AI for sales reps: getting 5 hours back every week

Most sales reps I know spend half their week on the wrap-around work. The actual selling, the calls and the conversations, takes maybe 15 hours. The other 25 go to research, follow-ups, CRM hygiene, and explaining to your manager what happened on a call they weren't on.

That's where the time goes. And that's where AI actually pays for itself.

Here's the breakdown of where reps bleed hours, and what to swap in.

Pre-call research: 15 to 20 minutes per call

Take 8 calls a week. That's already 2+ hours just opening LinkedIn tabs and scrolling 10-K filings.

Apollo and Clay handle the data layer. Apollo gives you the contact, the title, the org chart. Clay enriches it with funding rounds, headcount changes, tech stack signals, recent press. You can pipe Clay outputs into a Google Sheet and let it run while you sleep.

For the human layer, paste the LinkedIn profile and the company website into Claude or Perplexity and ask for a brief. Perplexity is better when you want sourced links you can click into. Claude is better when you want it written like a peer.

You're a senior account executive prepping for a 30-minute discovery call.

Below is the LinkedIn profile of the prospect, their company's About page, 
and their last earnings call summary.

Give me:
1. Three things they probably care about right now (with one source line each)
2. One opener that references something specific, not generic
3. Two questions I could ask in the first 10 minutes that would surface pain
4. One thing NOT to say (red flags from their public posts)

Keep it tight. I read this in the elevator before the call.

[paste profile]
[paste About page]
[paste earnings summary]

That prompt turns 20 minutes of scrolling into 3 minutes of reading.

Follow-up emails: 10 minutes each, and they all sound the same

This is the easiest win. You already have the call notes. You already know what to say. You just don't want to type it again.

Lemlist and Instantly handle the sequencing and deliverability side. For the writing, paste your last five replied-to emails into Claude and tell it to match the voice. Few-shot prompting beats every "write me a follow-up email" prompt by a wide margin.

Here are 5 follow-up emails I've sent that got positive replies. 
Notice the structure: short, one specific reference to the call, 
one clear next step, no "circling back."

[paste 5 emails]

The call I just had was with Sarah, VP Ops at a 400-person logistics company. 
She mentioned their current tool breaks every time they onboard a new warehouse. 
She liked our integration story. Next step is a technical demo with her IT lead.

Write the follow-up. Subject line included. Under 100 words.

Send the draft to yourself first. Edit the one line that sounds off. Send.

Call notes and CRM updates: 15 minutes per call

This is solved. Use Fathom, Avoma, Gong, or tl;dv. They join the call, transcribe, summarize, pull action items, and most of them push directly into Salesloft, Outreach, or your CRM.

Pick one and actually configure the integration. The rep difference between "I have Gong" and "Gong auto-updates my Salesforce next-step field" is about four hours a week.

Manager updates and pipeline reviews

Same transcripts, different audience. Once you have the recording, you can ask Claude or ChatGPT to summarize for your VP in a different shape: deal stage, blockers, asks. Don't write that summary by hand on Friday afternoons.

Reading the room

Crystal scores personality from public profiles and tells you whether to lead with detail or with vision. It's not magic. It's a useful nudge before a first call with someone you've never met.

What NOT to outsource

AI shouldn't write the relationship. It should write the friction.

Don't let it draft the message you send after a prospect's company has a layoff. Don't let it write your champion's birthday note. Don't let it pick which deal to walk away from. Don't let it talk to a customer in real time on a call, even if a tool offers it.

The pattern: AI does the parts where the prospect doesn't care that a human did it. Research briefs, templated follow-ups, CRM data entry, internal summaries. The parts where being human is the point, you keep.

If a rep can't tell whether your email was written by a model, the email was probably forgettable anyway. The reps who win with this stack use it to free up time so they can be more human in the moments that matter, not less.

Next up

You now have the sales rep playbook. Pillar 2 guide #2 covers the same approach for marketers: how to go from a fuzzy idea to a shipped campaign in a single day, using the same kind of stack but a very different workflow.