AI at work by function: sales, support, operations, admin and marketing

How to read this guide

For each use case you’ll see three things:

• What it does: the job the AI takes on.
• What it needs: the information or access it depends on.
• What a person checks: where human judgment stays in the loop.

None of these require a big IT team. Some are features in tools you may already pay for. Others need AI connected to your own systems or a small custom build. The SBA’s guidance for small businesses lists many of the same ideas, from summarising meetings to drafting review replies, along with its advice to have a person review what AI produces. ^[1]

Sales

Sales teams in small businesses lose hours to the work around selling: notes, follow-ups, quotes and updating the CRM.

1. Follow-up emails after calls and quotes

• What it does: drafts a follow-up from your call notes or the quote, in your tone.
• What it needs: the notes or quote, plus a few examples of good follow-ups.
• What a person checks: accuracy of anything promised, plus the tone, before sending.

2. Call and meeting notes into the CRM

• What it does: summarises a call or meeting into key points, next steps and dates, then adds them to the customer record.
• What it needs: a recording or notes (with the other party’s permission where required) and access to your CRM.
• What a person checks: next steps and commitments.

3. First-draft quotes and proposals

• What it does: assembles a first draft from your templates, price list and the customer’s request.
• What it needs: current templates and pricing, kept in one place.
• What a person checks: every number and every promise. AI should never set prices on its own.

4. Lead sorting

• What it does: reads incoming enquiries, tags them by type, size or urgency, and routes them to the right person.
• What it needs: a short list of categories and who handles what.
• What a person checks: the categories at first, until the sorting is reliable.

Customer support

Support is often the clearest win, because so many questions repeat.

5. Answers from your own information

• What it does: answers common questions (hours, policies, order status, how-tos) from your FAQ, policies and product information, on your website, by email or in chat.
• What it needs: up-to-date, approved information, plus a clear rule for handing off to a person.
• What a person checks: new or unusual questions, complaints, refunds and anything sensitive. The SBA lists a website chatbot that answers common questions as one way AI can improve customer service. ^[1]

6. Triage and routing

• What it does: reads each request, works out what it’s about and how urgent it is, and puts it in the right queue.
• What it needs: your categories and urgency rules.
• What a person checks: anything flagged urgent.

7. Reply drafts for your team

• What it does: drafts a reply for a staff member to edit and send, pulling in order details or the relevant policy.
• What it needs: access to the order or account information the reply depends on.
• What a person checks: everything, before it goes out.

8. After-hours first response

• What it does: acknowledges every request immediately, answers what it can, and tells the customer what happens next.
• What it needs: clear limits on what it may promise.
• What a person checks: the follow-up the next working hours.

We run our own support this way: 24/7 support. Every request gets an immediate AI response, and we guarantee a response within four hours, day or night.

Operations

Operations is where scattered tools and spreadsheets cost the most time, and where pulling work into one place pays off.

9. Order and request intake

• What it does: reads orders or requests arriving by email, form or PDF, pulls out the details, and enters them into your system.
• What it needs: a clear destination (order system, spreadsheet or portal) and the fields you need.
• What a person checks: spot checks early on, then exceptions only.

10. Scheduling help

• What it does: drafts schedules, confirmations and reminders, and flags conflicts.
• What it needs: access to the calendar or booking system and your rules.
• What a person checks: final schedules and changes that affect customers or staff.

11. Inventory, lot and supplier records

• What it does: pulls lot numbers, dates, quantities and supplier details out of paperwork and links them to the right records, so a look-up takes minutes instead of hours.
• What it needs: consistent record-keeping and one place for the records to live.
• What a person checks: anything used for compliance or a recall decision.

12. One portal instead of five tools This one isn’t a single AI feature. It’s a change in where work lives, with AI handling the repetitive steps inside it. It’s also where we’ve seen the biggest gains:

Frisco Brands replaced five tools with one portal. Staff love it: everything’s in one spot, with instant access to every tool they need, so they never have to go anywhere else. Each team member figures they’re saving several hours a week, and Frisco saves at least 30 hours a week on lot tracing alone.

Bookkeeping and admin

AI can take on the reading and typing here. The decisions about money stay with people, including your bookkeeper or accountant.

13. Invoices and receipts

• What it does: reads invoices and receipts, pulls out vendor, date, amount and line items, and suggests a category.
• What it needs: access to where invoices arrive and to your accounting system, or a review queue in between.
• What a person checks: every entry before it’s posted, at least at first, and anything that moves money.

14. Chasing paperwork

• What it does: drafts reminders for missing documents, overdue invoices or expiring certificates.
• What it needs: a list of what’s outstanding and from whom.
• What a person checks: tone and timing, especially with good customers.

15. Summarising contracts and documents

• What it does: produces a plain-English summary of key dates, obligations and renewal terms.
• What it needs: the document. Keep sensitive documents in tools you control.
• What a person checks: anything you’ll rely on. A summary isn’t legal advice.

16. Watching for payment fraud AI also helps criminals write convincing fake invoices and “we’ve changed our bank details” emails. The FBI has warned that generative AI makes fraud more believable and lets criminals reach more people, faster. ^[2] Whatever AI you use in admin, keep one rule: any change to payment details is confirmed by phone, using a number you already have.

Marketing

AI is good at first drafts. Your job is to make them sound like you.

17. First drafts in your voice

• What it does: drafts emails, social posts, product descriptions and web copy from your notes.
• What it needs: examples of your best writing and a short list of words you do and don’t use.
• What a person checks: facts, claims and tone. Never publish AI copy unread.

18. Repurposing

• What it does: turns one piece (an article, a talk, a product launch) into a newsletter, a few social posts and a short web update.
• What it needs: the original piece.
• What a person checks: that nothing was exaggerated in the shortening.

19. Review replies

• What it does: drafts courteous replies to online reviews, good and bad. The SBA lists this as a use for AI. ^[1]
• What it needs: your guidelines for responding.
• What a person checks: every reply to a negative review.

20. Product and service descriptions

• What it does: writes consistent descriptions from spec sheets.
• What it needs: accurate specs.
• What a person checks: every claim, especially anything regulated, such as health, safety or food claims.

Pick one: a quick scoring table

Use case	How often? (1 to 3)	Time per week (1 to 3)	Rule-based? (1 to 3)	Score (multiply)
e.g. “Where’s my order?” replies
e.g. Invoice entry
e.g. Quote follow-ups

The highest score is usually your best first project. Then follow the steps in How to start with AI without wasting money, and keep the data basics in mind for anything that touches customer information.