When your CEO asks what AI is doing for revenue, they want data. As a senior revenue operator, your job is to find the AI usage that should move a specific KPI, then show that KPI in a report your CEO knows, cut by AI usage versus non-AI usage.
As an example, imagine you built a post-demo follow-up skill to improve pipeline velocity. What you would do is take the pipeline velocity report your CEO loves and split it: deals the skill touched, and deals it didn't. If the skill works, the two lines separate, and you can measure the lift.
Do that across skills, tools, and the customer journey, and you've got a business case that AI is moving the KPIs you're accountable for.
1. Start with the report your CEO already reads
In our experience, the best practice is to reuse the reports you already have rather than inventing new measures, so that your frame of reference for the business, and your CEO's, stays familiar. You're starting from a chart you all already know and love, and simply adding nuance about AI usage to it.
This is how you turn a feeling into a number. When your reps say the post-demo skill is working and you believe them, but the board needs quantitative results, cutting the report they already trust by whether the skill was used gives you the separation, and the separation is the lift. (If you've already made the budget pitch, this is where the number on the first slide comes from.)
2. Tag AI usage by account, contact, stage, and skill
To produce these analytics, you first need to capture and warehouse your AI usage data at the conversation, agent, and tool level. Once that data set is in hand, augment it. Profile each conversation and extract the account and contact that appear in it (or the account a tool was used against), then pull the stage from the CRM as of when the conversation ran, and tag the skill, tool, or agent that was used.
The result is one row per conversation, tagged with those four things. It's a simplified picture from a data architecture standpoint, and it's where usage data and CRM data first meet.
What not to do: export your AI conversations once, slap them back into Claude or ChatGPT, and ask it to do the extraction. That works for a quick sense check, and it isn't how you operate a business. You need repeatable analytics you can measure against month after month.
3. Log AI usage against accounts and contacts in your CRM
Once the data is organized in your AI usage pipeline and warehouse, you get to decide what to expose in your CRM. We recommend evidence of the conversation, not the transcript: that it happened about an account or contact, the stage they were in at the time, and the skills used.
The join key is the unique account or contact identifier already in the row-level data that drives your report.
The transcript itself stays out for two reasons: it's far too much data, and it's potentially problematic for the reps having those conversations. Keep a way to get to it if you need it, and treat it as privileged.
What you end up with is evidence of a skill run at a point in time in the pipeline. That's what lets you measure whether it improved pipeline velocity, deal size, or speed of response to a prospect.
If this feels unnatural, or you're wondering why you'd log AI this way at all, think about how we log calls and website revisits today. That wasn't always the case: HubSpot popularized logging website revisits and Gong popularized logging sales calls, and now we couldn't imagine living without either one. Logging AI conversations puts you at that same frontier.
4. Rework your report to cut by "AI-touched vs not"
Now take your existing report and add a column for whether AI touched the deal at that stage. A single line showing pipeline velocity over time for one account type becomes two lines, AI-touched and not.
What you're looking for is meaningful separation, with enough sample behind each line for statistical significance.
We all believe AI can be incredibly impactful when applied correctly. This is the analysis that puts a number on that belief. Avoid showing data that's sparse or nowhere near statistically significant.
Think of it as a cohort analysis: the deals AI touched against the deals it didn't, with the data coming from reps' usage in the field rather than a test you configured. That's why you want a big impact before you call the skill effective at this stage.
What not to do: reach for deeper semantic analysis here. That's a separate job. This is used versus not used, and what it did to the KPI.
5. Analyze your new report for what changes with AI
Run this playbook across your skills and you'll find some have meaningful impact and some don't. You'll also find AI usage moving KPIs you care about that you never knew about before. Either way, the results sort into three kinds:
- Good results: scale the skill, or push it to more reps.
- Middle-range results: economize the cost but keep the skill, because reps like it or it helps their efficiency, even though it isn't moving the needle on the KPI.
- Net negative: people spending lots of time in Claude or ChatGPT, the general LLM use reps tend to do, where a purpose-built tool should be used instead.
Playbook summary
| Layer | What to capture |
|---|---|
| Warehouse | Every AI conversation, agent run, and tool use, one row per conversation |
| Tags on each row | Account, contact, stage at the time of the run, and the skill, tool, or agent used |
| What goes in the CRM | Evidence that a conversation happened about the account or contact, the stage at the time, and the itemized skills. Not the transcript. |
| Join key | The unique account or contact identifier in the row-level data behind your report |
| The report | The pipeline velocity report (or deal size, or speed of response) you already run, one line split into AI-touched and not, with significance |
| The read | Scale it, economize it, or redirect the work to a purpose-built tool |
As a leader, what you're building is an analytic system. Month to month and quarter to quarter, it shows you how the very organic, individually driven AI usage in the field is affecting the KPIs you care about, and whether the skills and tools you deploy centrally are earning their keep.
We recognize this is a lot of work, and that's exactly why we built SightLift. This is the right way for a revenue leader to analyze AI usage against the KPIs that matter to them, and a tool like SightLift is the fastest way to get there.