AI in sales: where it saves time in a real week, and where it doesn't

Think about how a rep's week actually goes. The calls and meetings take a fraction of it. The rest goes on prep, write-ups, chasing and records. That second part is where AI helps most today.

• Before a call, ask an assistant to summarise a company's website, recent news and the person's role, so you walk in with two good questions.
• After a call, transcription tools turn a recorded call into text and a summary with the next steps pulled out. You check it and fix it in two minutes, not fifteen.
• For first drafts of the follow-up email, the proposal outline or the reply to "can you send me some info?", AI writes version one and you make it sound like you.
• Some tools move what was said on a call into the right fields of your CRM, so the deal stage and next step are current without typing.
• With your history in one place, an assistant can list who's gone quiet, who's overdue a follow-up and which deals haven't moved in a month.
• A chat agent can answer a new enquiry at 10pm, ask a couple of qualifying questions and book a time, then hand the chat to a person.
• Roleplay tools let a rep rehearse a price objection with an AI buyer before trying it on a real one.

Each of those takes a task a rep already does and makes it shorter. None of them replaces the conversation with the buyer.

AI is good at producing text that sounds right. That's also how it fails:

• It makes things up: an assistant asked about a company will sometimes state a fact that isn't true, in the same confident tone as the true ones. Check anything you'd repeat to a buyer.
• It flattens your voice: a follow-up that reads like everyone else's AI follow-up gets the same reply rate as everyone else's: none. Edit the draft until it sounds like something you'd say.
• It can't read the room: whether to push on price, when to stop talking and what the buyer's pause meant are still judgment calls.
• Tools that send hundreds of "personalised" emails a day make it easy to annoy a whole market at once.
• It needs clean records: an assistant that decides who to call works from the history you've logged, and with no calls logged it's guessing.

1. Time your week. For five working days, note roughly how long you spend on research, notes, follow-ups and updating records.
2. Pick the one job that ate the most hours. Only one.
3. Try it with the AI you already have. Most CRMs, email apps and phones now include some; a general assistant covers research and first drafts.
4. Run it for two weeks, then time the same job again.
5. Keep it if it saved time you used for selling. Drop it if you spent the time fixing its mistakes.

Then do the next job. A team that adds one tool a month and checks each one ends up with three tools that work. A team that buys five at once often ends up using none of them by the end of the quarter.

Before anyone on your team pastes a client's details into an AI tool:

• Read what the tool does with what you type: whether it's stored, and whether it's used to train its models. Use a business plan that says it isn't, where one exists.
• Keep out anything a client told you in confidence: health, money, family matters and their ID numbers.
• Read every message before it goes out. You're the one who presses send.
• Say when a buyer is talking to a bot. A chat agent that answers enquiries should say it's automated and pass the chat to a person when asked.
• Check your own country's privacy rules and your company's own rules, and follow the stricter one.

If your team is trying AI, ask two questions every month, not "are they using it?":

• Did calls and meetings per rep go up? Time saved on admin should show up as more selling.
• Did anything get worse? More follow-ups that read alike, or a rise in messages from annoyed buyers, means someone's sending too much too fast.

Measure it against your own team's last quarter. Vendors' ROI figures come from their own customers, who are rarely a team like yours.