Better to sacrifice a part to preserve the whole.
— The 36 Stratagems, Sacrifice the Plum Tree to Preserve the Peach Tree
Previously on this series: #10: Lena Watched a Team Adopt Her AI Template. Leo Didn't Know the Knife Was in the Contract. — Lena joined CoreStack as a consultant to help Leo build his reporting system. He thought she was there to help. Five weeks later, he realized the contract he signed wasn't just a template renewal — six months of data baselines were already locked inside the VeriTest framework.
Meanwhile, the CoreStack consulting engagement was still running, and VeriTest's internal AI evaluation platform had already gone live — but today, Lena was dealing with her own company's business.
Marcus Reed walked into the office and set the folder on Lena's desk. The cover read "VeriTest AI Evaluation Platform · Q3 Review."
"The board wants an answer," he said. "Three directors asked me separately."
Lena didn't reach for the folder. She already knew what was inside.
"I'll have a proposal by Tuesday."
Marcus paused at the door. He didn't turn around. The door closed a little softer than usual.
Lena looked at the folder cover. The AI Evaluation Platform — the project she'd staked her credibility on in front of the board. Last year, she'd told them: "The future of consulting is AI-powered assessment. We need to build this capability ourselves."
The board approved it.
Twelve months later, the platform was running. The demo looked great. VPs loved showing it off when clients visited. But Lena knew what the directors were really asking — not "Is this platform good?" but "How much did this thing make?"
She didn't have an answer.
She opened the folder.
The Numbers
Lena spent two days pulling data. During the day, she had Leo's QBR at CoreStack to get through. At night, she switched over to VeriTest's own financials.
She broke VeriTest's product lines apart and ran the numbers on each one.
| Line | Quarterly Revenue | Quarterly Cost (People + Compute) | Management Attention |
|---|---|---|---|
| AI Evaluation Platform | $22K (3 trial conversions) | $200K | ✅ Board-level, constant |
| Standard Technical DD | $310K | $28K | ✅ Aware |
| Strategic Consulting | $460K | $32K | ❓ Limited visibility |
| Delivery Methodology & Knowledge Base | (Supports all delivery lines) | $9K | None |
The last row made her stop writing.
She pulled up the platform's admin dashboard:
Active users (last 30d): 12
Internal users: 7
External users: 5
Monthly inference cost: $15K
Q3 platform revenue: $22K
She closed the tab. The numbers didn't need explaining. Of those 12 active users, most were automated evaluation engine requests — not human interaction.
The methodology system — she'd always known it existed, and she'd always known it was underfunded. VeriTest's senior consultants had built their own delivery framework over the years: assessment templates, risk checklists, industry benchmarking databases, historical case references. New consultant ramp time, delivery quality consistency, client renewal rates — all of it rested on this system.
But it had no product form. No UI. No demo. No sales pitch. Nobody had ever mentioned it at a board meeting — because the board didn't know it existed.
It had been limping along on a marginal budget the whole time. Maintained by a semi-retired senior consultant, two half-days a week, $9K a quarter.
Lena compared the two sides.
$200K in, $22K out. An 89% negative margin.
$9K in, supporting $3M annualized delivery — renewal rates seventeen points above industry — and not a single director knew its name.
She stared at the notebook for a long time.
She already knew the answer. But she didn't close the notebook.
She knew what she was about to recommend — and this time, she wasn't cutting someone else's feature for a client. This time, she was cutting down what she'd built herself.
She picked up the pen. The signature box sat at the bottom right of the proposal's first page.
She didn't sign.
For a second, she ran through tomorrow's conversation with the platform's tech lead — the engineer who'd built this thing from zero with her, pulled all-nighters tuning models, spent weekends fixing pipelines. He wasn't the type to go back to delivery work. She knew where he'd end up after that talk.
She signed. The pen hit the table louder than the numbers ever did.
She didn't look at the signature again. Closed the folder. The sound carried in the empty office, louder than usual.
The Pitch
Lena walked into Marcus's office. The proposal sat face-down on the table.
"You read it?"
Marcus flipped it open. Two pages in, he stopped on the third.
"You're shutting down the AI platform?"
"Not shutting down. Stopping it as a standalone operation. Embedding the capability back into the delivery lines."
"Is there a difference?"
"The difference is — shut it down, the team scatters, the tech assets rot on the server. Embed it back, and we can pull it out again in three years if we need it. We just stop running it as its own product team."
Marcus closed the proposal. Said nothing.
"Aren't you curious why?"
"I don't need to see your numbers twice," he said. "What's the headcount plan?"
"Five back to delivery. Two stay on core engine maintenance. One — I'll talk to him."
"Who?"
"The platform tech lead. He won't go back to delivery."
Marcus nodded. "You talking to him, or me?"
"I'll do it."
Checkmate
The boardroom was quieter than usual.
Lena's proposal was only four pages. Page one: the AI platform's cost-revenue curve. Page two: the methodology system's coverage and actual support scope. Page three: the integration plan — core engine embedded into delivery lines, independent product team dissolved, $67K/month in redirected spend — $54K to methodology upgrades, $13K to cover core engine maintenance and operations. Page four: the execution timeline.
She turned to page one and started.
The AI platform's numbers. Twelve months of cumulative cost. The client conversion funnel. She'd memorized every figure.
"My recommendation: stop the AI platform's independent product operations. Integrate its core capabilities into the consulting delivery system. Redirect the freed resources into systematic methodology development."
A director spoke. "You pitched this platform last year."
"I did."
"You said it was VeriTest's future competitiveness."
"I meant it then," Lena said. "And I mean it now. But the path to that competitiveness isn't burning three years of runway on a standalone product team waiting for it to break even. It should live inside our delivery capability first, funded by client revenue, not investor patience."
Another director flipped through the proposal. "The methodology system — we haven't paid attention to this before. Can it hold?"
Lena glanced at Marcus. He didn't speak.
"It's been holding for four years," Lena said. "On $9K a quarter. It just never made it to the boardroom."
A pause.
"Lena."
The one who spoke was the director who'd been quiet the entire meeting — the oldest, sitting at the far end of the table.
"Last year you convinced us to fund this platform. Now you're convincing us to kill it. How do I know you won't be back in six months telling us to kill the methodology too?"
Lena held his gaze.
"Because I won't."
"You said the same thing last year."
"Last year I meant it. I mean it now too. The difference is — **last year I told you what I believed. This year I'm telling you what I saw. "**
The director watched her for a moment. Didn't push further.
The room was silent for a few seconds.
The directors exchanged glances. No vote. No formal decision. One director said "Proceed as recommended," and nobody objected.
Lena gathered her papers and walked out.
In the hallway, Marcus caught up.
"The board didn't say thank you."
"I don't need them to."
She walked into her office, closed the door, and sat down.
She opened the notebook to the project charter — the first line read "VeriTest AI Evaluation Platform · Initiation · Proposed by Lena." She looked at it for ten seconds. Closed it.
She never opened that folder again.
Aftermath
Ninety days later.
The AI platform team had gone from eight people to two — the core engine maintenance crew. The original five had been redistributed across the delivery lines. The evaluation models accumulated on the platform had been packaged into modules and embedded into the standard due diligence workflow. Clients no longer saw "AI Assessment" as a product name — but the deliverables now had an extra page of automated analysis. Nobody noticed that page had once been the core output of the now-defunct platform.
The methodology upgrades moved in two directions: the delivery knowledge base expanded from 40 cases to 120. New consultant ramp time dropped from ten weeks to seven.
The real validation came quietly.
Lena read a record during the quarterly delivery review —
Thursday afternoon, a junior consultant in his third month, preparing a client due diligence report, pulled up a historical case from the knowledge base. A similar architecture migration project six years ago — the same risk pattern hadn't surfaced until the fourth month after deployment, costing the client $200K. In the risk assessment section, he'd added a citation line: "See VeriTest Delivery Case Library #47."
As far as the records showed, nobody noticed the citation — not the client, not the project manager, not management.
But VeriTest's renewal rate had ticked up 1.2 points this quarter.
Nobody knew where that 1.2% came from. Lena did.
Lena went to the kickoff with Marcus. The fintech client's tech lead wasn't a talker — he exchanged pleasantries and opened VeriTest's due diligence report immediately.
"Your report goes way deeper than the other two," he said, flipping to the risk assessment page. "You flagged a similar case from six years ago — the other two firms didn't do that. Only yours did."
Lena nodded. Didn't respond.
Marcus glanced at her. He didn't say anything either — but she recognized that look. It was the same case, used in two different reports. But sitting in that room, she was the only one who knew the truth: that case was in the report because the methodology budget had been approved — funded by her own platform that she'd cut.
Lena was at Third Cup when Marcus's message came through.
"Q4 methodology budget approved. At your numbers."
She typed back:
"Don't tell me. Tell the team."
Marcus replied with one word:
"Done."
Lena took a sip of her coffee. It was cold.
She glanced at her phone. The lock screen was still a screenshot of the AI platform's launch page from the day it went live the year before.
Someone behind the counter brought a fresh cup over. Didn't ask what she wanted — he just put the right one down.
"Switch?"
"No. This is fine."
She took a sip. The coffee was cold. She didn't put the cup down.
She never changed the lock screen. Not because she forgot. Because she wanted to keep it.
This is what Sacrifice the Plum Tree to Preserve the Peach Tree means — not giving up, but cutting down what you built yourself, so the pieces can keep something more important alive.
🤖 AI Post-Mortem
[36 Stratagems Tactical Database v3.1] Loaded
[Tactic Match] Sacrifice the Plum Tree to Preserve the Peach Tree
[Analysis Mode] Full-field scan
━━━━━━━━━━━━━━━━━━━━
Tactic Match: ~91%
Operator: Lena
Action: Recommended dissolution of own AI platform project, redirected budget to consulting delivery methodology
Objective: Trade a visible initiative for an invisible but critical capability
Result: $67K/mo redirected. Delivery knowledge base: 40→120 cases. Ramp time: 10→7 weeks.
Renewal rate: +1.2pp (quarter)
Decision Evaluation:
- Target: Board's AI platform ROI concern
- Sacrifice: AI Evaluation Platform (Lena's own project, 12 months in production)
- Keeper: Consulting Delivery Methodology & Knowledge Base ($9K/qtr, zero board visibility)
- Net: -6 FTE, +$54K/mo for methodology, $13K/mo retained for platform maintenance.
Risk Assessment:
Personal cost: High. Ownership on record. Board questioned the reversal.
Institutional cost: Low. Team redistributed. Technology preserved.
Reputation cost: Medium. First reversal in VeriTest tenure.
Core Insight: The hardest sacrifice is the one you volunteer before anyone asks you to.
Next stratagem: Borrow a Corpse to Return the Soul
P.S. English isn't my first language. I use AI to polish the writing and smooth out the rough edges. Thanks for reading. ☕ Buy me a coffee
Top comments (20)
I thought the article was about me lol cuz of the title 😃
But good stuff anyway gg
Wait, you're telling me this WASN'T secretly your origin story? 😄 Next time I'll just use "EmberNoGlow" as the title and save everyone the confusion. Thanks for reading!
good luck!
The part that lands hardest is that the two systems were never measured on the same denominator. The platform was its own P&L line, so it had to justify $200K against 12 users. The methodology's cost was amortized invisibly across $3M of delivery, so it never had to clear a bar by itself. Same capability, two accounting boundaries, and the boundary decided which one lived, not the tech.
The adoption angle sits underneath it: a standalone AI product has to earn its own users, and that fight is brutal because the value has to beat the switching cost of the workflow someone already runs. Embedded AI skips it. The ember rides on adoption it never had to win, because it sits downstream of adoption instead of upstream. "What survives being invisible" is the same truth said in prose.
You nailed it — "same capability, two accounting boundaries" is exactly what I was going for. Lena figured it out too late. She was fighting on the wrong denominator, and by the time she saw it, the spreadsheet had already made the call.
Appreciate you reading this closely. Means a lot, especially from someone who's been around since the early ones. 🙏
"The spreadsheet had already made the call" is the real horror of it. By the time the denominator is visible in the numbers, the decision is retroactive, you are just reading the verdict. The only defense is drawing the boundary before the P&L does, which almost nobody does because the flashy version is the one that gets its own line. Glad to be along since the early ones.
The case citation detail is the one that lands. A junior consultant, probably on their first year, pulls a six-year-old failure pattern from a system nobody thought was production-critical. The client catches it. That's not AI, that's not automation, it's just institutional memory doing its job -- quietly, invisibly, on $9K a quarter.
The Stratagems arc across these episodes has a clear pattern: the expensive AI thing gets headlines, gets budget, gets killed. The cheap thing nobody notices survives and quietly creates value. I'm starting to think the "ember" isn't about resilience. It's about friction. High-friction systems attract attention and resources. Low-friction ones just work and nobody cares until they don't.
One thread I keep following: what happens when someone tries to AI-ify the KB? Turn it into a chatbot, a recommendation engine, a "smart" archive. Does that break the low-friction property? Or is there a way to add capability without adding attention?
P.S. Please forgive the quality of my English — this reply was written through AI translation from my native language. I'm not confident the translation fully captures what I mean. Looking forward to your reply.
You just put words to something I've been feeling but couldn't say — friction is the hidden axis. The expensive system wasn't inherently high-friction, it just lived inside a P&L line that demanded attention. The cheap one never had to justify itself because nobody tagged it as a line item. Same product, different accounting. That framing is sharp and I'm stealing it.
On the KB question: that's literally what the next arc is about. The ember survived by staying invisible. The moment you wrap it in a chatbot, a "smart" archive, a recommendation engine — you've rebuilt the attention magnet. Now it has users, uptime, a roadmap. The uncomfortable question I keep circling is: can you add capability without adding surface area? Or is the only way to keep something alive to leave it unfinished?
This is actually the exact persona I'm trying to build for Lena across the series — a strategic chess player who knows when to sacrifice a piece to protect the long game. The ember wasn't luck. She left it there on purpose.
I checked out your moteDB stuff. You're literally building the low-friction memory layer in Rust that says "yes" to that question. The robot doesn't know the DB exists — that's the whole point, isn't it?
Curious: has anyone ever successfully bolted an AI layer onto a working KB without breaking what made it work in the first place? Or is that just a trap people keep walking into?
That "Sacrifice the Plum Tree" frame keeps landing in this series. What struck me this time isn't the cut itself â Lena probably saw it coming the moment she opened the folder. It's the part where she spent two days pulling financials instead of one. The platform's numbers were clear on day one. She just didn't want to write them down.
I've watched a near-identical arc play out in a smaller company. The internal AI tool had three paying users after a year, and the founder kept saying "we just need one more quarter." It wasn't the math that killed it. It was the meeting where someone finally had to say the number out loud. The decision was already made the moment the spreadsheet was opened.
The "ember" framing is generous, by the way. Most platforms in that spot don't leave a reusable thing behind â they leave a wiki and a few PDFs that nobody opens. What's the actual artifact in VeriTest's case? A pipeline, a dataset, a customer who stuck around?
The artifact isn't a pipeline or a dataset — it's a knowledge base that nobody knew existed until the platform was gone.
Before the cut: $200K/qtr on the AI platform, $9K/qtr on a methodology archive maintained by a semi-retired consultant in two half-days a week. Zero board visibility.
After the cut: platform engine gets folded into delivery — customers don't see "AI assessment" on the menu anymore, but their reports gain an extra page of automated analysis. The $54K/mo freed up turns a 40-case knowledge base into 120 cases. New consultant ramp time drops from 10 weeks to 7. Renewal rate ticks up 1.2 points.
The real artifact? A case citation in a junior consultant's report — referencing a six-year-old failure pattern from a retired project that the knowledge base still had on file. Nobody flagged the citation. But the client noticed. That's the ember.
The ember is often the real product. Full platforms get cut because they carry too many promises at once. A small surviving workflow is easier to trust because it has a clearer job, clearer failure modes, and a better chance of becoming part of how people already work.
You read that deeper than most. I genuinely love Lena as a character — she first appeared in the old series, Chapter #15, and from the very beginning she treated AI as nothing more than a tool. She knows office politics inside out. Clean moves, dirty moves — if it wins, she'll use it. That puts her a step above the other five protagonists.
Three chapters in a row now, all from her side. But I don't want to write her as invincible. She's human. She runs into walls — external ones, internal ones, the whole range.
Hmm… saying this out loud, I feel like I'm not the same writer who started this series anymore. The characters keep growing. So do I. I couldn't have written lines like this back then. 😄 Okay, enough self-praise. Back to thinking about the next stratagem.
Early spoiler: a protagonist from early in the series is coming back. But… how do I even put this? I got a little choked up writing it. 🥹Ah, let me just save it for the next stratagem.
I only noticed this after posting: the recap for the first 6 went out on the 6th. Then #7, #8, #9, #10, and today's #11 — all landed on dates matching their numbers. Didn't plan it, just worked out that way.🤣
i really like that, you can check my article if you want, it's optional, i need some feedback
Keep it up. At 17, I was still buried in video games, completely lost about the future. But at 17, you're already building your own business. Looking forward to good news.
Thank you, I really appreciate that 🙌
I’m still figuring things out, but I’m trying to build something real and useful. Messages like this honestly motivate me a lot to keep going.
Good stuff - read the numbers, swallow your pride and change course, but don't throw the baby out with the bath water ;-)
"Read the numbers, swallow your pride" — basically the thesis statement of #11 summarized in 6 words. 🔥