Every CSM has done this: a champion forwards your update to their CEO, and what they send is your raw call notes โ forty bullets, three acronyms nobody outside your org uses, and a line that says "need to confirm w/ Priya." The CEO glances at it, decides it needs translation, and never opens it. Your quarter's best work, buried because nobody gave it a two-minute wrapper.
The executive summary isn't a nicety. It's the difference between your work being seen by the person who signs the renewal and it dying in a Slack thread. The good news: it's a compression problem, and compression is exactly what AI is good at. The bad news: most AI summaries read like they were written by a machine, because they were written from a prompt that asked for one. Fix the prompt and you fix the output.
What an executive summary actually has to do
An exec summary has one job: let someone who was never in the room make a correct decision in two minutes. That means it needs exactly five things, in this order:
1. The headline. One sentence: what happened and why it matters. "We renewed Acme at $240K, up 20%, with the new analytics module included." If the exec reads nothing else, this carries the news.
2. The decision or ask. If something needs approval โ a discount, a timeline change, a resource โ state it in the second paragraph. Execs scan for what they owe you; make it findable.
3. The three facts that matter. Not the ten facts that happened. Usage up 30%, two open P1s, champion leaving next month. The test: would a wrong decision get made without this fact? If not, it doesn't belong.
4. The risk. One line, plainly stated. "Risk: the economic buyer hasn't attended a call in six weeks." Execs trust summaries that name the bad news more than summaries that hide it.
5. Next step and owner. "Lisa to confirm implementation timeline by Friday." No owner, no next step โ it's a status report, not a summary.
That's it. No background section, no methodology, no "per our discussion." Every paragraph that doesn't serve these five is a paragraph that costs you the exec's attention.
The prompt that produces them
Here's the prompt I give CSMs. Paste it into whatever model your team uses, feed it the raw notes, and get something shippable:
You are a customer success manager writing for a customer executive (VP or C-level) who has 2 minutes and was not on the calls. Summarize the following call notes into an executive summary of 150โ200 words, structured exactly as follows:
HEADLINE โ one sentence: what happened and why it matters to their business.
ASK โ if the notes contain anything needing the executive's decision or approval, state it in one sentence. If nothing needs approval, write "No action needed."
KEY FACTS โ exactly 3 bullets: the facts most likely to affect a renewal or expansion decision.
RISK โ one sentence naming the biggest open risk. Never soften it; executives discount summaries that hide bad news.
NEXT STEP โ one sentence: the single most important next action, its owner, and its date.
Rules: plain business language, no internal acronyms or jargon (translate them). No hedging phrases like "it seems" or "potentially." No bullet longer than two lines. If the notes lack information for a section, mark it "Not in notes" rather than inventing it.
Call notes: [paste here]
The "Not in notes" clause is the most important line in the prompt. Hallucination in an exec summary isn't a quality problem โ it's a trust problem. One invented renewal date in front of a CEO and your AI initiative is done for the year. The prompt must be explicit that gaps get labeled, never filled.
What makes it sound like you, not a machine
Three habits separate summaries that get forwarded from summaries that get ignored:
Feed it voice examples. The first time you use the prompt, append two of your own best past summaries with a line that says "Match this tone." One-shot examples do more for voice-matching than any adjective in the prompt. After five or six runs, the outputs converge on your style and you can drop the examples.
Set a stakes variable. For high-stakes accounts, add one line to the prompt: "This account represents $X in ARR and is flagged [healthy/at-risk]. Weight facts accordingly." The summary for an at-risk $500K account should read differently from the summary for a healthy $50K account, and the model can't know that unless you tell it.
Always edit the risk line yourself. AI softens bad news even when you tell it not to. Read the risk sentence last, before you send, and ask: "Is this the bluntest true version?" If not, rewrite it. The whole point of the summary is that the exec trusts it โ and execs can smell laundered risk.
Building the weekly habit
The prompt is worthless if nobody uses it. The teams that actually get leverage from this build it into a Friday ritual:
Batch the notes. Don't summarize per call. On Friday afternoon, dump the week's call notes for each strategic account into the prompt โ five calls, one summary. Execs don't want five emails; they want one. Batching also gives the model enough material to pick real key facts instead of padding.
Send it to your champion first. Before it goes anywhere near an executive, your champion reads it. They correct facts, soften the risk line if needed, and โ critically โ they forward it themselves. A summary forwarded by the champion carries more weight than one sent by your team. It's their credibility lending to yours.
Track what gets read. If your tooling supports it, use tracked links or read receipts on the summary. Two months of data tells you which execs engage and which summaries die. The ones that die tell you where your fact-picking is wrong. This is the feedback loop that turns a good prompt into a team asset.
Where this goes wrong
The fourteen-fact summary. CSMs hate cutting. They'll take the three-fact output and add "just two more things." Now it's seven bullets and the exec stops reading at four. Enforce the three-fact rule like it's a law of physics.
Summary as theater. Some teams start writing exec summaries because "the playbook says so" and nobody reads them. If your champion isn't forwarding them, the format isn't failing โ the distribution is. Ask your champion directly: "Would a version of this be useful for your boss?" Build for the actual reader.
Feeding it garbage notes. The prompt compresses; it can't create. If your call notes are three words ("good call, no issues"), the summary will be honest fluff. This is a feature, not a bug โ it exposes which calls aren't producing information, which is itself worth knowing.
The math
A CSM with eight strategic accounts writing weekly summaries by hand: roughly 2 hours a week, most of it staring at notes trying to find the narrative. With the prompt: 20 minutes of paste-and-edit. That's 90 minutes back per CSM per week, but the real return isn't the time โ it's that execs who read your summaries start showing up to QBRs prepared, and renewals get easier when the decision-maker has been reading a two-minute story of your value all year.
Compression is leverage. This is one of the highest-ROI prompts in a CSM's toolkit because it converts invisible work โ the calls, the follow-ups, the saved escalations โ into visible narrative, at the exact altitude where renewals get signed.
Tomorrow: Calendar Tetris is a choice โ time-blocking systems that protect a CSM's deep work without dropping the ball on customers.
