It is 8:30 am. You have six shops to visit, two proposals to revise and yesterday’s notes to file. Start using AI on one of those jobs. A clearer visit brief or a follow-up ready to review can already make the day easier.
This guide is for sales reps visiting shops, pharmacies and specialist retail networks. The examples are fictional. Adapt the prompts to an assistant your company has approved; you do not need a particular vendor to test the method.
Start with a frequent task you can check
A good first use produces something you can verify. If you know a customer’s last three orders, you can check a summary of them. If you do not know their revenue or strategy, asking a model to guess will not prepare you for the meeting.
| When | Task for AI | What the rep checks |
|---|---|---|
| Before a visit | Summarise the supplied record and prepare three questions | Dates, sources and the customer’s last decision |
| Before a presentation | Role-play an objection using approved product information | No invented product claims or discounts |
| After the visit | Structure a dictated note | Products, quantities, agreements and deadlines |
| When following up | Draft a message about a specific commitment | Recipient, attachment and next action |
| At the weekly review | Group the reasons deals are stalled | Cited examples and cases that contradict the summary |
Keep negotiation, order approval and delivery commitments in their usual process. A well-written sentence is not evidence that a product is available or a commercial term has been accepted.
Prepare a visit from dated documents
Gather a small extract: the last visit report, relevant orders, the current proposal and an up-to-date product sheet. Date each document. Exclude personal or commercial information you are not authorised to give the tool. An outlet identifier is often enough for an initial trial.
For example, orders fell between June and August, but the shop closed for three weeks. The useful question is “How much stock is left for reopening?” Writing “customer in decline” in the CRM would be premature. Keep both the explanation and the scope of the figures in the summary.
Open the sources before using the brief
A real citation may still fail to support the sentence beside it. France’s data protection authority, the CNIL, explains that generative models can produce plausible but inaccurate answers (French). A short brief with two explicit unknowns is more useful than a confident page of untraceable figures.
Practise an objection without inventing an argument
Supply an approved product sheet and a real objection: limited shelf space, slow turnover or an already busy store team. Ask the assistant to play the buyer, one question at a time, then identify any answer you give that goes beyond the supplied information.
In a pharmacy, sales practice must not turn into unrestricted generation of health claims. Work from the documentation approved for the product and its regulatory status. For a shelf-space objection, discuss dimensions, the number of SKUs, opening stock and staff training. Those details can be checked.
Run a four-week pilot
- →Week 1: time your usual preparation and data entry on a small sample of visits. Record the corrections needed.
- →Week 2: test the pre-visit brief on comparable accounts. Keep the prompt and source documents.
- →Week 3: add structured notes, reviewed before they are saved in the CRM.
- →Week 4: compare total time, errors and actions completed. Keep useful steps and revise the rest.
Include an experienced rep and someone less familiar with the accounts. Their difficulties may differ. In Generative AI at Work, revised in 2024, Brynjolfsson, Li and Raymond study 5,172 customer support agents and report roughly 15% more issues resolved per hour on average, with different effects by experience. That finding concerns customer support; it does not measure a rise in pharmacy sales.
- Week 1
Measure normal preparation and data entry.
- Week 2
Test visit briefs on comparable accounts.
- Week 3
Add structured notes, then review them.
- Week 4
Compare total time, errors and completed actions.
Include review and correction time in every measurement.
Calculate net time saved, including review
Worked example: for 12 visits a week, preparation falls from 15 to 9 minutes and reporting from 8 to 4 minutes, including checks. With both uses active, the weekly saving is 12 × (6 + 4) = 120 minutes. In this staged pilot, week one establishes the baseline; week two saves 12 × 6 = 72 minutes on preparation; weeks three and four each save 120 minutes. Total: 312 minutes. Subtract three hours of setup and training, and the first month saves 132 minutes, or 2 hours 12 minutes.
If the nine minutes exclude source verification, the calculation is incomplete. Include time spent fixing a bad record or a message sent too quickly. Measure commercial outcomes separately: appointments secured, confirmed agreements and repeat orders. Saved time does not automatically become revenue.
Agree on straightforward usage rules
Before using real account records, clarify permitted data, access, retention and the supplier’s reuse terms with your company. The CNIL’s deployment FAQ (French) stresses defined uses and training. An entirely fictional account is enough for the first test; exporting your whole customer base is unnecessary.
Start with a real day in the field
Bring your visit template, an example follow-up and your data-entry constraints. We can examine the sales workflow and the steps worth simplifying.
Choose the next task: prioritise visits and plan a route, turn a voice note into a CRM report or write a follow-up after an appointment. Each trial should produce something the rep can review, correct and use that day.
By Pierre Lerebours
By Anatole Briant
By Pierre Lerebours
CEO of Salesia
Anatole Briant is CEO of Salesia. He is your contact for demonstrations, pricing and partnerships.