What students learn
1. Find the biggest leak first
Squads can fix only two of six problems, so they must rank them. A heavy video and a vague headline lose far more visitors than a quiet logo. Buying more ad traffic pours more water into the same leaky funnel, and it costs money every week.
2. Measure the stage you fixed
The +2 bonus after each round asks which number proves it: engagement rate (GA4) for the homepage, checkout completion for the checkout, repeat purchase rate for loyalty. Followers, time on site, email opens and chatbot chats are the tempting wrong answers.
3. Surprises kill checkouts, and some are illegal
Fees that appear at the last step and forced accounts drive abandonment. A fake countdown lifts sales a little, but it is misleading, and in the UK drip pricing has been banned since April 2025.
4. Bots for speed, people for judgement
Order tracking, a clear policy and refill reminders suit a chatbot. A smashed jar and an accessibility need deserve a person. The business is responsible for what its bot says.
Built for the customer experience and website evaluation week of a digital marketing or e-commerce module, and for sustainability marketing classes. It complements Site Inspector (an in-depth audit of one website) and Happy Returns (a service journey and Net Promoter Score).
Five golden rules
- Reloading is safe.The game remembers where you are in this browser tab. Closing the tab clears it.
- Two fixes per squad in rounds 1 and 2.Tap a squad on the right, then tap its two fixes on the phone or in the list. The next squad lights up by itself. Tapping a fix again removes it; a third tap swaps out the oldest pick.
- Round 3 answers go on screen.Tap a squad, then its Bot or Human answer for each of the five messages. Each squad's answers fill in on the right for everyone to see, and the next squad lights up by itself. Scores are counted automatically when you reveal.
- Points.Each round: 10, 7, 5, 4, 3, 2, 1, 1 by rank, plus 2 for the right metric (squads hold up 1, 2 or 3 fingers; you tap the ones who got it). The twist then gives each squad +3, 0 or −3. Ties go to the 12-week value.
- "Back" undoes a reveal.If a squad's picks or score were entered wrongly, go back, fix them and reveal again.
The model
Engaged = visitors × engagement rate · baskets = engaged × 12% · orders = baskets × checkout completion
Customers who come back = orders × repeat rate (each buys twice more within 12 weeks)
12-week value = (orders × £24 + returners × £48) × 12 − ad spend × 12 − one-off costs
Before the game Fill Good turns 1,000 visitors into 350 engaged visitors, 42 baskets, 10.5 orders and 1.6 returning customers a week: a 12-week value of £3,931.
Round 1 · Spot the leak (scored on engaged visitors; engagement starts at 35%)
| Leak | Fix | Effect |
|---|---|---|
| 1. 8.4 MB autoplay video | Light photo of the jars | +12 points |
| 2. "Welcome to our journey ✨" | Say what you sell in one line | +10 points |
| 4. Cookie wall with no Reject | Small bar, equal Accept and Reject | +6 points |
| 3. "Join our tribe!" pop-up | Remove it | +5 points |
| 5. Lowercase logo | Bigger, bolder logo | 0 (not a leak) |
| 6. "Not enough people visit" | £300 a week for 500 ad visitors | +75 engaged, −£3,600 over 12 weeks |
Best pair: 1 + 2 = 570 engaged. Then 1 + 6 = 545 (but it costs £300 a week), 1 + 4 = 530, 2 + 6 = 525, 1 + 3 = 520. Worst: 3 + 5 = 400. Right metric: 2. Engagement rate (GA4).
Round 2 · Checkout surgery (scored on checkout completion; starts at 25%)
| Problem | Fix | Effect |
|---|---|---|
| 1. Deposit and delivery appear at step 3 | Full price on the product page | +12 points |
| 2. "Create an account to continue" | Guest checkout | +10 points |
| 3. Card only | Apple Pay, Google Pay, PayPal | +6 points |
| 6. "Nobody feels in a hurry" | Fake "Only 2 left! 09:59" | +5 points, punished by two twists |
| 4. Four checkout steps | Cut to two | +4 points |
| 5. Grey Continue button | Big green button | +1 point |
Best pair: 1 + 2 = 47%. Then 1 + 3 = 43%, 1 + 6 = 42%. Worst: 4 + 5 = 30%. Right metric: 1. Checkout completion rate.
Round 3 · Bot or human? (scored on repeat purchase rate = 13% + 5 points per right answer)
| Message | Answer | Why |
|---|---|---|
| 1. "Where's my order? It said Tuesday" | Bot | Factual lookup, instant, 24/7 |
| 2. "My jar arrived smashed…" | Human | Upset customer, judgement call |
| 3. "What's your returns policy…?" | Bot | Only by quoting the real policy page (Air Canada) |
| 4. "I use a wheelchair and can't get to the post office…" | Human | Accessibility need: a flexible, respectful answer |
| 5. "Can you remind me when I'm about to run out?" | Bot | Automatic refill reminders bring customers back |
5 right = 38% repeat purchase; 0 right = 13%. Right metric: 3. Repeat purchase rate. Message 3 sparks the best debate: if a squad argues well for "human", press Back to the inbox and change its answer before you go on.
The twist wheel (+3, 0 or −3 for every squad)
| Twist (real case) | +3 | −3 |
|---|---|---|
| 🤖 The chatbot made up a refund policy (Moffatt v Air Canada, 2024) | 4–5 right in round 3 | 2 or fewer right (and £400 in refunds) |
| ⚖️ The regulator checks your prices (DMCC Act, 6 April 2025) | Full price shown early, no fake timer | Fake countdown (and £1,000) |
| 🔌 The website is down for a week (M&S cyber attack, 2025) | 4–5 right in round 3 | 1 or fewer right |
| 🛒 An AI shopping agent comes to buy (ChatGPT Instant Checkout, 2025) | Full price shown early (+15% orders) | Fake countdown (−10% orders) |
Design assumptions and simplifications
The same list appears in the game under "How this game simplifies reality".
- Every number is invented for teaching.Visitors, engagement, add-to-basket, checkout completion and repeat rates are round, plausible figures chosen so that good decisions win clearly. Real figures vary by sector, device and season.
- The funnel follows RACE.Reach, Act, Convert, Engage (Chaffey and Smith). Add-to-basket is fixed at 12% so that each round isolates one stage. Ad visitors are colder traffic, so only 15% of them engage.
- Engagement and bounce follow GA4.An engaged session lasts 10 seconds or more, has 2 or more page views or a key event; bounce rate is the share of sessions that were not engaged.
- Page speed is part of CX.Core Web Vitals measure loading (LCP, good at 2.5 seconds or less), responsiveness (INP, which replaced FID in March 2024) and visual stability (CLS). Google retired its Mobile-Friendly Test in December 2023 and points to Lighthouse and PageSpeed Insights.
- Checkout effects echo research, not exact figures.The Baymard Institute's studies put average cart abandonment at about 70%, with extra costs at checkout the most common reason and forced account creation also high on the list.
- Round 3 is a proxy.Repeat purchase rises by 5 points for each message sorted correctly. It stands for how well the service is designed, not a real formula.
- The law is simplified.Drip pricing is banned under the UK's Digital Markets, Competition and Consumers Act 2024 from 6 April 2025; fake urgency and false scarcity are misleading practices. The costs in the twist are illustrative, not legal penalties.
- The twists are stylised.The real cases are summarised accurately; their effects on Fill Good are invented. Fill Good is fictional.
Results you can predict
Every result is fixed by the picks and scores; only the twist is random. Squads don't affect each other's numbers, only each other's rank.
- The best path is 1 + 2, then 1 + 2, then 5 out of 5, with the right metric every time: 12 points a round, plus +3 from any twist (39 in total), and a 12-week value of about £15,000 to £19,000 depending on the twist, roughly four times Fill Good's starting £3,931.
- Buying ads often comes second or third in round 1 (545 engaged with the video fix), which is the point: it wins on a traffic number but loses £3,600 over 12 weeks. Its 12-week value is usually well below the squads that fixed the page.
- The fake countdown looks clever in round 2 (1 + 6 = 42%, close to the best), until the regulator or the AI agent lands on the wheel.
- The logo and the grey button are the "pretty" traps: they feel like design improvements but plug almost nothing.
The screenshots below come from one four-squad game: Bubbles (1 + 2, 1 + 2, 5/5), Suds (6 + 3, 6 + 5, 2/5), Splash (5 + 4, 5 + 3, 3/5) and Drip (1 + 4, 1 + 3, 4/5), with the chatbot twist. Final: Bubbles 39, Drip 30, Splash 14, Suds 12.
How long will it take?
Every time in this guide is a range. The low end is a tight run; the top end is what most classes take once squads discuss, decide and react to the results.
In class: about 15 minutes of play; plan 15 to 25Solo: 8 to 12 minutes at the player's own pace
What moves you up or down the range
- How strictly you run the clock. Stopping each round when the timer ends (2:00, 1:30, 1:30) keeps you near 15 minutes; letting squads finish talking moves you towards 25.
- Number of squads. More squads means more picks to tap in and more results to read out.
- Round 3 entry and reveal. Five taps per squad, then five messages revealed one at a time. Asking a squad to defend its answer before you reveal adds a minute or two, and is worth it.
- Interaction in between. Cheering at the results, the metric bonus, reading the twist and its real case aloud, questions.
- The debrief. One question, or a full discussion of the final funnel.
- Your group. Familiarity with the topic, reading in a second language, and the room layout.
Short on time? Use the timers, keep to four squads, and take one debrief question. Want the richest discussion? Let squads argue about message 3 (the returns policy) before you reveal it.
Step by step
0 Set the scene and form squads
1 to 2 min
The client, the leaky funnel, the three mini-games and the winning rule on one screen. - Click
- Set the number of squads with − and +; type names if you like.
- Full screen, then Start round 1.
- Say
- "You're the CX team for Fill Good. 1,000 people visit the website every week and about 10 buy. Where are they leaking out? Three rounds, three different games, and a twist at the end."
1 Round 1 · Spot the leak
3 to 5 min · timer 2:00
Numbered leaks on the phone, the fixes in the list, and each squad's two picks on the right. - Click
- Start timer. When time is up, each spokesperson calls out two numbers. Tap the squad, then its two numbers (the next squad lights up). Then Reveal: who stayed?
- Watch for
- A squad that picks 6 (buy ads) and does well. Ask: "What will that cost you every week, and did the homepage get any better?"

Each fix is labelled as a big leak, a medium leak or not a leak. On the right: the +2 metric bonus. - Click
- Squads hold up 1, 2 or 3 fingers. Show the answer, then tap the squads that got it right (+2 each).
2 Round 2 · Checkout surgery
2.5 to 4 min · timer 1:30
£18 becomes £26.45 at the last step. Cut two problems. - Say
- "These people wanted to buy. What stopped them? You can only operate twice."
- Watch for
- Option 6, the fake countdown, is tempting. After the reveal, ask: "Was the stock really running out? Is that allowed?" Then read the real-world box on drip pricing.
3 Round 3 · Bot or human?
3 to 5 min · timer 1:30
Tap a squad, then its five answers. Everyone sees each squad's answers fill in on the right. 
One message at a time: Reveal the answer, and every squad's answer turns green (right) or red (wrong). 
Scores are counted for you. Then see the results and the metric bonus. 4 Spin the twist wheel
1.5 to 3 min
The wheel slows gently (no flashing). Every squad sees +3, 0 or −3 and why. - Click
- Spin the wheel. Read the real case aloud, then To the podium.
5 Podium, then watch the funnels
1.5 to 3 min
Jar trophies and falling droplets (no flashing or shaking). 
The visitors walk each squad's funnel. - Say
- "Look at the right-hand column. The squad that bought the most traffic is not the squad that made the most money. Small fixes at every stage multiply."
Solo mode

Share lab.yingwu.co.uk/leaky-funnel/#solo for homework or revision. The rivals always make the same choices: Traffic Hogs buy more traffic and push hard, Pretty Pixels make it look amazing, and Journey Nerds fix the basics and measure them, but not perfectly. Finishing any game earns the "Leak fixer" badge; beating all three rivals in solo mode earns "Golden jar".
Debrief questions
- Which leak would you fix first on a real website you use, and why? Good answers name a problem, the stage it hits (Act, Convert, Engage) and the metric that would show the fix worked. Use them to rehearse the website evaluation criteria from the lecture: aesthetics, mobile and speed, accessibility, usability, content and trust.
- Which metric would you put on Fill Good's dashboard first, and which is a vanity metric? A RACE dashboard has one or two measures per stage: engagement rate (Act), checkout completion or conversion rate (Convert), repeat purchase rate and jars returned (Engage). Followers, raw visitors, email opens and chatbot chats are vanity metrics unless they connect to sales.
- Which customer messages would you never hand to a bot? Emotional complaints, judgement calls (refunds, exceptions) and accessibility needs. Bots suit fast facts and routine tasks, but the business is responsible for what they say (Moffatt v Air Canada), and Klarna began hiring human agents again in 2025.
- What changes if your next visitor is an AI agent? Clear, complete prices, accurate product data and clean pages matter even more; pop-ups and fake urgency get in the way of robots as well as people.
Cheat card: the button order
- Set squads, Full screen, Start round 1
- Start timer (2:00); tap each squad, then its two leaks; Reveal: who stayed?; Show the answer; tap the +2 squads; Next round
- Start timer (1:30); tap each squad's two fixes; Reveal: who checked out?; bonus; Next round
- Start timer (1:30); tap each squad, then its 5 Bot/Human answers; Lock answers and reveal; for each message: Reveal the answer, Next message; See round 3 results; bonus
- Spin the twist wheel; Spin; read the real case aloud; To the podium
- Point at the funnels and the 12-week value; ask one debrief question
Behind the rules and the real cases
- Chaffey, D. and Smith, P.R. (2022) Digital Marketing Excellence. 6th edn. Abingdon: Routledge. (RACE: Reach, Act, Convert, Engage.)
- web.dev (2024) Interaction to Next Paint becomes a Core Web Vital on March 12. web.dev
- Search Engine Land (2023) Google officially drops Mobile Usability report, Mobile-Friendly Test tool and Mobile-Friendly Test API. searchengineland.com
- Baymard Institute (n.d.) Cart abandonment rate statistics. baymard.com
- CMS (2025) The DMCC Act: consumer elements come into force from 6 April 2025. cms.law
- Competition and Markets Authority (2025) CMA secures changes from Ticketmaster following Oasis tickets investigation. gov.uk
- Moffatt v. Air Canada, 2024 BCCRT 149. Summary: Business Law Today (2024). businesslawtoday.org
- Customer Experience Dive (2025) Klarna changes its AI tune and again recruits humans for customer service. customerexperiencedive.com
- The Register (2025) M&S warns of £300M dent in profits from cyberattack. 21 May. theregister.com
- Insurance Journal / Reuters (2025) After 46-day cyberattack pause, British retailer M&S resumes online orders. insurancejournal.com
- TechCrunch (2025) OpenAI takes on Google, Amazon with new agentic shopping system. techcrunch.com
- Digital Commerce 360 (2026) Shopify integration with ChatGPT changes. digitalcommerce360.com
A simplified classroom model: all rates and figures for Fill Good are illustrative teaching devices. Fill Good is fictional.