Read Time: 5 Minutes
Your reps spend an hour before every call digging through call recordings, Salesforce, and old emails.
So you buy an AI tool.
And it gives them confident answers... that are wrong.
Short answer: Let AI read your data long before you let it write to your CRM. Feed it your definitions, your metrics, your call recordings, and your deal data first, and use it for call prep and deal questions. Only let it write to CRM fields after you test it on 10 past calls and check every field by hand.
This one comes from my podcast with Sarabeth Scott. She ran RevOps at B2B SaaS companies for 10 years, most recently at Camunda. Now she builds the RevOps function at Camber Partners, a growth equity firm, and works with portfolio companies from $4M to $35M in revenue.
It's one layer that pulls your call recordings, CRM deals, and marketing data into one place. Your team asks it questions in plain English. It tells them deal status, churn risk, and where to expand. Nobody has to pull a report. The AI isn't the smart part. The context you give it is.
One Camber portfolio company built one. In Sarabeth's words, they're "creating what we call like a second brain. And it is that context layer, right?"
The result = faster meeting prep, faster follow-up, and a clear view of every deal. All of it comes from data that used to sit in different tools.
Her favorite part: "so that people can get what they need, they can get their customer 360 without even having to go into Salesforce."
She built one for herself too. "I built like an intern, if you will." It knows every stakeholder and how she works with each one. So she never has to explain things twice.
AI without context is just a fast guesser.
Not until you test it. Letting AI read your data is low risk. Letting it write to your CRM is not. If it pushes call notes into the wrong fields, your pipeline report is now fiction. And nobody can tell which numbers a rep typed and which ones the AI guessed. Start with read only.
Reading is safe. Writing is where it breaks.
Sarabeth's warning: "So before you start writing anything into a CRM specifically, I'd be very careful about how you set that process up, leveraging AI."
She's not against it. She thinks it's a big win once it works. Her example was pushing call notes from Gong into the CRM. You can do it. You just have to tune it first: "You're pushing the right pieces of information into the right places in your CRM."
Her short version: "I think it's all about training agents".
Your definitions. If your team can't agree on what an SQL is, your AI can't either. Write down what counts as an MQL, SQL, and SQO. Then list every key metric, how you get it, and which CRM fields feed it. Give that to your AI before anything else. Then add calls and deals.
Sarabeth says the best thing you can do is "to clearly define what is a marketing qualified lead, what is a sales qualified lead, what is a sales qualified opportunity."
She keeps it simple: "we have a glossary, we have a metric index". At Camber, she goes one step more. She maps "these are the CRM fields that map to these metrics".
Here's the order I'd use. Read only, for now:
Safe first jobs for it:
Run it on 10 past calls before you let it touch live deals. Check every field it fills in by hand. Fix the mapping where it got it wrong. Then run it again. Keep going until it gets them right. Only then turn on writing, and start with low-risk fields, not deal amounts.
Sarabeth's process: "And you can test and test and test and refine and tune so that you're making sure that it's structuring those notes in the right way."
Don't let it write these until it's tested:
The payoff is real. She says "you can automate like call prep for reps research on the company." The rep input part comes next. But "it has to be in a reliable way."
Do this today (20 minutes):
Less than you think. Pick the core numbers that run the business, like sales cycle length and deal size. Get clear on who your best buyer is. Don't build heavy process too early. A small, clean set of data with clear names gives AI more to work with than a big messy one.
Sarabeth: "Less is more. And that is always my point of view, especially in your CRM. Like you don't need to track everything."
She also warned about your first few wins. Don't assume they tell you who your best buyer is. In her words, "don't mistake like early wins".
And clear names give you room to try new things: "When you're not questioning definitions or data, it gives you more freedom to to try things out".
More on this: How do I know if my Salesforce org is Agentforce-ready?, Gong + Salesforce Integration: Playbook for Clean Data, Lower Risk, and Real Adoption, and Under $10M? Stop optimizing. Seriously.
Most teams think they have an AI problem. They have a context problem.
Write down your definitions. Let AI read first. Test on 10 past calls before it writes a single field. Fix that, and the AI part gets easy.
It's one place where AI can see your call recordings, CRM deals, marketing data, and definitions all at once. Your team asks it questions in plain English, like where a deal stands or which customers might churn. It saves reps from digging through 3 tools before every call.
Yes, but test it first. Run it on 10 past calls, check every field by hand, fix the mapping, and repeat. Until it passes, keep it away from deal amounts, stages, close dates, and anything that feeds a board number.
What counts as an MQL, SQL, and SQO. Then every key metric, how you get it, and which CRM fields feed it. One page is enough to start. Give it to your team and your AI at the same time.
No. Start with 1 open deal and 1 folder. Put the call transcripts, notes, CRM export, and your definitions doc in it. Ask the AI about the deal. If the answer is weak, you just found what your data is missing.
Ask it about a live deal, then ask the rep. If the two answers don't match, the gap is the context your AI doesn't have yet. Add that to the folder and ask again.
Want more of this? The Simply Scale newsletter drops every week: Subscribe here
Need help with Salesforce? Book a call: https://calendly.com/jordan-gosimplyscale/30min?utm_source=blog&utm_campaign=use-ai-on-crm-without-breaking-data
Let's talk about how we can build a Salesforce system that grows with your business.
