AACE AI SW
👤ParticipantACE AI SW
DAY 02 · TUESDAY · REAL PROJECT 1

Break down a service
with growth hacking,
rebuild it with AI

Use growth hacking (PMF · North Star Metric · AARRR · growth loops) to see why well-built services grow, then rebuild that structure with AI. You will see for yourself why what you design and what AI builds turn out differently.

AI seems like it can build anything, but it can't build the invisible growth structure for you. That's why growth-hacking analysis comes first.

HTML Presentation

Day 2 Slides · Growth Hacking

Use the arrow buttons or keyboard arrow keys (←/→) to move. Press the button (or the N key) to open the lecture notes for each slide.

ACE AI SW Startup Training · Day 2

Growth Hacking ×
AI Reverse Engineering

Break down the growth structure of a well-built service with growth hacking, rebuild it with AI, and experience "why it doesn't come out the same".

WHY?Why isn't it the same?01LEARNConcepts · tools02OBSERVEObserve growth features03REBUILDRebuild with AI04COMPAREProve the gap
01 · LearnGrowth hacking concepts and tools
02 · ObserveObserve growth features in a real service
03 · RebuildReverse-engineer and rebuild with AI
04 · CompareProve the gap between design and result
Opening question

"AI, build me an app like Jumia."
Will you get Jumia?

AI builds a product list, cart and review screen in minutes. But will that app grow?

AI-made screenReal serviceFLASH SALE 02:14:09★4.6 (120)OFFICIAL★4.5 (157)★4.4 (194)★4.3 (231)≈?Screens look alikeReasons differDotted lines = growth features hidden in the real service (diagram)

What AI does well

  • Visible screens (UI layout, buttons, cards)
  • Common features (search, cart, login)
  • The structure of an "average" online store

What AI can't create by itself

  • Why information is shown in that order (intent)
  • What brings users back (loop)
  • Decisions refined by years of experiments (data)
You can copy the screen, but not the reason it grows. Growth hacking is the skill of finding that reason.
HANDS-ON 0110 minSolo · AI tool

First, just ask AI as-is

GoalSee with your own eyes the "average app" AI makes without analysis
1
Open AIChatGPT · Claude · Gemini — any one
2
Type it as-isDon't edit the prompt
3
Save a screenshotFile name BEFORE-01
4
Count trust featuresRating · reviews · delivery, etc. — 5 items

What do you count? · 5 trust features

Rating
Number of reviews
Official store
Delivery · returns
Verified seller
Build an app like Jumia. Product list and cart.

Can you see it in the AI result? · Tap to mark (seen → not seen → reset)

BEFORE-01 trust features
0 / 5
0/4
Tick the boxes
to check yourself
Career linkA common interview question: "AI can build apps, so why do we need you?" — today's BEFORE vs AFTER difference is your answer.
What is growth hacking?

Growth hacking =
fast, data-driven experiments for growth

A term first used by Sean Ellis in 2010. It crosses marketing, product, data and development to find "what creates growth" through hypotheses and experiments.

"Hacking" doesn't mean breaking in illegally — it means a creative shortcut to growth with few resources.
MarketingProductDataDevelopmentGROWTHFast experimentsGrowth experiments happen where the four areas overlap
GoalJust one: sustainable growth
EvidenceData and user behavior, not feelings
MethodRepeat small, fast experiments
MeansChange the product itself, not just ads
Traditional marketing vs Growth hacking

Growth features hide inside the product

ACQUISITIONAcquisitionACTIVATIONActivationRETENTIONRetentionREFERRALReferralREVENUERevenueScope of traditional marketingScope of growth hacking — the whole user journey

Traditional marketing

  • Goal: awareness, exposure, traffic
  • Scope: the top of the funnel (Acquisition)
  • Method: big-budget campaigns, results checked afterwards
  • Owner: marketing team

Growth hacking

  • Goal: the whole path — acquisition → activation → retention → referral → revenue
  • Scope: the whole user journey (AARRR)
  • Method: small fast experiments, judged by metrics
  • Owner: planning, dev, design and data together
  • Review count and rating → reduces worry about buying
  • "Up next" autoplay → keeps people watching longer
  • Verified-seller badge and feedback → trust in the other party
  • Invite-a-friend reward → referral and new users
Step 0 · Product-Market Fit

Before growth, check:
Do people really want it?

Using growth tactics on a product people don't want is like pouring water into a bucket with a hole.

First question when reverse-engineering: "If this service disappeared, what would users miss most?"
Answers to "What if you couldn't use it anymore?"Example numbers for illustrationService APMF signal46%34%20%Service BNot yet18%42%40%40% lineVery disappointedSomewhat disappointedNot disappointed
Q."How would you feel if you could no longer use this product?"Sean Ellis PMF survey
40%+If more than 40% of users say "very disappointed", it's a PMF signal.Baseline
Why?The core value the "very disappointed" group names = the service's real growth engineWhere reverse-engineering starts
HANDS-ON 0210 minWhole class · vote

"What if this app disappeared tomorrow?" Class PMF vote

GoalDecide for yourselves, using the 40% line, whether our class really loves this app
"How would you feel if you could no longer use this app from tomorrow?"
Very disappointed
Somewhat disappointed
I don't care
1
Pick one appThe app the class uses most
2
Everyone raises a handOnly one of the three
3
Check 40%Is "very disappointed" 40% or more?
4
Hear the reasonsCore value in one line
Very disappointed
0
Somewhat disappointed
0
I don't care
0
40% line
Enter the votes
0/4
Tick the boxes
to check yourself
Startup tipYou can test a startup idea with the same question. Asking 10 friends is the cheapest and fastest market test.
North Star Metric

The one number everyone watches:
the North Star Metric

A metric that shows both the value users get and the company's growth. Every feature and experiment exists to move this number.

North Star MetricOrders completed after checking trust info (hypothesis)Value users getCompany growthVisit · browseCheck trust infoPayment completeInput metrics — levers the team moves with experimentsJumia learning hypothesis example
Jumia (hypothesis)Orders completed after checking trust info such as ratings and official storeBrowse → trust → buy
YouTube (hypothesis)Total time users spend watching happilyDiscover → watch → next video
Jiji (hypothesis)Number of contacts (chat or call) started between buyers and sellersLocal → trust → contact
  • Sign-ups, downloads → "vanity metrics" that don't show value
  • These examples are learning hypotheses based on public info — don't present them as facts
HANDS-ON 0310 minPairs · quiz

Guess the North Star: value or vanity?

GoalExplain in your own words how to tell a good North Star from a vanity metric
User valueCompany growthNorth Star

2 questions to ask yourself

① Did the user get value?
They traveled · saved money · got their food
② Did the company grow?
It led to sales, revenue or return visits
If either answer is "No" → vanity metric
e.g. downloads, sign-ups, followers
BoltRide-hailing · travel
PiggyVestSavings · investing
ChowdeckFood delivery
※ Learning hypotheses from public info · not official metricsCorrect 0 / 3
0/4
Tick the boxes
to check yourself
Career linkA common PM/growth interview question: "What is this service's North Star Metric?" — picking a vanity metric is a red flag.
AARRR · Pirate Metrics (Dave McClure)

Break the user journey into 5 stages

ACQUISITIONAcquisitionWhere and how do users come in?Visits · share by traffic channel
ACTIVATIONActivationDid they feel the "aha!" value on their first visit?First key action completion rate
RETENTIONRetentionDo they come back? What brings them back?Return rate · N-day retention
REFERRALReferralDo they tell others and bring them in?Shares · invite conversion rate
REVENUERevenueDoes value turn into money (or sustainability)?Purchase conversion · average order value

▼ Drop-off Where users leave between stages = the biggest growth opportunity

HANDS-ON 0415 minSolo → pairs · phone

Find AARRR features on your phone

GoalFind 5 growth features in an app you use every day, as facts (what you can see)
1
Open one appJumia · Jiji · OPay · Kuda · PiggyVest …
2
One per stageFind features in five stages
3
Screenshot · numberStart from OBS-001
4
Compare with partnerThe 1 most surprising feature

Look for things like these

ACQUISITIONAcquisitionApp store reviews · a link from a friend
ACTIVATIONActivationFirst-transaction bonus · sign-up tutorial
RETENTIONRetentionPush notifications · savings streak
REFERRALReferralInvite code · share button
REVENUERevenueFee notice · paid top placement
ACQUISITIONAcquisition
ACTIVATIONActivation
RETENTIONRetention
REFERRALReferral
REVENUERevenue
My AARRR Map
0 / 5
0/4
Tick the boxes
to check yourself
Career linkOrganize your screenshots + interpretation and it becomes the "Product Teardown" section of your portfolio.
From funnel to growth loop

A funnel ends, but a loop keeps spinning

In successful services, user actions grow the service's value, and that value brings in new users again — a cycle.

Always write this too: the condition that stops the loop (e.g. if there are no listings in my area, nobody comes back)
① User action② Value grows③ Other users respond④ New & return visitsGROWTHLOOP ↻Stop conditionIf listings, reviews or videosrun short, the loop stops
YouTubeWatch → taste data builds up → better recommendations → watch longer
JijiSellers post ads → more local listings → more buyers visit → more sellers post
JumiaRate and review after buying → more trust info → new buyers worry less → more purchases and reviews
Classic growth hacks

Cases where one small feature created explosive growth

InviterInvitee+GB DropboxInvite a friend and both get extra storage. The reward is the product's core value itself.REFERRAL loop
Existing big platform AirbnbEarly on, it let hosts post listings to Craigslist too, bringing users from a place that already had crowds.ACQUISITION shortcut
×N HotmailA "Get your free email" link at the bottom of every email. The more people use it, the more it advertises itself.Viral loop
What they share: not ad spending, but small design choices inside the product. And AI won't think of these from a simple "build it" request.
The growth experiment cycle

Growth isn't an idea — it's repeated experiments

01 · AnalyzeFind drop-off and opportunity points in metrics and user behavior
02 · IdeasLots of "if we do X, Y will go up" hypotheses
03 · PrioritizeICE score: Impact · Confidence · Ease
04 · ExperimentSmall and fast, like A/B tests — learn from results

A good experiment hypothesis

EXP-001
Hypothesis: if we show the review count on product cards,
            more people will open the detail page
Metric: card click rate (side effect: bounce rate)
Success: click rate +10% or more
ICE score (example)Impact8Confidence6Ease9Average 7.7 → test first

Today's key insight

  • The finished screens we see = the survivors of many experiments
  • Failed experiments and abandoned versions don't show on screen
  • AI doesn't know these experiment records and data → it gives an average answer
HANDS-ON 0515 minTeam · ICE

Team experiment design: pick #1 with ICE scores

GoalJiji buyers worry about scams, so they don't contact sellers. Pick 1 experiment to increase contacts
IImpactIf it works, how much
will contacts grow?
CConfidenceIs there evidence
(OBS · interviews)?
EEaseCan we do it now with
little time and cost?
FormulaICE score = (I + C + E) ÷ 3  →  test the highest-scoring idea first

Ideas · scores (1–10)

IdeaICE

Live ranking

1
Show verified-seller badge in the list
–
2
Safety-tips popup before first contact
–
3
Idea 3
–
4
Idea 4
–
5
Idea 5
–
0/4
Tick the boxes
to check yourself
Career linkExperiment hypothesis docs are the core of a growth marketer or PM portfolio. "We set priorities with evidence" is much stronger than "we have lots of ideas".
Benchmark targets · Choose one

Services to break down today: Jumia · YouTube · Jiji

Choose just one service, one typical user and one typical task.

★4.5★4.4★4.3OFFICIAL-40% Jumia · jumia.com.ngGeneral online storeFind product → check trust → buy → deliveryWhat to look at: flash sales, official stores, ratings and reviews, pickup stations and delivery info
NEXT YouTube · youtube.comVideo platformDiscover → watch → next video → return visitWhat to look at: recommended feed, autoplay, subscribe/notifications, Shorts, comments
₦ 45,000Verified₦ 12,500₦ 230,000 Jiji · jiji.ngLocal trading marketplaceChoose area → find listing → check trust → contact sellerWhat to look at: area and price filters, seller feedback and verification, safe-trading tips, paid top listings
Write internal algorithms, revenue and policies as hypotheses, not facts, unless there is an official source.
Mission 01 · Observe

Find growth features by observing: fact ≠ interpretation

First visit 30 minTypical task 45 minReturn features 30 min0 min30 min75 min105 min
First visit · 30 minLogged out: record your 5-second impression, value proposition and CTA
Typical task · 45 minRecord clicks, scrolls and inputs, trust elements and drop-off points
Return features · 30 minWhere saving, subscribing, notifications, reviews and sharing lead to the next visit

Observed fact (OBS)

  • "Product cards show the rating and number of reviews."
  • Record date · URL · device · screenshot together

My interpretation (hypothesis)

  • "It seems to reduce buying worries before opening the detail page."
  • Mark which AARRR stage it affects
Reverse engineering workflow

Observe → Document → Build with AI → Compare

The 4-step mission of Real Benchmarking PBL. Every decision links to evidence by ID.

01 · Observe10+ OBS · AARRR map · growth loop · EXP hypothesisOBS-###
02 · DocumentOPPORTUNITY · BUSINESS_MODEL · BASE_ARCHITECTUREPRD · BM · ADR
03 · Build with AIUse the docs as a contract and build 1 core flow with AIAI-### work log
04 · CompareOriginal vs build — prove the gap with the same task and conditionsCMP-###
OBS-001Observed factsPRD-001RequirementsBM-001Business modelADR-001Design decisionsCMP-001Comparison resultsEvery comparison result can be traced back to its observed evidence
Why AI can’t rebuild it the same way

AI builds from only the tip of the iceberg

What AI seesScreens · UI · featuresHistory of experimentsData · networkOperations · trust policyUser contextBusiness limitsDesign intent (Why)What AI can't see
What's visible (what AI builds)Screen layout · buttons · cards · common features · familiar patterns
History of experimentsOrder, wording and placement that survived thousands of A/B tests
Data · network effectsValue users built up, like reviews, listings and watch history
Operations & trust policyInvisible operations like moderation, reporting, seller verification and official-store guarantees
User contextWho comes in, when, and with what worries
Business limitsDecisions shaped by fees, ads, partners, laws and regulations
Intent (Why)The design intent behind "why this feature is here"
What isn't in the prompt doesn't exist for AI. So the result becomes a "plausible average".
HANDS-ON 0610 minTeam · game

Guess what's under the iceberg: what's behind this feature?

GoalTrain your eye to find the 6 hidden elements behind one feature on the screen
ExperimentsMany A/B tests
DataRecords users built up
OperationsModeration · reports · verification
ContextWho · what worries
ConstraintsFees · deals · law
IntentWhy is it here
1
Look at the feature1 feature on screen
2
Team discussionPick hidden elements
3
Check answerHit · missed · wrong
4
Note it for the promptUse in the next lab
JijiVerified-seller badge
Pick the hidden elements (more than one is OK)
Operations ID checks · moderators · report handling
Context Used-goods buyers worried about scams
Intent A signal that "it's safe to contact"
JumiaFlash-sale countdown
Pick the hidden elements (more than one is OK)
Experiments Results of testing timer position and wording
Constraints Brand discount deals · stock
Intent Urgency: "buy now"
YouTubeAutoplay next video
Pick the hidden elements (more than one is OK)
Data Recommendations learned from watch history
Intent Keep watching the next video
Context Worry about data costs → option to turn off
0/4
Tick the boxes
to check yourself
Startup tipWhen you start a business, anyone can copy your screens. Your edge comes from below the iceberg — operations, data and understanding local users.
Prompt = your analysis

Prompt quality = depth of growth-hacking analysis

Request without analysis

Build me an online store like Jumia.
With a product list and a cart.
  • Result: an average store you've seen everywhere
  • No growth or trust features
No trust infoSeen everywhereAverage card

Request with analysis

Role: Next.js development partner
Goal: first-time visitors see trust info and open the detail page
Evidence: OBS-003 (card rating & review count), OBS-007 (delivery & returns info)
Docs to follow: PRD-001 / BM-002 / ADR-001
Done when: empty and error states, 360px mobile
Don't: copy original logos, images or text; add new features
First, propose a plan and the files to change.
★4.6 (200)Free returns★4.5 (160)Free returns★4.4 (120)Free returnsOBS-003 appliedRating · review countOBS-007 appliedDelivery · returns info
HANDS-ON 0720 minSolo · AI tool

Rewrite the prompt: BEFORE → AFTER

GoalCompare for yourself how the AI result changes when you add today's evidence (OBS · EXP)
BEFORE-01 · no analysis
Build an app like Jumia. Product list and cart.
AFTER-01 · with analysis
Goal: first-time visitors see rating and review count and open the detail page Evidence: OBS-003, OBS-007 Experiments: EXP-001 verified-seller badge Completed: 360px mobile · error states Don't: copy original logos or text
1
Get your BEFOREHands-on 01 screenshot
2
Fill the blanksWith OBS · EXP numbers
3
Put it into AISame AI · same conditions
4
Compare countsBEFORE vs AFTER
BEFORE
AFTER
Enter the number of trust features (0–5) to compare

Full prompt · live

0/4
Tick the boxes
to check yourself
Career linkA good prompt is a good planning document. You can use it as-is when you outsource development or work with a team.
Core experience · GAP experiment

My design vs the AI result:
where did it come out differently?

① Design firstBefore using AI, draw the core flow and 3 growth features yourself
② Rebuild with AIRequest the same thing in a prompt and get the result
③ Record the GAPFor each feature, write "what I designed / what AI built" side by side
④ Find the causeWhy is it different? Missing from the prompt? AI averaging? My design error?
My design (before AI)AI resultExample · Jumia (for illustration)Rating & review count on product cardShows rating & review countRetentionDelivery info at bottom of detail pageDelivery & returns summary on the cardImprovedOfficial-store badge(none) No seller typesLostFlash-sale countdownGeneral discount % bannerChanged
Kept The design intent was built as-is Improved AI suggested a better option (check the evidence) Lost A growth feature was dropped or replaced by a common pattern Changed A different form than intended — a person decides
Day 2 mission

Today's Outputs

Output 1 · OBS
At least 10 observations (fact/interpretation separated, with screenshot and URL)
Output 2 · AARRR map
Features for 5 stages + growth loop + loop stop condition
Output 3 · AI rebuild
1 core flow, with the prompt and AI-### work log
Output 4 · GAP table
At least 5 design vs AI differences + causes + next experiment
  • Fix scope: 1 service · 1 user · 1 task
  • Don't copy original logos, images, text or code
  • Use only public screens and official materials
  • People check and record any claim AI makes
Takeaway

AI is the fast hands,
growth hacking is the eyes and brain.

AI may seem like it can do anything, but it won't tell you what to build or why. The people who see the growth structure clearly use AI best.

AI · fast handsSpeed · building+People · eyes and brainDirection · evidence · judgmentA growing product ↗
Eyes to seeFind the growth features and loops behind the screen
Power to explainCommunicate your analysis clearly in docs and prompts
Responsibility to judgeAccept, change or reject AI results based on evidence
1 / 25
LECTURE NOTE

Hands-on Lab

Today's Lab Flow

See detailed missions, templates, checklists and grading in the Real Benchmarking PBL Workspace.

STEP 01 · 40 min

Choose a service & observe

Pick one of Jumia, YouTube or Jiji and find growth features in 3 observation sessions.

OBS-001 ~ 010
STEP 02 · 30 min

AARRR & growth loop

Place your observations in the 5 stages and draw the repeating loop and its stop condition.

AARRR · LOOP
STEP 03 · 60 min

My design → AI rebuild

Design it yourself first, then have AI build the core flow with a prompt that includes your analysis.

PRD · AI-###
STEP 04 · 40 min · Key

GAP comparison & retro

Sort the differences between your design and the AI result into kept, improved, lost or changed, and write the causes.

GAP-### · CMP

GAP Record Template

Write one block per feature. At least 5.

## GAP-001 — [growth feature name]
- Evidence: OBS-___
- AARRR stage:
- What I designed:
- What AI built:
- Type: kept / improved / lost / changed
- Cause of difference: missing from prompt / AI averaging / my design error / other
- Human decision: accept / change / reject + reason
- What to add to the prompt next time:

Retro Questions

Answer these together in a team discussion.

1. How many growth features did AI make "on its own"?
2. Which features were missing because I didn't state them?
3. Where did AI swap in a common pattern?
4. What couldn't you learn from the original service's screens alone?
5. If you did it again, what else should you observe?
6. Prove "AI is not magic" in one sentence using our results.

10-Day Course Navigation

Click the buttons above to go to the Day 1–10 learning pages.

Learning Workspace

Growth Hacking Videos & Day 2 AI Helper

Pick a reference video by topic on the left, and ask the AI chat on the right about growth hacking and the reverse-engineering lab.

Related YouTube VideosDay 2 recommended materials
  • Growth hacking basics
  • 1. Growth hacking in 10 minutes: from definition to processIntro to growth hacking · concepts and work process KR
  • 2. Growth hacking basic conceptsKey terms used in real meetings KR
  • 3. Growth hacking – building services that grow through data and experimentsData- and experiment-driven growth methods KR
  • 4. Growth hacking from a working growth hacker (the customer is the answer)Industry view · why understanding customers comes first KR
  • 5. Growth Hacking: The Science of Growth with Sean EllisThe science of growth, explained by Sean Ellis who coined the term EN
  • 6. The original growth hacker reveals his secrets | Sean EllisLenny’s Podcast · the PMF 40% test and growth experiments EN
  • AARRR · metrics
  • 7. Growth hacking AARRR and funnel analysis5-stage metrics and funnel drop-off analysis KR
  • 8. Startup marketing: succeed with AAARRR metricsUsing stage metrics from a startup view KR
  • 9. Pirate Metrics for Startups - AARRR Framework by Dave McClureThe framework explained by the creator of pirate metrics EN
  • Growth cases
  • 10. Growth hacking cases and methods – from startups to solo businessesCases at many different sizes KR
  • 11. How Did Airbnb And Dropbox Succeed With Surprisingly Simple MVPs?Early MVPs and growth features of Airbnb and Dropbox EN
  • 12. Case Study: How Airbnb Used Growth Hacking to Overcome ChallengesAirbnb growth-hacking case analysis EN
  • Design reference (earlier Day 2 videos)
  • 13. Intro guide to UML Use-Case & Sequence DiagramsDocumenting reverse-engineering · drawing flow charts EN
  • 14. Data Flow Diagram (DFD) concepts and design lab (Level 0/1)Reverse-engineering data flow EN
  • 15. System Design: core architecture and client-server flowReference for writing BASE_ARCHITECTURE EN
  • 16. System error handling and Fault Tolerance designReference for designing failure states EN

Videos are embedded public YouTube videos. If playback is restricted by the uploader, open them directly on YouTube.

Day 2 · Daily Feedback

Today's Course Satisfaction Survey

After Day 2 class, please answer a short survey. Your answers help us improve the next class and the course.

Open Day 2 SurveyJoin with the button or QR code
Day 2 Daily Feedback Survey QR codeSCAN TO OPENNigeria AI Training · Day 2