Small businesses can use AI in sales to sort enquiries, prepare calls, draft replies, assemble proposals, explain options, plan follow-up, record actions, review stalled enquiries, analyse lost sales and check promises before sending. Start with one repeated task that produces a reviewable draft, then measure the time spent checking it.
These are real sales workflows presented through illustrative businesses, messages and figures. They are not ten claimed client case studies. Product capabilities are identified where relevant; the examples show how to put a capability to work and where a person still needs to decide what happens.
1. Turn a picture-framing enquiry into a usable job brief
A picture framer receives a message containing several useful facts and several gaps: 'Two prints, roughly 30 by 40, black wood if possible. One is a gift for the 20th. Could you send options? I can bring them in on Friday.' Before discussing a price, the framer needs to distinguish measurements, preferences and deadlines.
The old process is to reread the message, copy details into a notebook and remember which questions still need asking. An AI assistant can draft a structured enquiry brief. The framer checks it, adds the missing questions and keeps it beside the original message.
Extract a draft framing brief from the message below.
Use only stated facts. Return: items, dimensions, frame
preference, deadline, visit preference and questions to ask.
Preserve words such as roughly and if possible.
Do not choose materials, calculate a price or confirm a date.
Message: [approved enquiry text]An illustrative output reads: 'Items: two prints. Dimensions: approximately 30 by 40; units and measurement basis missing. Preference: black wooden frames if available. Deadline: one gift needed for the 20th; month unconfirmed. Visit preference: Friday, not booked. Questions: units, paper or image size, month, glazing and mounting needs.'
This is useful because it makes uncertainty visible. A weaker output might convert the dimensions to centimetres without evidence and treat Friday as an appointment. The reviewer fixes those changes before the brief becomes a customer record. The task succeeds when the next conversation begins with the right questions rather than a polished guess.
Here is the main worked example using illustrative operating figures. The framer receives 40 similar enquiries in a month. Manually preparing each brief takes six minutes, or 240 minutes in total. With AI, reviewing and correcting each brief takes three minutes, or 120 minutes. A further 30 minutes goes into checking the saved instructions and reviewing awkward cases.
The new total is 150 minutes, leaving 90 minutes of monthly capacity. At an illustrative internal time value of $28 an hour, that capacity is worth $42. This is not necessarily a cash saving. It matters if the owner uses the time for consultations, production or work that otherwise goes unfinished.
Budget an initial two hours to agree the fields and another hour to test ten old, minimised enquiries. Keep that three-hour setup cost separate from the monthly calculation. Any new software charge must also be included; use an approved tool already available for the trial if possible.
Over the first month, the framer tracks missing dimensions, invented appointments and the number of briefs needing major correction. The chosen release requirement is zero invented commitments. If mistakes persist, the framer narrows the task to extracting item descriptions and leaves date handling manual.
Once the brief is checked, ordinary business steps continue: inspect the prints, confirm specifications, calculate the quote and agree collection. AI has improved one small part of the sale. It has not replaced material knowledge or made an unmeasured price accurate.
2. Draft the first reply to a music-lesson enquiry
A music teacher repeatedly answers questions about lesson length, availability and preparation. AI can draft a reply using approved facts and the prospective pupil's actual question. The teacher checks the details before sending and keeps control of booking decisions.
For an illustrative enquiry, the customer writes, 'I am a complete beginner. Do I need to read music before starting, and are evenings possible?' The teacher supplies two approved facts: beginners can start without reading music, and evening availability must be checked individually.
A useful sample reply is: 'You do not need to read music before starting beginner lessons. I can check evening availability if you tell me which days usually work for you. We can also discuss what you would like to learn at the first conversation.' It answers the question before asking for more information.
A poor draft adds, 'I have several evening spaces and you will be playing confidently in a month.' Remove both claims. Neither follows from the teacher's two approved facts. This is the review that determines whether quicker writing is actually useful.
If the enquiries already live in a CRM, the drafting may be built in. Pipedrive's AI email creation adds a "Write my email" option to the composer: you describe the email, choose a tone and length, and it generates a draft you can then edit. Its help page limits the feature to the upper plans rather than the entry tier, an admin has to switch on email content generation in the Email AI settings, and it doesn't work for group emails. Check your own plan before relying on it, and review the draft in exactly the same way. The feature can write the email; it cannot tell you what the teacher is able to promise.
Begin with five common enquiry types and save approved examples. Measure total drafting and review time, not just the seconds until text appears. A saved reply may outperform AI when the answer rarely changes.
3. Prepare a tutoring consultation without rereading every note
A tutoring agency may hold several conversations before arranging a suitable tutor. Staff need the learner's broad level, subject, preferred times and the family's unresolved questions. They should not have to read every old message while the next call is waiting.
AI can prepare a short call brief from approved sales notes. HubSpot, for one, offers record summaries on all its plans, including the free CRM: Breeze Assistant summarises a contact, company, deal or ticket from its properties, notes and logged activities, once an admin has switched on the relevant AI settings. Treat the summary as an aid to preparation, and keep the underlying records available for anything that affects a promise or decision.
Illustrative example: the record contains an initial request for weekend maths, followed by a later message saying weekday evenings would now be better. A useful brief says, 'Current preference: weekday evenings, replacing earlier weekend request. Subject: maths. Level: needs confirmation. Main question: whether the same tutor can continue next term.'
The reviewer checks the chronology. If the summary lists weekends as current, the agency risks spending the call offering the wrong options. Give dated updates priority and flag contradictions instead of combining all preferences into one confusing list.
Keep sensitive educational details out of routine sales preparation unless they are needed and handled appropriately. The brief should help staff ask relevant questions, not label the learner or predict their ability.
A useful check is whether staff can name the customer's unresolved question after reading the brief. Count errors in timing and previous commitments, as well as minutes saved. The tutorial on updating CRM records after calls explains how better notes at the end of one conversation improve preparation for the next.
4. Assemble a language-school proposal from approved parts
A language school preparing a small-group proposal may reuse the same course description, attendance conditions and joining process. AI can assemble those approved parts around a particular request. The owner still verifies the timetable, tutor capacity, fees and any exceptions.
Illustrative example: a group asks for six weekly sessions of 90 minutes. The school supplies an invented approved rate of $120 per session and a materials charge of $15 for each of eight learners. The arithmetic is $720 for sessions plus $120 for materials, giving $840.
The assistant drafts the scope and a price table from those inputs. It should not turn six sessions into an eight-week programme because another document contains an older course description. Ask it to show every quantity and rate so the reviewer can compare the proposal with the source.
A plausible draft says, 'Includes ongoing support between sessions.' If no approved source offers that support, delete the sentence or ask the owner whether it should become a separately defined service. Pleasant language can create extra work just as easily as an incorrect total.
Keep one approved set of commercial conditions and date the proposal. If the timetable changes, create a revised version rather than editing an old copy without a record. For a fuller drafting process, see preparing business proposals from approved material.
5. Explain picture-framing options without inventing an upgrade
Customers often need help understanding a choice rather than another sales pitch. A framer can use AI to turn approved specifications into a plain comparison: what each option includes, what changes and what remains the same. The product facts must come from the framer's current catalogue.
Illustrative example: the owner provides three complete approved totals for the same job: $95, $130 and $165. The differences are a specific frame profile and specified glazing options. The assistant produces a short table with the exact options and prices, leaving performance claims out unless supplied.
A weak sample output labels the highest price 'museum quality' or claims it prevents all fading. Those phrases need evidence and may misrepresent the product. Replace them with the actual product description and the limits supplied by the manufacturer.
The useful sales action is to ask which concern matters to the customer: appearance, reflections, handling or budget. AI can help phrase the question. It should not decide that the customer can afford the most expensive option based on their writing style or address.
Test the comparison by asking a colleague to explain the differences without seeing the original catalogue. If they cannot tell what the extra $35 buys, revise the wording. Clarity is the result to measure; an increase in the average order value is not automatically evidence that the advice improved.
6. Match a nursery follow-up to the parent's unanswered question
A parent who has requested visit information may still be deciding whether the daily arrangements fit their needs. AI can help staff draft a relevant follow-up from that question and the nursery's approved information. A fixed sequence of booking nudges is less useful when each parent needs something different.
Illustrative example: the parent asks how settling-in visits work. A draft says, 'You asked about the settling-in process. Our team can explain the current arrangements during your visit and discuss your questions. Would you like us to check a suitable visit time?' Staff include the actual policy information where available.
Remove any guarantee that a child will settle within a certain period. Do not let the assistant infer developmental needs or use private details to make the message more persuasive. The point is to answer the parent's question and invite an appropriate conversation.
Before sending, check whether the parent has already booked, declined or requested no further contact. Stop the sales follow-up when any of those events makes it inappropriate. A booking made by telephone must update the same record used to control messages.
Measure replies that lead to useful conversations and messages correctly stopped. Read the replies themselves: 'Please stop asking' is not a positive engagement result. The lead-nurturing tutorial shows how to build a journey around customer needs and clear exits.
7. Turn a clinic enquiry call into administrative actions
A physiotherapy clinic can use AI to organise administrative notes after an enquiry call. Keep this sales task separate from clinical assessment and treatment recommendations. The output should say what reception needs to do, who owns it and which details still need confirmation.
Use staff-written notes for an initial trial. Recording calls introduces additional permission, privacy and operational questions; use an approved process and appropriate advice before doing that. You can test action extraction without recording a single conversation.
Illustrative input: 'Caller wants appointment information. Can attend after 17:00. Asked whether a referral is needed. Reception to check the relevant booking requirements and call tomorrow afternoon. No appointment booked.'
An illustrative output is: 'Task: confirm booking requirements and later appointment availability. Owner: reception. Due: tomorrow afternoon, date to be confirmed. Status: enquiry only.' A bad output changes the last line to 'initial assessment booked'. Correct that before it enters the diary or triggers a reminder.
Do not store a guessed medical interpretation in the sales record. Where a question needs clinical judgement, route it to the appropriate clinician. Check each extracted task against the notes, and verify that it appears in the actual work queue. An accurate action hidden in an unused summary does not help reception or the caller.
8. Find language-course enquiries with no next action
A small pipeline review can reveal enquiries that have simply been forgotten. AI can help inspect a limited export or approved record set for missing owners, contradictory stages and absent next actions. Begin with a review list, then let staff decide what to change.
Illustrative example: a language school has 32 active enquiries. Twelve have a future appointment, eight have an agreed later contact date, seven await a customer reply within the school's chosen window, and five have no next action. The useful output identifies those five records and explains the missing field.
Do not instruct the assistant to invent a follow-up date for every blank. One of the five may already have declined in an unrecorded phone call. Staff should check the evidence, close the record or agree an appropriate action before anything is sent.
An illustrative review note reads: 'Enquiry 024: stage says considering; last note says requested contact after the next timetable release; next-action field blank. Ask owner to confirm timetable date.' This is more useful than 'high-potential prospect: contact immediately'.
Run the first review on a copy or read-only view. Measure how many flagged records were genuinely missing an action and how long staff spent checking them. If most flags come from a field nobody uses, change the record design rather than generating the same noisy list every week.
9. Group the reasons a music teacher loses enquiries
A teacher may remember the most frustrating lost enquiry and overlook a repeated practical obstacle. AI can group short, approved notes into themes, provided it separates what customers actually said from the teacher's guesses. This is a way to organise evidence, not discover hidden motives.
Illustrative example: 20 closed enquiries contain these recorded reasons: six timetable mismatches, four unsuitable instruments or lesson types, three price objections, two travel difficulties and five unknowns. The assistant should preserve the unknown category. Silence is not evidence that the lesson price was too high.
Ask for the source note behind every assigned reason. If a customer says, 'Tuesday does not work and I cannot manage the fee this month', allow more than one theme or agree a primary-reason rule. Do not present overlapping categories as percentages that must total 100.
The teacher can then test a practical change, such as explaining available days earlier in the enquiry process. Cutting prices would be a poor first response if the most common recorded obstacle is timing. A small set of 20 records helps identify questions to investigate; it does not establish a stable market trend.
Keep the review close to the original words. A theme called 'low commitment customers' adds judgement without useful evidence. 'Requested days unavailable' points to a concrete issue the teacher can act on or explain more clearly.
10. Check a tutoring proposal for unsupported promises
AI can provide a second reading of a proposal before a person approves it. Give it the approved source material and ask it to identify claims that lack support, contradictions and missing conditions. This is a checking role, not permission to silently rewrite the offer.
Illustrative example: a tutoring proposal says 'weekly sessions with the same tutor throughout the year'. The approved offer promises a named tutor for the initial six-session block only. A useful output quotes the relevant proposal sentence, identifies the narrower source commitment and suggests asking the owner before sending.
A second mismatch concerns a price: the proposal states $270, while the approved calculation is six sessions at $50, totalling $300. The assistant can flag the disagreement, but the reviewer should use a calculator or checked spreadsheet to establish the correct figure and confirm whether a discount was authorised.
Test the checking prompt with one deliberately inserted error in a copy. Then include a correct proposal to see whether the assistant invents problems. A checker that challenges every ordinary sentence wastes time and encourages staff to ignore its warnings.
Keep final approval with someone who understands the offer. AI reviewing AI-written text is still fallible. For numerical checking, use a source-based check of proposal figures and require every price, quantity and date to match approved evidence.
Pick a trial that your team can judge in one sitting
Choose a task with frequent repetition, a clear input and an output someone can check. Preparing a brief or drafting an answer usually fits this test better than giving a tool permission to negotiate discounts. Write a one-sentence acceptance rule before you compare products.
For example: 'The enquiry brief must preserve every stated measurement and flag every missing one.' Another useful rule is: 'The reply must answer the question without inventing availability.' These are easier to test than 'make our sales process better'.
Build a set of ten approved examples: six ordinary cases, two incomplete ones and two with conflicting information. Run them through the proposed process and record preparation, review and correction time. Count the entire job, including copying material between tools and entering the approved result into the customer record.
Use a filled test record so reviewers agree on what a pass means. For the framing brief, write: 'Input says roughly 30 by 40. Expected result preserves roughly and flags missing units. Actual result says exactly 30 by 40 centimetres. Decision: fail; remove the invented precision and rerun.' This identifies the error more clearly than a rating of three stars.
Also test what happens when the usual reviewer is absent. Give the draft and original input to the person covering the inbox, without explaining the intended answer. If they cannot identify which facts are approved or where to record a correction, improve the handover instructions before expanding the trial. A process that works only because its creator remembers every exception will be difficult for the rest of the team to maintain.
Choose an owner and a stop condition. If a draft invents a price, promise or confirmed booking, pause customer-facing use until that failure has been understood and retested. A wrong field in a private draft is an opportunity to improve the task; the same error sent to a customer becomes a business problem.
After the trial, keep the smallest version that helps. You may find that an approved reply library solves most questions and AI is useful only for unusual wording. Or summaries may save time while automatic follow-up adds unnecessary complexity. The worthwhile result is a sales process staff can run accurately, with more attention available for the conversations customers actually need.
Further reads
- AI Lead Generation vs Hiring a Salesperson: A Cost Comparison — Compare software costs with the work a salesperson actually does.
- What Is an AI CRM and Which Features Actually Save Time? — Choose CRM features by the administration they remove.
- AI Proposal Mistakes That Cost You the Job — Catch promises and omissions before a proposal reaches a customer.
- Signs Your Business Is Ready to Automate Sales Follow-Ups With AI — Check whether your process is ready for automatic follow-up.
- How to Turn Sales Call Recordings Into Coaching Notes With AI — Turn agreed sales-call evidence into useful coaching notes.
- How to Find B2B Prospects With AI Research Tools — A seven-step method for turning your best customers into a checked list of look-alike companies, named buyers and timely reasons to get in touch.
- How to Write Sales Call Scripts With AI That Don't Sound Scripted — Swap word-for-word scripts for an AI-built call map: openers, questions and objection cards drawn from your own best calls, then rehearsed aloud with AI.
- How to Write a Sales Deck With AI That Wins Meetings — Research one buyer, draft the storyline in plain text, turn every headline into a claim, and only then let an AI slide tool do the design.
- How Small Businesses Actually Use AI in Marketing: 15 Examples — Fifteen specific marketing jobs small businesses hand to AI, each with the steps, a sample, the time it takes and the catch to watch for.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Pipedrive Knowledge Base, AI email creation (updated September 2026); HubSpot Knowledge Base article on record summaries. Checked 28 September 2026. Business scenarios and operating figures are illustrative.