I Built a System to Grade Calls With AI. My Average? Yikes.

I run most sales for Rocket Media. I'm also on a lot of calls every day. Sales calls, client calls, vendor calls, internal meetings, the works.

And for years, my quality control system was a feeling. I'd hang up and think "that went well" or "that felt off," and that was the entire performance review. No data. No pattern. No way to know if I was actually getting better or just getting comfortable.

So I got curious: how effective are my calls, really? Where am I weak? Where am I strong? Could I grade calls with AI the way a coach with a clipboard would, except honest and available at 11pm?

Here's the thing I want you to catch before we go further: this is not a sales-only experiment. Everything below works for any call you'd like to evaluate with a genuine desire to learn how you're doing. Internal meetings. Vendor negotiations. Client check-ins. Honestly, even conversations in your personal life. If it can be transcribed, it can be measured.

You can't improve what you don't measure. It turns out measuring a conversation is now shockingly easy.

What actually goes into analyzing a call?

Three ingredients. That's it.

1. A transcript. You need the words. I use a Plaud recorder for in-person conversations and the built-in recording on Zoom for the rest, but Fireflies, Granola, or your meeting tool's native transcription all work fine. Two rules: always ask permission to record, and pick the tool you'll actually use every single time. Consistency beats quality.

2. A rubric. This is the magic ingredient, and it's the part people assume is hard. A rubric is just 5 to 10 categories of observable behaviors, each scored 1 to 10, with anchors describing what a 9 looks like versus a 5. Mine has ten categories covering the arc of a sales call: preparation, opening, discovery and listening, storytelling with the numbers, expertise, differentiation, engagement, objections, connecting value to price, and the close.

3. A scorer with no ego. AI applies the rubric the same way every time. It doesn't care that you were tired, that the prospect was difficult, or that you really felt like it went well. And it enforces the one rule that makes this whole system work:
   
   Every score must cite a direct quote from the call. "Good opening" is banned. "Asked recording permission, walked the agenda, confirmed the decision-maker was in the room" is a score. The first time AI quotes your own words back as evidence that you skated past something important, you will never un-see it.

Can AI build the rubric for you? Yes. I checked.

You don't need to be an expert in what to measure. Flip the workflow: have AI interview you. What kind of call is this? What does success look like? What do you suspect you're bad at? What do the best people you've seen do on this kind of call? Answer those questions and AI will draft your categories, your observable behaviors, and your scoring anchors in one sitting.

I put the exact prompt for this in the playbook at the end, along with the scoring prompt and starter categories for internal meetings, vendor calls, client calls, and personal conversations.

Seven weeks of report cards: what the data said about me

I've scored 7 diagnostic sales calls since May 20. My average is a 71.3. My range runs from a 50 (yes, a D, we'll get there) to an 89 that closed two days later.

Here's what measurement showed me that feeling never did.

I'm not as good a listener as I thought. On 4 of my 7 calls, the scorecard caught me skating right past a prospect bombshell. One prospect told me they were shifting their entire sales model from outbound to inbound. That's a strategic earthquake, and the transcript shows me acknowledging it and moving to the next slide. Another mentioned mid-call that they were actively redoing their website, which is literally a core service we offer. I answered by offering to send them an article. I read those scorecards with my head in my hands.

I keep opening a loop I don't close. On 5 of 7 calls, the same flag: pricing was never tied back to the revenue payoff. I'd spend forty minutes building the math for a $2M growth opportunity, then present the fee and just... stop. Never connected the two. It's about as fundamental as sales conversations get, and I couldn't see it until it showed up in writing five times in a row.

My strength is real, and now I know to lean on it. My expertise and credibility category averaged over 9 out of 10 across all seven calls. Measurement isn't only about finding weakness. Knowing what's working tells you what to build the call around.

The part that made me a believer. In late June, a call with a prospect scored a 50. The scorecard flagged exactly why: I mined past a revenue-decline disclosure and left the next step open-ended. Five days later I was back with the same company for the follow-up presentation, and I deliberately fixed those two things and nothing else. That call scored an 81.

That's the loop. Score, fix one thing, score again. My grades are trending up, and honestly, so is my confidence, because it's not based on a feeling anymore. I'm becoming a better student of my own process.

How to set this up yourself, this week

1. Capture. Pick your recording tool. Get permission. Every call.

2. Build your rubric. Run the Rubric Builder prompt from the playbook. Tune what AI gives you.

3. Score one call. Paste rubric plus transcript into the scoring prompt. Brace yourself.

4. Log it. A simple spreadsheet works: date, score, category scores, top 3 fixes. The patterns live across calls, not inside any single one.

5. Fix ONE thing. Then score the next call. One deliberate fix per call beats ten forgotten resolutions.

What this doesn't do (honest results, as always)

AI can't see the room. It doesn't know the relationship history, the politics, or that the prospect was distracted because their warehouse flooded. It reads words.

Right now AI is also my only grader. We're layering in human peer scoring next, and I still owe myself a consistency test: score the same call twice on different days and see if the grades match. I'll report back.

And there's a quirk worth naming: once you know you'll be graded, you behave differently on the call. So far that's been entirely to my benefit, since "behave differently" has meant "actually book the next meeting before hanging up." But a rubric shapes behavior, so build one that rewards what you actually care about.

The takeaway

I spent years assuming call skills were something you developed through osmosis and gray hair. Seven weeks of report cards taught me more about how I communicate than the previous several years of "that went well."

You can't improve what you don't measure. And the cost of measuring just dropped to a transcript, a prompt, and the willingness to look.

Try it right now: the Rubric Builder prompt

You don't even need the download to start. Paste this into Claude or ChatGPT and it will interview you, then build your custom scoring rubric in one conversation:

You are an expert performance coach. I want to build a scoring rubric to
evaluate my own performance on calls, and I want you to build it by
interviewing me.

Ask me these questions one at a time, waiting for my answer before the next:
1. What type of call is this rubric for? (sales call, internal meeting,
   vendor negotiation, client check-in, difficult personal conversation)
2. What does a successful version of this call look like? What outcome am
   I hoping for?
3. What do I suspect I'm good at on these calls?
4. What do I suspect I'm weak at, or what feedback have I gotten before?
5. What do the best people I've seen do on this kind of call?

Then build me a rubric with 5-8 categories. For each category:
- A name and one-sentence description of what it measures
- 4-6 observable behaviors (things visible in a transcript, no mind-reading)
- Scoring anchors: what a 9-10 looks like, what a 5-6 looks like, what a
  1-2 looks like

Each category is scored 1-10. Also give me a total-score-to-letter-grade
conversion. Format the final rubric so I can save it and paste it into
future scoring sessions.

Once you have your rubric, you need the other two prompts: the Call Scorer that grades your transcripts with quoted evidence, and the Trend Review that finds the mistakes you repeat across calls. Both are in the free playbook, along with the full setup roadmap and starter rubrics for five call types, including personal conversations.

→ Grab the AI Call Scorecard Playbook