Diagnostic sample · public information only

A Diagnostic, on a company I’ve never touched.

The most common question before anyone buys: “What do I physically get for €2,900?” Fair. So here is the deliverable’s exact shape — the same six artifacts, in the same format — worked end-to-end on a company you can check me against: Duolingo (NASDAQ: DUOL).

I have never worked at Duolingo and have no inside access. Every observation below is inference from the public record — the app, visible lifecycle and growth tactics, earnings disclosures, and press coverage. Read it as a demonstration of the method and the format, not as a claim about their internal systems. A paid Diagnostic replaces this guesswork with your actual stack.

Read this first

Public-info only. No NDA data, no analytics access, no employee conversations. If I can’t see it from the outside, it isn’t here.
Inference, framed as inference. Phrases like “from the outside, it looks like…” are load-bearing. They mean a hypothesis, not a finding.
Not affiliated with Duolingo. Not endorsed by or reviewed by them. Used only because their growth machine is unusually well-documented in public.

How to read this page

The same six artifacts, rebuilt from public information.

A real Diagnostic ships six artifacts. Below, each section is labeled with the artifact it stands in for — so you can see the format, even though the data underneath is public rather than yours.

Artifact 01 · 02: Executive map — what’s strong, what’s at risk → §1
Artifact 03: Lifecycle / CRM architecture read → §2
Provenance: Deliverability & data note → §3
Artifact 04 · 06: Prioritized opportunities + hypotheses → §4
Artifact 05: The 90-day operating sketch → §5

§1

Artifact 01–02 · Executive map & system view

What looks strong. What looks at risk.

In a paid Diagnostic this is the two-page brief your CFO reads first, drawn from your own stack. Here it’s assembled from what’s visible publicly: the app’s mechanics, the shareholder letters, and the growth press. Treat every cell as a public-info read, not a finding.

Looks strong — from the outside

A habit loop that reportedly compounds. Streaks, the home-screen widget, and timed notifications form a closed daily loop. Public write-ups credit the iOS streak widget with a step-change in daily opens, and Duolingo’s own commentary ties 7-day streaks to materially higher long-term retention. From the outside, this is the engine — and it appears genuinely well-built.
Engagement converted into a public scorecard. Q3 2025 disclosures put DAU at ~50.5M and paid subscribers at ~11.5M, with paid penetration around 9% of trailing MAUs — a freemium funnel that is clearly working at scale.
A credible AI upsell tier. Duolingo Max (Video Call with “Lily”, Roleplay) gives the premium ladder a top rung that earnings calls have linked to higher-priced-plan strength. A public-info read suggests monetization has somewhere new to go, not just more of the same.

Looks at risk / worth a question — inference only

The notification voice carries brand risk. The “guilt owl” is beloved and mocked in roughly equal measure. From the outside it’s impossible to tell whether pressure-toned nudges are capped per user or whether some segments get over-messaged into resentment. This is exactly the kind of question a Diagnostic exists to answer with your data — here I can only flag it.
Penetration at ~9% means the 91% is the real estate. A huge free base is a strength and an exposure: the conversion story leans on the same loop continuing to convert. A public-info read can’t see where that funnel leaks — only that the leak, wherever it is, is where the money is.
“Explain My Answer” went free in early 2026. Publicly, that narrows Max’s differentiation toward the two conversation features. From the outside it raises a fair question about upgrade-prompt clarity — answerable only with internal trial-to-paid data.

Reminder · none of the above is a statement about Duolingo’s internal systems. It is what a careful outsider can infer from the public record, arranged in the deliverable’s format.

§2

Artifact 03 · Lifecycle / CRM architecture

The lifecycle, stage by stage.

The real artifact is a journey-by-journey gap table built against your CRM: what exists, what fires, what quietly died. I can’t see Duolingo’s internal journeys, so this reconstructs the visible lifecycle by using the product and reading the public record, then marks where an outsider simply can’t know. That “can’t-know” column is the point: it’s precisely what the paid version fills in.

Stage	Visible mechanic (public)	Outside read	Can’t know without access
Activate	First lesson before signup; goal-setting; early streak start.	Friction pushed late — value before the ask. Looks deliberate.	Day-1 → Day-7 activation curve; the first-session step that loses the most people.
Habit	Streak, freeze, widget, daily push (“Don’t let Duo down”).	The core loop. Streak-freeze softens loss aversion so the streak survives a missed day.	Per-user notification caps; whether nudge tone is segmented or one-size-fits-all.
Engage	XP, Leagues, leaderboards, weekly competition.	Social-comparison layer on top of the solo habit. Adds a second reason to return.	Whether Leagues lift retention net of the users they discourage.
Monetize	Light interstitial ads · Super · Max · Family · 14-day trial.	Ad load looks deliberately low to protect the upgrade moment. Tiered ladder, clear rungs.	Trial-to-paid by source; where the upgrade prompt converts vs. annoys.
Resurrect	“Your streak may be gone, but your progress isn’t” win-back; lapsed-user pushes/email.	Public talks frame the first 3–4 days after lapse as the window. Progress-ownership framing over pure loss.	Win-back send cadence, channel mix, and actual reactivation rate by recency.

The right-hand column is the honest part. For a paying client, those blanks are filled from the CRM and become the gap table — with what fires, what’s dormant, and what should exist but doesn’t.

§3

Provenance · Deliverability & data note

What I’d test before I trust a single number.

In a real engagement, this note lists the deliverability and data-integrity checks I run before I believe any reported metric — sending domains and authentication, consent and suppression hygiene, identity stitching across app and email, and which instruments are known to lie. From the outside I can run almost none of these on Duolingo, which is itself the lesson: this is the part of the deliverable that is impossible to fake from public data.

What a public-info read can say is method, not verdict. A notification- and widget-led machine like this one lives or dies on signals that don’t survive naïve measurement: push opt-in rates by OS, iOS Mail Privacy Protection inflating any email open rate, and the gap between “notification sent” and “human actually came back.” On the deliverability fundamentals — SPF, DKIM, DMARC alignment, list hygiene, the unsubscribe path — I would assume nothing and verify everything against the actual sending infrastructure. None of that is visible here, so none of it is asserted.

The honest gap

The shortness of this section is the tell. Deliverability and data-trust work needs your DNS records, your ESP, your consent logs, and your event stream — the things a Diagnostic gets and a public-info sample never can. If a vendor claims a deep deliverability read of a company they have no access to, be skeptical.

§4

Artifact 04 · 06 · Prioritized opportunities + hypotheses

Three I’d test first — stated as hypotheses.

On a paying client these are quick wins and prioritized bets, each anchored to something in your stack with a cost next to it. From the outside I can’t size them — so they’re framed as falsifiable hypotheses with the test attached. That framing is the method: a Diagnostic doesn’t hand you opinions, it hands you experiments you can run.

Opportunity 01 Retention

Notification fatigue capping, by segment.

Hypothesis: a meaningful slice of lapsed users left partly because of nudge volume or tone, not loss of interest. Public signal: the “guilt owl” is a recurring punchline in reviews and social — affection and irritation in the same breath. Test: per-user weekly notification cap and a softer tone variant, with a holdout; read 28-day retention and reactivation as deltas against control. Why it’s honest: from the outside I can’t prove fatigue exists — but it’s cheap to falsify, and that’s the point.

Opportunity 02 Monetization

Re-grounding the Max upgrade story.

Hypothesis: with “Explain My Answer” now free, the Max prompt may be selling features users already have. Public signal: reporting that the grammar-explanation feature moved to the free tier in early 2026, leaving Video Call and Roleplay as the live differentiators. Test: trial-to-paid A/B that leads the upgrade moment with the conversation features specifically, surfaced right after a user’s first real speaking attempt. Outside read: a public-info read can flag the message-market fit risk; only internal trial data can size it.

Opportunity 03 Resurrection

Win-back inside the 72-hour window.

Hypothesis: the highest-ROI reactivation spend sits in the first three to four days after a lapse, and is easy to under-invest in. Public signal: Duolingo’s own people have discussed this window publicly, and the “progress isn’t gone” framing is visible in their win-back. Test: recency-banded win-back (0–3d / 4–7d / 8–14d) with progress-ownership creative, measured as incremental reactivation against a holdout per band. Why list it if they may already do it: a Diagnostic confirms what’s actually firing — the public version can only nominate the bet.

§5

Artifact 05 · The 90-day operating plan

If I were inside: the first 90 days.

The real artifact is sequenced with owners, KPIs, and dependencies — executable by your team, me, or anyone. This is a sketch of the shape: no owners, because I’d need to meet the team; no KPI targets, because I’d set those against your baseline, not a guessed one.

Days 0–30 · See clearly

Instrument the truth before touching anything. Establish the retention and trial-to-paid baselines as deltas on clean cohorts, audit deliverability and consent end-to-end, and map which notifications fire to whom. Deliverable: one honest baseline everyone agrees on. No change ships this month.

Days 31–60 · Test the cheap ones

Launch the three §4 hypotheses as controlled experiments with holdouts — notification capping, Max-prompt re-grounding, recency-banded win-back. Read deltas, not absolutes; kill what doesn’t move and let it stay dead. Deliverable: two or three significance-read results, not a dashboard.

Days 61–90 · Bank the winners

Productionize what won, document what lost so it isn’t re-litigated, and hand over a roadmap tied to LTV-against-CAC and the shape of the retention curve — not opens and clicks. Deliverable: a plan the next quarter runs on, owned by your team.

Sketch only · public information. On a paying client every box above carries an owner, a number, and a date.

What this was, and wasn’t

The format is real. The access was the missing half.

Everything above is what a careful outsider can build from the public record, arranged in the deliverable’s exact shape. The thing it can’t do — the thing you’re actually paying €2,900 for — is replace inference with your data: your stack, your sends, your consent logs, your retention curve, and 2–3 short conversations with the people who touch CRM. That’s the half that turns “from the outside it looks like…” into “here is what’s leaking, where, and what it costs.”

The honesty rules I held myself to here are the same ones every number on a real Diagnostic must pass. If that’s the bar you want held against your own systems, that’s the work.

Scope the Diagnostic — 20 min See what the real deliverable contains

Built from public information only · no inside access · not affiliated with, endorsed by, or reviewed by Duolingo, Inc. · figures cited from public earnings disclosures and press coverage as of late 2025 / early 2026 and may have since changed. Shows the deliverable format, not a real audit.

Think this maps to your situation?

Start with the Diagnostic or see how we’d work together