Skip to content
Academy
Marketing Academy · Field Work●Growth Marketing
All projects·From: Building an Experimentation Program
MiniAudit· 25 minutes

Grading a Growth Team's Experimentation Maturity

Nubank (Nu Holdings)

Objective: Given a raw backlog export from a real growth team, score their experimentation process against the five-gear framework and identify the single weakest gear.

You're a growth analyst embedded with Nubank's app growth team for a quarter. They've handed you a backlog export and asked you to diagnose why velocity is stuck at 2-3 tests a quarter.

Score hypothesis compliance and ICE-scoring coverage from the raw backlog, then name the weakest gear with a concrete fix.

Before you start

What you'll need

Free path (everything below is enough to finish)

Google Sheets
FreeScore and rank the backlog

Free, no account friction, sortable columns for ICE ranking

The process

2 steps

Step 01 of 02

Filtering ideas through a structured hypothesis format before backlog entry

Gear 1 requires every backlog submission to state 'If we change X for users Y, metric Z will move because rationale R.' Without that structure, ideas are wishes, not experiments.

Nubank's backlog has 18 entries. Only 4 use the full hypothesis format; the other 14 are one-line feature requests like 'Test a new home screen banner.' What's the actual state of Gear 1?

Google Sheets— Open backlog-export.csv, add a column flagging which rows contain all three hypothesis components (change, audience, metric).

Procedure

  1. Import backlog-export.csv into Sheets
  2. Add a column 'Has Hypothesis' and mark TRUE only for rows naming a specific change, a specific user segment, and a specific metric
  3. Count TRUE vs FALSE to get the compliance rate
Sample output
18 backlog rows
  4 rows: full hypothesis (change + audience + metric + rationale)
  14 rows: title only, no metric named
  Compliance rate: 22%

Healthy

Compliance rate above 80% — most ideas already name a specific metric and audience before they reach the backlog.

Unhealthy

Compliance rate under 30% — the backlog is really a wishlist, and prioritization scoring later will be guessing, not measuring.

What this means

A low compliance rate means Gear 1 is broken even if Gear 2's scoring template looks polished; you can't score a hypothesis that was never written.

So what do I do about it?

SymptomActionEffort
Backlog entries are one-line feature requests with no named metricAdd a required 'hypothesis' field to the intake form and reject submissions missing it30 min
YouYou can do this yourself, no engineering access required.

Step 02 of 02

Scoring backlog items with ICE before committing engineering time

Gear 2 requires every hypothesis to be scored (ICE or PXL) before anyone commits engineering time. 58% of teams skip this entirely, and that single gap kills velocity more than anything else.

None of Nubank's 18 backlog rows carry an Impact, Confidence, or Ease score. Engineering picks the next test from whichever PM asked most recently. What does this tell you about Gear 2, and what's the fix?

Google Sheets— Add Impact / Confidence / Ease columns (1-10 each) next to the 4 rows that already have a real hypothesis, and compute an ICE score.

Procedure

  1. Add three columns: Impact, Confidence, Ease (1-10 scale)
  2. Score only the 4 rows with a real hypothesis, since the other 14 can't be scored honestly yet
  3. Sort descending by ICE score to produce a ranked shortlist
Sample output
ICE-scored shortlist (of 4 valid hypotheses)
  1. Simplify KYC upload flow   I:8 C:7 E:6 = 336
  2. Reorder home screen cards  I:5 C:4 E:8 = 160
  Remaining 14 backlog rows: unscored, unranked

Healthy

A ranked shortlist exists and engineering pulls the next test from the top of that list, not from whoever asked last.

Unhealthy

Zero rows carry a score, and the next test is chosen by seniority or urgency of the request, not evidence.

What this means

Missing ICE scores plus a HiPPO-driven pick order is the exact failure mode the lesson calls out: 58% of teams skip prioritization, and it is the single most common reason velocity collapses.

So what do I do about it?

SymptomActionEffort
Engineering picks tests by whoever asked most recentlyRequire an ICE score on every backlog row before it can be assigned to a sprint30 min
YouYou can do this yourself, no engineering access required.

Final deliverable

A one-page maturity scorecard for Nubank's experimentation program: hypothesis compliance rate, ICE-scoring coverage, and the single weakest gear with a named fix.

See a reference example
Sample output
Grab experimentation maturity scorecard (reference)

Gear 1 (Intake): 91% hypothesis compliance — healthy
Gear 2 (Prioritization): 100% ICE-scored before sprint assignment — healthy
Gear 4 (Review): no weekly council, SRM checks ad hoc — WEAKEST GEAR
Recommended fix: stand up a 30-minute weekly experimentation council.

Success criteria

You're done when you can:

  • Correctly computes hypothesis compliance rate from the raw backlog
  • Identifies missing ICE scores as the primary velocity blocker, not a tooling problem
  • Names one concrete fix tied to the weakest gear