The AI Sales Checklist: 30 Checkboxes That Turn Dabbling Into an Operating System
Every sales team is “using AI” now, which mostly means reps paste things into a chatbot when they remember to. The difference between that and an AI-operating sales team is not a better tool, it is a system: the right things automated, in the right order, with humans where judgment lives. This is that system as a working checklist, six sections and thirty items, each with why it earns a place and how to actually do it, ordered so each one makes the next one work.
Read it as a sequence, not a menu. The order matters: data before drafting, prompts before automation, skills before scale. Skip ahead and the later items break.
TL;DR
- One workflow at a time. Human reviews every output for two weeks, you measure one number, then you scale or kill it.
- The stack order is load-bearing: connect data first, build a prompt library second, automate third, encode skills before you scale.
- AI owns reading and drafting; humans own judgment and every send.
- A workflow that does not move a metric in a month is a hobby. Cut it.
Section 1: Foundation, choose your stack deliberately
1. Choose a primary assistant and know why
Why it matters: Reps who bounce between three chatbots build no depth in any. A default assistant means shared prompts, shared skills, and compounding skill instead of scattered dabbling.
How to do it: Pick one primary (Claude for long-document and strategy work, ChatGPT for roleplay and speed) using our July 2026 comparison. Standardize the team on it; a second tool is fine once the first is mastered.
2. Connect live data before optimizing prompts
Why it matters: The gap between models is now smaller than the gap between a connected model and a disconnected one. A connected average model beats a brilliant disconnected one, because it works from your truth instead of its memory.
How to do it: Wire your CRM and a contact-data source to the assistant over MCP. This is item two for a reason: every later workflow assumes the AI can query real data, not guess.
3. Decide what data may not touch which tools
Why it matters: One careless paste of customer PII into the wrong tool is a compliance incident. Deciding the boundary once, up front, prevents the panic later.
How to do it: Write a one-paragraph policy: what customer data may go into which assistant, and what stays out. Match it to each vendor’s data-retention terms. Share it before you roll anything out.
4. Give each rep the paid tier, not five free logins
Why it matters: Free accounts hit context limits on long transcripts, lack connectors, and lose saved projects. The productivity difference pays for the seat in a week.
How to do it: Budget one paid plan per active rep. The limits, MCP connectors, and persistent projects are the difference between a toy and a tool.
Section 2: Prompt library, stop re-typing your five best prompts
5. Save standing prompts for your five most-repeated tasks
Why it matters: Reps re-write the same research and email prompts a dozen times a day, each slightly worse than the last. A saved prompt is consistent quality at zero marginal effort.
How to do it: Identify the five tasks you prompt for most (research, cold email, objections, call prep, follow-ups) and save a tested prompt for each where the whole team can reach it.
6. Upgrade your best prompts into skill files
Why it matters: A prompt is per-conversation; a skill file is permanent and consistent across every rep. It is how one rep’s best approach becomes the team’s default.
How to do it: Turn your top prompts into skill files that encode ICP, voice, and stage conventions. Templates are in our Claude skills for sales guide.
7. Keep a shared ‘what worked’ doc
Why it matters: Prompt folklore, the good prompt one rep discovered, dies when it lives in their DMs. Captured, it compounds across the team.
How to do it: Maintain one shared doc of prompts that produced real outcomes (a booked meeting, a saved deal), with the output attached as proof. Review it monthly.
Section 3: Prospecting and research
8. Run pre-call research as a workflow, not a habit
Why it matters: ‘Do your research’ is advice nobody follows at volume because it used to take twenty minutes per prospect. As a two-minute AI workflow, every meeting can get a brief.
How to do it: Point a research workflow at each meeting on tomorrow’s calendar; our prep checklist shows the exact blocks, and deep research automates the gathering.
9. Summarize the 10-K, earnings call, or press before executive outreach
Why it matters: Executives dismiss generic outreach instantly. One specific, current reference from their own filings proves you did the work and earns the reply.
How to do it: Before any senior outreach, feed the assistant the prospect’s latest filing or earnings transcript and ask for the three points relevant to what you sell.
10. Build lists by asking, not filtering
Why it matters: Plain-language search against a live database returns rows in seconds; manual filtering across tools burns an afternoon and still misses the trigger.
How to do it: Query a connected contact database in plain English (“heads of RevOps at US fintechs, 50-500 employees, verified emails”). Dedupe against the CRM before the list reaches a rep.
11. Verify every address before you send
Why it matters: An AI-personalized email to a dead address is polished waste that also dents your domain reputation. Verification is the floor under every other email metric.
How to do it: Verify at send time, not at import; contact data decays 25-30% a year, so a list verified last quarter is already stale. Our email diagnosis covers the deliverability layer.
Section 4: Calls
12. Record calls and extract the moments that matter
Why it matters: The insight in a 40-minute call, the objection, the next step, the competitor mention, is lost to memory by Friday. AI turns every call into a searchable, structured record.
How to do it: With consent, record calls and have AI pull key moments, agreed next steps, and competitor mentions into a summary you actually reread.
13. Roleplay the hard call before the real one
Why it matters: Practicing on a live prospect is expensive. An AI playing a skeptical CFO lets you fail in private until your answer stops sounding defensive.
How to do it: Have the assistant play your buyer and pressure-test your pitch twice before the meeting; the setup is in our Claude guide.
14. After every call, analyze what the buyer revealed vs what you assumed
Why it matters: This is the highest-leverage and most-skipped prompt in sales. The gap between what you heard and what was actually said is where deals quietly slip.
How to do it: Paste your notes or transcript and ask: what did this buyer reveal about criteria, timeline, and internal politics, and what did I assume they never actually said? Then, what should I ask next?
15. Score your own calls monthly against your methodology
Why it matters: Reps plateau without feedback, and managers cannot ride along on every call. AI scoring gives consistent, judgment-free feedback at scale.
How to do it: Run transcripts against your chosen methodology encoded as a skill file; the gaps become your coaching topics.
Section 5: Pipeline and CRM
16. Turn transcripts into proposed CRM updates with human approval
Why it matters: Reps spend roughly 28% of the week on data entry, and the CRM still lags reality. AI drafts the update; the human just approves the diff.
How to do it: Feed the transcript in, get a proposed field update out, approve it before it writes. The full guardrailed setup is in Claude + CRM workflows.
17. Run a weekly read-only pipeline audit
Why it matters: Deals die in the gaps between reviews, quietly, with a stale stage and no next step. A scheduled audit catches them while there is still time.
How to do it: Every Monday, have AI (read-only) flag stale, inconsistent, slipped, and momentum deals from the CRM. No write access needed, which makes it the safest workflow to start with.
18. Let AI draft follow-ups from CRM history; let humans send
Why it matters: The follow-up is where deals go generic, because writing a good one means re-reading the whole thread at 4:50pm. AI reads it instantly.
How to do it: Have the assistant draft follow-ups grounded in the engagement history, as drafts, never auto-sent. The human adds the human line and sends.
19. Run a monthly enrichment and dedupe sweep
Why it matters: Databases decay 25-30% a year; left alone, your segments lie and your bounce rate climbs. A job change, though, is a buying signal, not a correction.
How to do it: Monthly, re-enrich active contacts and propose merges (never auto-merge). Route job changes to the rep as opportunities rather than silently overwriting the record.
Section 6: Habits that make it stick
20. Block 15 minutes daily to practice one workflow
Why it matters: AI skill is built by reps, not bought by managers. Fifteen deliberate minutes a day beats a full-day training that is forgotten by Thursday.
How to do it: Put a daily 15-minute block on the calendar to practice one workflow until it is boring. Boring means it has become muscle memory.
21. Review every AI output before it goes out
Why it matters: AI gets you 80%; the last 20%, the detail only you know, is what proves a human did the work. Unreviewed output is how ‘AI emails are generic’ becomes true.
How to do it: Make it a rule: AI drafts, you decide. No AI output reaches a customer without a human edit that adds something the model could not have known.
22. Measure one number per workflow, and kill what does not move
Why it matters: AI that does not change a metric is a hobby dressed as productivity. A single tracked number tells you which workflows earn their place.
How to do it: Assign one metric per workflow (reply rate, prep time, next-step completion), baseline it for two weeks, and retire any workflow that has not moved it in a month.
23. Share wins in a channel; adoption spreads by example
Why it matters: Mandates create compliance; visible wins create demand. Reps adopt what they see a peer succeed with, not what a memo tells them to.
How to do it: Create a channel where reps post AI wins with the output attached. The best rep’s workflow becomes the team’s, without a rollout deck.
What this checklist cannot do
A checklist makes the judgment work systematic. It does not build the execution layer underneath, and no chat assistant does either: sequencing across email, calls, and LinkedIn tasks; deliverability and warm-up; reply detection that stops a sequence the moment a prospect answers; verification at send time. That is infrastructure, and it lives in a sales engagement platform built for it. The division of labor that works in 2026: AI for the thinking, the platform for the doing, one data layer under both. If you would rather have the whole loop run as a managed service, that is the job description of an AI SDR. And if your team still needs the foundational skills, start with the free AI courses for sales.
Frequently asked questions
Start with one workflow, not a tool rollout: pick the highest-friction task (usually pre-call research or post-call CRM updates), run it with AI for two weeks with a human reviewing every output, and measure one number. Teams that start with 'everyone gets a license' get scattered dabbling; teams that start with one workflow get a repeatable win they can copy to the next.
AI owns the reading and drafting: research, summarization, first drafts, data formatting, and pattern-finding across calls and pipeline. Humans own the judgment and the relationship: what to say to whom, the final edit that adds the detail only you know, every send decision, and everything on a live call. The failure mode in both directions is the same, automating judgment or manually grinding through reading.
Ground it in data the model cannot invent: connect live research and CRM context so drafts reference true, specific facts; encode your voice rules in a skill file so every rep's output follows them; and enforce the 20% rule, no draft goes out without a human adding the one detail that proves a person did the work.
AI drafting is fine; AI volume-blasting is where deliverability dies. The rules that matter are unchanged: verified addresses, warmed inboxes, per-inbox sending limits, real unsubscribe handling, and relevance. AI raises the ceiling on personalization quality but also lowers the floor on spam production; which one you get is a process choice, not a tool choice.
Pick one number per workflow and track it against a two-week baseline: reply rate for outreach, prep time per call for research, next-step completion for CRM hygiene. If a workflow does not move its number in a month, kill it, do not keep it out of guilt. AI that does not change a metric is a hobby, not a tool.
Prem is a B2B sales technology founder passionate about helping teams build better outbound systems. His writing explores AI-powered prospecting, hyper-personalization, cold email, deliverability, and the future of outbound sales.
