Looked up current stock and lead times for each line
• Copied it all into a quote template
• Formatted the totals, the VAT, the delivery
• Written the covering note
• Sent it

For a routine quote, that's half an hour to an hour. For something with a few unusual lines, it can run to half a day. Multiply by the number of quotes leaving your business each week and you have one of the bigger pieces of quiet work in the sales operation.

The interesting thing is: almost everything in that list is already in your data. The customer, their history, the prices, the stock, the lead times, the templates. The salesperson is mostly finding information that already exists, then assembling it into the right shape.

That's exactly the shape of work that becomes a draft on arrival.

Drafting from your own data

We've covered the inbound version of this twice already in this series — drafting replies in customer services and extracting structure from inbound documents. Quoting is the outbound mirror.

A customer enquiry lands — by email, web form, phone note, or eventually through a portal. The AI reads it, works out what's being asked, and starts looking things up.

The lookups it has access to — via MCP servers, the same plumbing as the other posts — typically include:

• The customer — who they are, their account history, their pricing tier, anything currently flagged
• Their order history — what they've bought before, at what price, how often
• Your product catalogue — matching their wording (or their codes) to your SKUs, with specs and substitutes if needed
• Your contracted pricing — for this customer, on these products, today
• Stock levels and lead times — live, from your warehouse and supplier systems
• Production capacity — if you make to order
• Similar quotes you've sent — for related products, to similar customers, recently

It assembles all of that into a structured draft quote that lands on the salesperson's desk. They open it and see:

• The customer's enquiry, parsed into clean line items
• Each line filled in with the right product, the right contracted price, current stock, current lead time
• A suggested margin position based on the customer's tier and recent quote history
• Notes on anything unusual — "this customer last ordered this in October at £42, current contracted price is £44.50", or "line 3 is a non-standard item, suggest checking with production"
• A covering note drafted in your house style

The salesperson reads it, adjusts the margin where their judgement says so, edits the wording if they want, sends. Total time for a typical quote: a few minutes instead of half an hour.

What the salesperson actually decides

The thing AI won't do is set the price. Or rather: it will suggest a price based on the contracted tier and recent quote history, but the call is still the salesperson's.

That's deliberate. Margin is the bit of the quote where commercial judgement matters most:

• Is this customer about to walk away to a competitor? Win it cheap.
• Is this a strategic account where you want to set a precedent? Hold the price.
• Is this a one-off enquiry from someone you've never heard of? Quote your standard rate, no discount.
• Is the customer asking for something that'll tie up production for two weeks? Price it accordingly.

The AI can lay out the data — "this customer's last 12 quotes averaged 28% margin, this product line is currently running at 31% across the book, two of the last five enquiries on this product line we lost on price" — but the decision is the salesperson's. The system supports the call; it doesn't make it.

Same pattern as the other AI posts. The AI does the looking-up, the cross-referencing and the drafting. The person makes the calls that need a person.

AI in the back office isn't about quoting more. It's about quoting better, faster, and putting the salesperson's day on the bits where their judgement actually moves the number.

Where this naturally goes next: portals

Once you've got AI drafting quotes from your own data, an interesting thing happens to the email itself.

Email is a poor interface for quoting. Most of what bounces back and forth between you and your customers — "can you quote on the attached?", "what's your stock on X?", "what's the lead time on Y?", "can I get an updated price on the order I placed in March?" — is the customer trying to do something they could do themselves, if you gave them a place to do it.

A customer portal or quote portal is that place. The customer logs in, sees their account, their orders, their invoices. They can request a quote directly — pick from your catalogue, add quantities, hit Request quote. The same AI machinery that drafts quotes from inbound emails drafts the quote here too — but instantly, while the customer is still on the page.

For routine quotes within the customer's normal pricing rules, the portal can show the price there and then, ready to convert to an order. For anything outside normal patterns — unusual quantities, non-standard items, a price below the salesperson's threshold — it routes to a person for sign-off, with all the same context already assembled.

This changes the customer experience meaningfully. "Can you quote on this?" stops being a 24-hour wait and becomes a 30-second look. The 80% of quotes that are routine flow through with no human intervention on either side. The 20% that genuinely need judgement still get it — but the salesperson opens them with the work already done.

We'll write more about portals in their own right — what they actually look like, how to think about pricing visibility (the "don't show my price to a competitor" problem), where the boundary should sit between self-service and human oversight, and how to roll one out without upsetting the half of your customers who like email. For now, treat AI-drafted quotes as the precursor: the same machinery that drafts a quote into a salesperson's inbox today is the machinery that drafts it onto a portal screen tomorrow.

What it changes

For the sales team:

• Time per quote drops to single-digit minutes for routine work
• Salespeople spend their day on the quotes that need their judgement, not the typing
• Quote turnaround time falls — usually from days to hours, sometimes to minutes
• Quotes are more accurate (right contracted price, right stock, right lead time, every time)
• The team starts seeing patterns they didn't before — "we lose 60% of enquiries from this region on lead time, not price"

For the customer:

• They get a quote when they want one, not when somebody got round to it
• It's right the first time (no "actually, that price has changed since" follow-ups)
• Eventually, with a portal, they don't have to ask at all

Where to start

Same pattern as the rest of the series:

• Pick one quote type — usually the one that's both highest-volume and most repetitive. Routine reorder quotes are the obvious starting point
• Map the lookups — what does the salesperson currently dig up before drafting that kind of quote? Each one is an MCP lookup
• Put the draft in the tool the team already uses — Outlook, Gmail, the CRM, the quote tool, wherever the work happens today. Don't make them learn a new system to get the benefit
• Keep the salesperson on the margin — let the AI suggest, let the human decide
• Tune from edits — every margin adjustment, every reworded sentence is a signal the system can learn from

A first useful version is usually four to six weeks for a single quote type. Once that's working, the same machinery extends — to other quote types, to inbound emails (the customer-services pattern), to inbound documents (the document-understanding pattern), and eventually to a portal where the customer drives it themselves.

If your sales team spends more time finding information than thinking about it, say hello.