Boss Fight B-B: Own a POD's AI uplift for a month

Status: drafted · Time: 1800 min · Audience: platform-builder Outcome: Embed with a POD outside your own for one month; intervene with the full Black Belt toolkit; ship a signed-off metric lift and a one-pager case study contributed back to this playbook.

Belt progress: the closer of Black Belt Time budget: ~1 calendar month elapsed; ~2-4 active hours per week of pairing + the case-study writing Prerequisite: Black Belt Parts A, B, and C read; Quests B-1 and B-2 claimed; nomination registered What you’ll prove: that the leverage you have built propagates with you in the room — that a POD outside your own can move materially faster on a metric they care about because of the moves you make over a month, and that the lift is real, measured, and reproducible

---

What this boss fight is

The Black Belt capstone. INDEX commits the success criterion: a signed-off metric lift plus a one-pager case study contributed to this playbook. The boss fight is to embed with a POD outside the candidate’s own for one calendar month, pick a shared metric with the POD lead, intervene through the full Black Belt toolkit (Parts A, B, and C in concert), measure the lift, write the case study, and present in an all-hands or the program’s main forum.

This is the moment where Black Belt’s promise (I am a force multiplier for Razorpay) is tested with evidence rather than self-report. Quests B-1 and B-2 prove cross-POD adoption (B-1) and cross-layer ownership (B-2). The boss fight proves cross-POD impact.

---

What does NOT count

• Embedding with the candidate’s own team or with a team that shares the candidate’s manager.
• Picking the metric unilaterally; the metric must be the POD’s, signed off by the POD lead before the embed begins.
• A metric that moves on its own (e.g., “PRs merged” during a quarter where the POD’s headcount doubled) without controlling for the underlying change.
• A lift claim that cannot be reproduced: no baseline measurement, metric tracked differently pre and post, or one-off conditions during the embed window.
• A case study that names individuals, exposes regulator-protected data, or relies on numbers the POD lead has not approved for sharing.
• A purely advisory intervention — “I told them to use X.” The candidate must have built or shipped something during the embed (a shared skill, an MCP, a custom agent, a documented practice that the POD now uses).
• A boss fight that lacks a Black-Belt-or-above reviewer in the chain (per Appendix L).

---

How to do it

Phase 0 — Pre-embed (~1-2 weeks before)

Find a POD. The candidate identifies a POD outside their immediate team that has a measurable challenge they could plausibly help with. Channels: the office-hours queue (B.12), the embedded-sprint debrief logs (B.13), direct outreach by a POD lead, or a cohort-lead-suggested match.

The POD lead must be a willing partner. A reluctant POD will not produce signal; the boss fight is not a one-sided mandate.

Pick the metric. With the POD lead, in writing. The metric is the POD’s business metric — what they are already tracking or what they would track if the data were available. Examples (illustrative):

• median time-to-first-commit for new builders joining the POD;
• PRs merged per builder per week (controlled for headcount);
• cycle time on a defined ticket class;
• skill-trace count (how often the POD invokes shared skills);
• a Voice-of-Customer metric the POD already publishes.

The metric cannot be one the candidate proposes unilaterally. The POD lead’s sign-off ensures the metric is one the POD genuinely cares about, which means the lift will be one they will defend in the all-hands.

Establish the baseline. With the POD lead, in writing, before the embed begins. The baseline is what the metric was over the most-recent comparable window — typically the four weeks preceding the embed. Documented in the same place the metric is documented. Numbers shared with the candidate.

Scope the intervention. A rough sketch (the full plan emerges during the embed) naming what the candidate expects to do across the four weeks. Common shapes: a shared skill the POD adopts (per Quest B-1’s pattern); an MCP integration the POD did not have; a set of office hours and one or two embedded sprints (per B.12 and B.13); a documented practice the POD then runs.

Calendar. The four weeks are blocked on the candidate’s calendar. The POD lead’s standups and sprint reviews are on the calendar. The all-hands or program-forum slot for the case-study share is booked tentatively.

Phase 1 — Week 1 of the embed: orient and baseline-confirm

Day 1. The candidate joins the POD’s standup. Reads the POD’s CLAUDE.md (if it exists; if not, that may be one of the interventions). Reads the recent PRs. Reads the most relevant Slack channels. Pairs with the POD lead for 60 minutes on the metric: confirming the baseline, agreeing on the measurement methodology, agreeing on what would falsify the lift claim.

Days 2-5. The candidate observes. Joins meetings. Pairs with builders on existing work. Takes notes on what the POD finds hard. The first week is mostly listening — interventions that bypass the listening phase will land badly because they will not be the right interventions.

End of week 1. The candidate names what they think the right interventions are in a thread with the POD lead. The POD lead validates or challenges. The intervention plan is set.

Phase 2 — Weeks 2 and 3: build and ship

Build. Shared skills, MCPs, custom agents, documented practices — whichever the intervention plan named. The candidate builds with the POD’s builders, not for them — the same ship-with-not-for discipline as B.13’s embedded sprints.

Run cultural moves. Office hours for the POD if relevant; embedded-sprint patterns for specific deliverables; tool teaching where it lands.

Measure continuously. The metric is tracked through the two weeks. Spikes and dips noted; possible causes named. The candidate is not trying to game the metric; they are trying to make it move.

Document. A running log: what was done, what worked, what did not. This becomes the raw material for the case study.

Phase 3 — Week 4: consolidate, measure, write

Days 1-2. Last interventions. Deprecate anything that did not work. Hand over the things that did to the POD’s builders so the lift is not dependent on the candidate’s continued presence.

Days 3-4. Final measurement. The lift (if any) is calculated against the baseline. The methodology is documented; the numbers are shared with the POD lead for sign-off. If the lift is positive but small, that is fine — small lift is real lift and should be claimed honestly. If the lift is zero or negative, that is also fine for the boss fight — the case study is what matters; an honest “this did not move the metric, here is what we learned” is more valuable than a lift claim the reviewer cannot defend.

Day 5. Write the case study. One page (~600-800 words). Sections per the template below.

Phase 4 — Post-embed: present and submit

Present. The case study is shared in an all-hands or the program’s main forum. The presentation is short — 10 minutes plus questions. The POD lead is in the room (in person or async); the lift, if claimed, has their sign-off.

Contribute the case study back to the playbook. The case study lives in the playbook’s case-study archive (a case-studies/ directory adjacent to the appendices, created when the first case study lands). Future Black Belt candidates read past case studies as part of their pre-embed scope.

Reviewer routing. Per Appendix L. For Boss Fight B-B specifically: at least two reviewers, one of whom is out-of-team (per Appendix L’s standard rule for Black Belt) and one of whom is Black-Belt-or-above (the boss-fight-specific stricter rule). The two reviewers can be the same person if that person satisfies both criteria; usually they are different.

---

The case study template

A one-page (~600-800 words) artefact. Six sections.

# Case Study: [POD identifier or anonymised name] AI uplift

**Embed window:** YYYY-MM-DD to YYYY-MM-DD (4 weeks)
**Candidate:** <handle>
**POD lead:** <handle>
**Reviewers:** <out-of-team reviewer handle>, <Black-Belt-or-above reviewer handle>

## 1. Problem

What was the POD trying to move? Why did they care? What was the baseline?

## 2. Intervention

What did the candidate do across the four weeks? Shared skills, MCPs, agents, cultural moves — name them.

## 3. Result

The lift, with the methodology. The baseline number, the post-embed number, the percentage change, the confidence (if applicable).

## 4. What would not have worked

A short paragraph naming an intervention that *did not* move the metric, or that the candidate considered and rejected. The honesty is part of the artefact's value.

## 5. What the next embed should know

One paragraph for future Black Belt candidates considering a similar embed. Lessons that compound across the program.

## 6. Sign-offs

POD lead's name, date, and a one-sentence "I confirm the baseline, the intervention, and the result as described."
Reviewers' names, dates, and a one-sentence "I have read the case study and the underlying evidence and find it consistent with the boss-fight rubric."

The template is shape-only; the content is the candidate’s. A case study that strays significantly from this shape needs a reason in the reflection.

---

Worked example — a case study shape

A representative shape (no real POD; numbers illustrative).

Problem. The POD owns a recurring weekly status report that takes ~3 hours of a senior builder’s time. The metric the lead picked: median time spent producing the weekly report.

Intervention. Week 1 spent observing the report-writing flow. Week 2-3: authored a weekly-status skill that pulls from the POD’s tracker and primary Slack channel and produces the report in the POD’s standard format; paired with a mid-level builder on the authoring; merged it to razorpay/agent-skills after owning-team review. Week 4: helped the POD author a small weekly-status-config document that the skill consumes; deprecated three internal scripts the POD had been using ad-hoc; ran two office hours for the POD focused on shared-skill adoption.

Result. Baseline: 3 hours per weekly report (4-week mean). Post-embed: 35 minutes per weekly report (2-week mean during weeks 5-6 post-embed). 80% time saving. Methodology: timed by the report-writer; recorded in the POD’s tracker; signed off by the POD lead.

What would not have worked. An off-the-shelf reporting tool. The POD lead and I considered it; the report’s shape is specific enough that a generic tool would have produced a 50% time saving at best, and the team would have invested time learning the tool that they could re-invest in their own work instead.

What the next embed should know. The POD’s CLAUDE.md was unusually well-written, which made the embed possible in four weeks. A POD with a poor CLAUDE.md would need at least a fifth week, or the candidate should narrow the intervention to fit the four-week box. The pattern of “shadow week 1, build weeks 2-3, hand-over week 4” is robust; deviating from it (e.g., starting to build in week 1) cost a peer Black Belt their lift on a different embed.

Sign-offs. POD lead, 2026-XX-XX. Reviewers: out-of-team reviewer 2026-XX-XX, Black Belt reviewer 2026-XX-XX.

The artefact is ~500 words. It is small on purpose. The next Black Belt candidate reads ten of these and understands the embed pattern across ten different POD shapes.

---

Common pitfalls

Picking the metric the candidate wants. The lift then “happens” because the metric was reverse-engineered. Reviewers see through it. Fix: the POD lead picks the metric, in writing, before the embed.

Measuring once instead of continuously. The baseline and post-embed numbers compete; without continuous measurement, attribution is impossible. Fix: track the metric weekly through the embed and at least two weeks after.

Over-claiming the intervention’s contribution. The metric moved 30%; the candidate claims credit for all 30%; in fact, the POD’s headcount also grew 10% in the same window. Fix: name confounders honestly; if the metric moved for multiple reasons, name the multiple reasons.

Skipping the case study because “the metric speaks for itself.” It does not. Future Black Belt candidates need the narrative that converted the intervention into the lift. Fix: 600-800 words; the case study is the artefact, not the metric on its own.

Embedding with a POD that does not have a willing lead. The POD goes through the motions; the lift is artificial; the case study is shallow. Fix: pre-embed conversation surfaces the willingness; if the POD lead is not enthusiastic, find a different POD.

Treating the embed as a side-project on top of normal work. The candidate is “embedded” in name; in practice still doing 100% of their original POD’s work. Fix: the embed needs meaningful time — typically 50-80% of the four weeks is not realistic; 30-50% is. The candidate’s home POD must accept the load shift.

Confusing this boss fight with a permanent transfer. The candidate enjoys the new POD and stays. That is a transfer, not a boss fight. Fix: the four-week box is binding; if a transfer makes sense, it is a separate conversation after the boss fight closes.

Writing the case study in private and not contributing it back. The lift exists for the candidate’s tracker row only. Fix: the case study is contributed to the playbook’s archive — that is the propagation mechanism for future candidates.

Skipping the all-hands share. The lift exists in the case study but not in the org’s collective memory. Fix: the all-hands share is required by the boss-fight rubric; book the slot in pre-embed.

Reviewer routing skipped. A peer rubber-stamps. Fix: two reviewers per Appendix L’s stricter rule for the boss fight; one out-of-team, one Black-Belt-or-above.

---

Reviewer routing

Per Appendix L. For Boss Fight B-B specifically: two reviewers, one out-of-team and one Black-Belt-or-above. The two reviewers can be the same person if that person satisfies both criteria; usually they are different. Reviewers attest:

• the embed began and ended on the named dates;
• the metric was picked by the POD lead, signed off in writing, before the embed began;
• the baseline measurement is documented and methodology is sound;
• the intervention is specific (built or shipped, not advisory);
• the lift claim is consistent with the documented evidence;
• the case study has been contributed to the playbook’s archive;
• the all-hands share has happened (or is scheduled within two weeks of the boss-fight close).

If the lift is zero or negative, the boss fight can still be claimed if the case study honestly names what was attempted, what did not work, and what the next embed should learn. The reviewer’s job is to evaluate the honesty and rigor of the artefact, not the size of the lift.

---

Evidence template

Copy into your tracker or LEARNER.md:

## Boss Fight B-B — Own a POD's AI uplift for a month

- Builder: <handle>
- Embed start: <YYYY-MM-DD>
- Embed end: <YYYY-MM-DD>
- POD identifier (or anonymised): <handle / identifier>
- POD lead: <handle>
- Metric chosen: <metric name>
- POD lead's metric sign-off: <link to written confirmation>
- Baseline: <value, methodology>
- Post-embed value: <value, methodology>
- Lift: <%, with confidence if applicable>
- Intervention summary: <link to running log or artefact list>
- Case-study link: <link to the playbook's case-study archive>
- All-hands share confirmation: <link or date>
- Reviewers: <out-of-team reviewer handle>, <Black-Belt-or-above reviewer handle>
- Reviewer attestations: <links>

The case study itself is the primary artefact. The reviewer attestations are the proof of the chain. The all-hands share is the proof of propagation.

---

What you can say after this boss fight

“I embedded with a POD outside my own for a month, picked a metric they cared about, intervened through shared skills and cultural moves, shipped a measurable lift signed off by the POD lead, wrote a case study contributed back to the playbook, and shared the result in an all-hands. The leverage I built propagated with me in the room; my Black Belt is what I built, not what I was given.”

---

Previous: ← B.16 Plugin + skill governance · Next: → Black Belt badge

Further reading

• B.12 — Running office hours — the cultural move that often surfaces the right POD to embed with.
• B.13 — Embedded sprints — the one-week pattern; the boss fight is the four-week version.
• B.10 — Cost + observability — the measurement discipline the boss fight depends on.
• Appendix L — Certification — the reviewer rules for the boss-fight close-out.