Document it before you automate it
Every sales leader I talk to has the same mandate right now: get AI into the sales process. Usually by next quarter.
So they buy an AI SDR or a research agent.
Six weeks later, reps have gone back to their own habits, and someone asks whether the tool was a waste of money.
The tool usually isn’t the problem. The process it was supposed to automate was never written down.
Your process lives in people’s heads
Ask five reps how they work a new inbound lead. You’ll get five answers.
One checks LinkedIn first. Another skips research entirely if the title looks junior.
Your top performer does something nobody else does, and she couldn’t explain it if you asked.
That works while humans do the work, because humans fill in what’s missing with judgment.
Software can’t. Point an AI at five versions of a process and you get a sixth that matches none of them.
What “documented” actually means
A documented process isn’t a slide with boxes and arrows. It answers these questions for every step.
- What triggers this step? A lead assigned, a reply received, a meeting no-show.
- What decision gets made here, and on what criteria? “Good fit” isn’t a criterion. “200+ employees, uses a contact center platform, visited pricing twice” is.
- What happens on each branch? Every yes needs a next step. So does every no.
When you can answer those three questions, you have a decision tree. A decision tree is something a machine can follow.
Then draw the human line
Once the steps are on paper, each one gets one of three verdicts:
| Verdict | What it means | Example |
|---|---|---|
| Automate | Software does it end to end. No review. | Enrichment, routing, logging activity |
| Assist | Software drafts, a human approves. | First-touch emails, call prep briefs |
| Human | A person owns it. Software stays out. | Discovery calls, pricing exceptions |
The verdict isn’t about what AI can do. It’s about what a mistake costs. A bad enrichment field costs nothing. A tone-deaf email to a Fortune 500 VP costs the account.
Why this order matters
Skip the documentation and you automate a mess faster.
Do it first, and tool demos turn into a checklist. You know exactly which steps you’re buying software for.
Adoption goes up too. Reps trust a system when they can see the rules it follows, especially rules they helped write.
Being AI-native isn’t about how many AI tools you own. It’s about having a process clear enough that a machine can run the parts it should.
Where to start
Pick one workflow that eats your team’s week. Inbound lead handling is usually the best candidate because it’s frequent, measurable, and close to pipeline.
Sit with two reps and write down every step and every exception.
Then draw the tree, and mark which steps software should handle.
That one page will tell you more about where AI belongs in your sales process than any vendor demo.
If you want help drawing it, that’s what the first call is for.