Case study · AI build · 2026

How I compressed a four-hour QBR into 90 seconds. By building it myself.

I was carrying enterprise coverage for a 50+ account ANZ book, with most customers contracted for two to three Quarterly Business Reviews per quarter. The math didn't work. I built a QBR automation skill, from inside the CSM seat, that generates a first-draft QBR in about 90 seconds. It saves me four to five hours every week. More importantly, it lets me spend those hours on the judgement work the customer actually pays for.

9-slide narrative Problem → Approach → Outcome Built on OpenCode + frontier LLMs

  1. 01 Problem · The math wasn't working

    Minimal CSM coverage. 50+ enterprise accounts. Two to three QBRs per quarter each.

    I was carrying enterprise coverage for a mid-size book of ANZ customers. Most accounts were contracted for two to three Quarterly Business Reviews a quarter. Do the math: 50+ accounts times 2-3 QBRs times four hours of prep each equals a job nobody could do well. Something had to give. The traditional answer is to lower the bar or skip the prep. Neither of those is a CSM I want to be.

    50+ enterprise accounts · 2-3 QBRs per quarter each · ~4 hours of prep per QBR · one CSM

  2. 02 Problem · What was actually broken

    QBR prep was a recurring four-hour artifact that nobody had four hours for.

    The QBR isn't optional. It's the moment where the customer sees what they paid for, where their leadership decides whether to keep paying, where the relationship gets reset for the next cycle. Skipping it is malpractice. Phoning it in is worse. The reality was that doing it well, at the cadence the contract specified, simply did not fit in the working week. I wasn't going to keep losing weekends to slide assembly.

    Four hours per deck · pull data · build charts · assemble slides · QA · all manual

  3. 03 Approach · The frame

    Build it from the inside of the role. Not as an engineer who happens to be a CSM.

    Look, here's the thing most automation tooling misses. When an engineer builds for a CSM, they build what they think the CSM needs. When a CSM builds for themselves, they build what the work actually needs. Different artifact. I built this as a CSM who happens to know how to build. That perspective drove every decision about what goes in, what stays out, and what the output actually has to do for the customer in the room.

    CSM-as-engineer · build for the work, not for the abstraction · fit for function

  4. 04 Approach · The context the automation needed

    Five context streams. Pulled in, synthesised, and shaped into a QBR.

    I worked out what a great QBR actually grounds itself in, then made sure the automation could see all of it. Previous customer conversations and meeting context. Current and recent platform changes from the vendor side. New products and capabilities that might fit the customer's roadmap. Account team input from the AE, SE, and CSE. Customer-side stakeholder context and prior business priorities. Five streams. The automation pulls from each, synthesises against the customer's own past QBRs, and produces a draft.

    Previous conversations · platform changes · new products · account team input · customer priorities

  5. 05 Approach · What the output covers

    Eight QBR sections. Generated as a first draft in about 90 seconds.

    The draft covers business priorities, what worked and what didn't, scheduling, engagement signals, new products, product changes, training adoption, and utilisation. A little bit of everything a QBR is supposed to land. It's a first draft, not a final deliverable. The CSM (me) reads it, fixes what's wrong, sharpens what's vague, and ships. The 90-second number is the time-to-first-draft. Total time-to-customer-ready is closer to 30-45 minutes, which is still an order-of-magnitude compression on the manual baseline.

    8 sections · 90s to first draft · 30-45 min to customer-ready · ~5-8x compression

  6. 06 Approach · Stack

    OpenCode plus a rotation of frontier LLMs. Claude Opus, Gemini Pro, the rest.

    Built on OpenCode as the agentic harness. Underlying models rotate across Claude Opus, Gemini Pro, and the other frontier LLMs OpenCode routes to. Data inputs from the company's CRM (the kind of system anyone running customer success has), meeting transcript history, and internal platform documentation. Chart rendering through a separate skill that pipes HTML/CSS templates through headless Chrome to produce PNG slide assets. The stack matters less than the principle: assemble the cheapest reasonable tools, build what's missing, ship.

    OpenCode · Claude Opus · Gemini Pro · headless Chrome · HTML/CSS templates · CRM as data layer

  7. 07 Outcome · The compression

    Four to five hours a week back. Thousands of dollars a year in time recovered.

    The headline number: four to five hours a week of QBR prep recovered. At fully-loaded CSM cost, that's thousands of dollars a year per CSM, every year, forever. Multiply across a team and the savings compound fast. But the dollar figure isn't the most important outcome. The more important one is below.

    4-5 hours/week recovered · thousands of dollars annually per CSM · compounding across team

  8. 08 Outcome · What the time bought

    The customer got a sharper QBR. Not just a faster one.

    When the four hours of slide assembly went away, the time didn't disappear. It moved up the value chain. The CSM (me) spent it on the actual judgement work: which insights matter to this specific stakeholder, what the customer's real next-quarter ask should be, how to frame the recommendations so they land. The QBR got better because the prep got smarter, not because the prep got skipped. That's the move. The customer gets more, not less.

    Time reallocated · prep got smarter, not skipped · customer outcomes improved

  9. 09 Takeaways · What another CSM or builder could use

    Build through the lens of the role. That's the part nobody else can do for you.

    Look, here's the lesson. Through effective planning and through the lens of a CSM specifically operating as an engineer, you can build tools that automate high-impact artifacts like QBRs. That compresses real hours every week and saves thousands of dollars a year, if you're committed to the craft and you take your time to build things fit for function. Oftentimes that's not what you get from a general engineer building for someone else. This is what happens when a CSM is able to build from the engineering perspective, and really deliver something fit for function.

    Three lessons · Build from inside the role · Compose cheap frontier tools · Commit to the craft