Tizzy makes Roblox games that pull tens of thousands of players and six figures a month, and explains how on YouTube. We took 20 of those videos apart, merged what he repeats, kept his exact numbers and his own words, and arranged it in the order you actually use it. Every card links to the video it comes from. Not affiliated with him; all credit is his.
Never invent a loop from nothing and never clone one 1:1. Take a loop players already understand, then change roughly a fifth of it — a mechanic, a theme, a market — so it is familiar but not a copy.
Only build where you can already point at players who want this. Brand-new markets with no proven audience are the riskiest thing you can do.
Title, thumbnail and art direction are what players judge before they ever load the game. Two games with the exact same loop live or die on packaging alone.
Real player data beats anything you imagine. Launch the MVP before it feels ready and then patch daily — sometimes twice a day — during the first week.
Pick your worst number, guess what will fix it, ship exactly that, and measure. If the stat does not move, revert and guess again.
Roblox ranks engagement above monetization, and pay-to-win games die fast. Make the game fun and sticky first, then add money on top of it.
Most of the platform is young kids and about 80% of them play on mobile. Every mechanic must be understandable without reading and usable with a thumb.
Tizzy gives ideas away on purpose because they are abundant. The real bottleneck is finding talent that ships on time and polishing what everyone else leaves broken.
Cross-reference qualitative feedback, dashboard numbers and competitor research before every big decision. Three agreeing sources make your guess strong enough to bet on.
Before anything else, name the fantasy your game sells: the role, the world and the feeling. Picture what a 7-year-old daydreams about before sleep — animals running a hospital, being a firefighter, a tiny person in a giant kitchen.
Old Flash games, Minecraft mini-games, Garry's Mod modes, Steam hits and console games are loops that already proved people love them. Take one that costs money or needs a PC and make it free on Roblox.
Addicting Games, Coolmath Games, Crazy Games, Newgrounds, Xbox Game Pass and the iOS/Google Play charts are Tizzy's idea sources. Crazy Games is his favourite because many of its games are already 3D Unity builds.
Coal took Garry's Mod's Prop Hunt — a $20 Steam game — made it 6v6 and free on Roblox. Kids who had no PC and no $20 had been asking for it for years.
Scroll TikTok, find a viral challenge, and build it as a mini-game so kids can do it with friends who live too far away. These are Tizzy's favourite ideas because dev time is tiny.
Bomb Click, the Lying Challenge and Rubber Band Watermelon are all TikTok challenges rebuilt as Roblox mini-games. Simple to build, and they ride a cultural moment straight to huge reach.
Instead of inventing mechanics, take a game that already works and reskin it with whatever content is trending. You can also strip the scary or adult parts off a proven loop to reach kids.
Look for fantasies with hundreds of millions of visits where the top game is years old and stuck at 1K CCU, or where every game is the same genre. Bring a trendy genre — checkpoint, extraction, clean-the-blank — into that fantasy.
The airport fantasy was all roleplay games. One dev made a Papers Please style checkpoint game inside it and had zero competition. Tizzy called it at 50 CCU and it grew to 30K.
Blue ocean means you are first, red ocean means everyone is already there, purple ocean means huge demand but everyone executes badly — that is where you want to be. Hide and seek, airport and firefighter fantasies were all purple oceans. Blue oceans also come from serving an audience nobody serves, like first responders in Emergency Roleplay.
Different fantasies have different kids. A loop that is beaten to death in the brain-rot crowd can be brand new to the SCP crowd, so the same clone that dies in one market wins in another.
Loops from 5-10 years ago still work if you modernise the art, the UI and mobile controls. Even better: find a live game with a good loop, a community page full of bug complaints and a dev who stopped updating — those players are waiting for you.
Copying a big trending game or a modded fad gets you a month before it dies, and the same kids play every brain-rot, squishy and lucky-block variant so you are fighting for the same playtime. The opposite mistake is just as bad: a fantasy nobody plays yet has no demand to capture.
One side: Tizzy openly reskins hits — he saw Steal Cookies and made Steal Sigma's RZ, and Coal reskinned My Singing Monsters. Other side: he also says template games and straight clones never work long-term and the new algorithm punishes derivative work. His resolution is a meaningful twist to the core loop, but where 'meaningful' starts is a judgement call he does not fully define.
First question on the checklist: has this been done outside Roblox and did it work? Then find a countable proof — ratings, votes, wishlists, visits, or a big YouTuber who played it. Older proof is fine; if demand existed four years ago and nobody has made a modern version, it is still there.
The numbers Tizzy treats as real signals when scanning other platforms. Anything at or above these means the core loop is proven.
Never bet on one signal. Stack Steam wishlists plus YouTube views plus Creator Exchange visits plus creator coverage — the more sources that agree, the more confident you can be before you spend months building.
The site that shows what is blowing up right now: CCU over time, visits, revenue estimates and how often a game updates. Tizzy calls it a gatekept secret and found Alien Ranch with it. Its revenue numbers under-report, sometimes by half.
Filter for games created this month, then read the 90-day CCU chart for ones climbing but not yet at their maximum. Also check CCU against total visits — 560 CCU on 100k visits beats a huge visit count with nobody online.
Type a fantasy or keyword and sort by visits to see who owns that space, how saturated it is and which sub-genres nobody has taken. Also the fastest way to study the thumbnails of the top games in your category.
On any game's Roblox detail page, this section shows what else that audience plays. If your loop already appears there, those players know it; if it does not, you have found an untapped market.
Borrowed from e-commerce: name exactly who plays this, what else they play and what they want. Then ask Cole's three questions — is the idea clickable, is it relatable, does it already have a passionate fan base?
Read the reviews of similar Steam games and the subreddits and Discords for that fantasy. You are looking for what they praise, what they complain about and the sentence 'why is every X just a Y?'
Ten minutes tells you nothing. Play at least two hours: hit the progression walls, see the mid-game, try the monetization, and write down the core loop, the progression and the price points. Matia studied games 12 hours a day before his breakout.
Pick rising games off RoTrends, log each one in a Google Sheet with your notes and a predicted CCU ceiling, then come back in a week or a month and check whether you were right. When you are wrong, name the exact bottleneck — social, first-time experience, bugs, or mid-game. This costs nothing and builds real design intuition.
Tizzy found it on its first spike at 1.2k CCU. Deep mechanics and freedom, but game-breaking bugs, no progression and a weak first-time experience — so it stalled. It still makes about $500/month with no updates, which proves the loop, and Farmer Die was built from it.
Scope only the minimum viable product — core loop plus a bit of progression, no live ops, no seasons. Get it running with AI, parts and free models yourself, so you know the game is fun before you pay anyone.
Hand it to your little cousin on their phone and say nothing. Every place they hesitate is a friction point. Most devs have never even opened their own game on a phone, and that is where the majority of players are.
If you plan to build it entirely with AI, the genre must pass three checks: low dev time, high demand with low competition, and at least one successful vibe-coded game already in it. Quiz, clicker and simulator pass; Action RPG is underserved but too complex to vibe code.
Put the whole game on paper: what players do, how they level up, the economy, the progression. Then hand it to AI and ask what the programmer specifically has to build — that becomes your task list.
Post what you are building and what you are learning until developers trust you, then post the job. With a reputation you can hire the same day, sometimes within an hour. A top dev told Tizzy he will not hire anyone with fewer than 30 mutual Discord connections.
Two to four people who each go deep in one skill and wide across others — a programmer who also does UI and animation, a builder who also models. Keep them in your time zone so a request is answered in hours, and pair a producer who plans with a dev who builds.
Never work with someone who will not sign. The SOW lists every frame, model and UI system so nobody argues later. Pay on completion, not upfront, and do not add work mid-project — if you want more, write a new contract.
Money without reputation does not work in a clout-based space. Tizzy's first game cost $4K and never launched; Farmer Die cost $12K with full-time staff and peaked at 1.8K CCU. Give applicants a paid sample task, and fire people who miss what they promised — if the team is slow, that is on you.
Break the design document into Trello columns per role with explicit tasks. Explain each one in a Loom video with pictures and a link to a live game that already has that system — video beats text, and if they build it wrong your brief was unclear.
Phase 1 is builds, models and UI, finished. Phase 2 is the programmer walking into a world that already looks real and wiring the core loop into it. Nobody is motivated by an empty baseplate full of free models.
Once the MVP exists, download a backup and tune the feel yourself with AI for two or three rounds. Only then tell your programmer exactly what changed, so he verifies it works in the real codebase instead of doing five rounds of guesswork.
Every system you add — upgrade board, buy-one/buy-max buttons, locking, plots — should already exist in a game with billions of visits, so players know how to use it on sight. Blend two or three of those games and change roughly 20% to get your own identity.
Understand why a mechanic works in its original genre before you paste it in. Build a Base and Steal copied auto-rolling from AFK games; kids hit end-game content in 3 hours and the economy broke, so they deleted auto-roll and roll speed entirely instead of tuning them.
One simple mechanic that interacts with another creates endless strategy — building plus stealing, or voxel building in Build a Plane. Complex feature stacks just confuse kids. Small tight maps and level design that creates gameplay beat big empty worlds.
Show, do not tell: press the button, see the result immediately. Kids do not read tutorials. Taking an adult game and dumbing the mechanics down until a kid gets it instantly is a real skill top devs have.
Within the first 60-120 seconds a player must know what the game is, how to progress and what to be curious about next. By 5 minutes they should already feel like they are getting somewhere. Then keep curiosity loops open — what unlocks next, how big can this get, and RNG rolls that tease rare rewards you cannot have yet.
About 80% of Roblox is mobile and 30% of Build a Base and Steal's players are on console — console players also play often and for long sessions. The core mechanic has to be frictionless with a thumb: voxel building is clunky on touch, which Tizzy thinks costs Build a Base and Steal 100-150k CCU. Fixing console UX measurably raised their D1.
One annoying thing is survivable; five to ten small ones are not. Spam-clicking to upgrade, slow transitions, broken proximity prompts, unclear buttons — they compound. UX obsession is the difference between a 1K CCU game and a top-100 game, and most devs are simply too lazy to do it.
Interaction with other players should not be optional — PvP or co-op belongs at the centre of the game, like Missiles vs Cities or 99 Nights. If the game is 10x more fun with friends, players recruit their own friends for you.
Picture a YouTuber recording your game and build the moments they would cut: ragdolling into the void, a stolen pet, an insane build. Then watch their videos and note exactly which seconds they clip — that is free advertising and it tells you what is actually fun.
Obby is the easiest, then simulator and tycoon; quiz and clicker are the safest for AI-built games. Modern AI can carry a whole game in these genres, so there is no excuse for never shipping.
One side: vibe-coded games prove kids forgive rough UI as long as the game is playable, not broken, and on-theme — Carpet Cleaning made $4K/month looking obviously AI-made. Other side: the June 2026 algorithm explicitly rewards high-quality unique games, and Tizzy says obsessive polish is what separates a 1K CCU game from a top-100 one. It depends on whether you are chasing a quick $4K/month or a flagship.
A bought template skips the learning, so when the game needs iterating you do not understand your own mechanics. Build it from scratch — or from a prototype you made — instead.
Built from Alien Ranch's loop with guns, NPC curing and a class system added. It cost $12K with a full-time programmer and modeler and still only peaked at 1.8K CCU — but months later, with no updates, it holds 80-150 CCU and pays about $800/month. Money did not make it big; the niche made it durable.
Sometimes the strongest design fights the fantasy. An outdoor cafe would show off other players' plots far better than an indoor one, but indoor is what a gaming cafe actually is. Tizzy says both choices are valid — you pick which one you are optimising for.
Use words a kid can picture — no 'data', no 'corporation'. Build from templates like Verb + Noun (Grow a Garden), Be + Noun (Be a Nurse) or Escape + Noun, and make the title imply a challenge, like 99 Nights in the Forest. Validate the keyword in Creator Exchange first: several high-visit games sharing it means the fantasy is strong, one is just luck.
'Sell Your Data' confused kids and got 1.54% CTR. Renamed 'Hack a Business' — a fantasy they instantly understood — CTR more than doubled to 3.27%. Its best thumbnail, a before/after split screen, did 60 million impressions, and cutscenes copied from competitors dropped bounce from 22% to 15-16%. The game reached 1K CCU and about $30K/month.
The thumbnail has to show what the player will feel — power, fear, outsmarting someone — and give context to the title so the two together explain the game. Tease the one thing that makes you different, like 'steal at night' or a visible buy-and-sell loop.
One or two characters, rarely three, with one very clear action and readable emotion on their faces. Working formats: before-and-after or Level 1 vs Level 999 for progression games, first person for 'be a ___' games, a swarm for chaos games.
Bright saturated colours on a simple background, clouds and clutter stripped out, logos scaled up huge, text on empty sky rather than over buildings. Use characters kids already recognise — a bacon, a noob, the red guy from Rivals, the McDonald's arches.
Free tool that previews your thumbnail as it will appear on mobile, desktop and console. The most common mistake Tizzy sees is a thumbnail that looks great on a monitor and is unreadable on a phone — this is the last gate before you go live.
Rough out the layout in Canva with screenshots of your assets, then feed that to an AI generator (he uses Row Thumbs) and iterate precisely — shorter hills, bigger buildings, clear blue sky. Be exact in the prompt or it gives you a fruit apple instead of an Apple Store. If it still misses after several tries, start a fresh chat. Budget 10-15 minutes per thumbnail.
The volume of creative Tizzy runs. At least 10 at launch, 20 if you want to win, 25 per ad campaign; he tested 50 across two campaigns for one game. Then 2-3 new ones every week forever, because CTR always decays as impressions pile up. AI-made thumbnails have never lost to a hired artist for him.
Roblox's own discovery guidance says repetitive titles and reused images rank lower, and clickbait like 'mysterious game' is now punished. Make your own thumbnails, icons and name, keep every personalised variant honest about the game, and fill the detail page with your best thumbnails showing different parts of the game — most devs ignore that page entirely.
Identical core loop to a wave of clones. It mutated the title and thumbnail and held 9.5K CCU while the clones sat at 31, 11, 2, 7 and 12 CCU. Same for fishing: Ice Fishing and Cage Fishing survived, generic Fishing Simulator died.
In Beta Mode your game is kept out of home recommendations, so you can pour ad traffic in and fix your numbers without burning your one shot at the algorithm. Big studios do this and spend heavily — 10K in a day — so they exit beta with top-tier stats.
Stop polishing the version in your head. Live players give you data that is worth more than any assumption, and the real work starts after launch when you update daily off what they actually do.
If the game is called Build a Base and Steal, they should be building within 10 seconds and feel the PvP threat immediately. Bounce rate is literally the gap between what your title and thumbnail promised and what the first minute delivers.
Find the single most engaging thing in your game and let them do it inside 30 seconds — building, planting, whatever it is. One mechanic at a time; teaching two at once overloads them. Use the mechanic itself to teach: making them lock their base signals a threat without a word of text.
Get players emotionally invested in the first 90 seconds — a cutscene or world-building beats a list of instructions, and cutscenes are low-key overpowered for bounce rate. Do not throw starter packs or 2x offers at them in those first moments.
In launch week, put out a new tutorial version daily and read the funnel and D1 after each one. Build a Base and Steal went through nine versions before the onboarding was right.
$16 a day for at least 3 days is the floor — below that the algorithm has no signal. Convert Robux into ad credits instead of paying USD, set a daily budget not a lifetime one, pick All Ages (it serves 16+ anyway) and accept every setting Roblox recommends.
Tizzy's real numbers: $16/day from May 20th, threshold passed in 11 days for about $180 total. Farm and Die's launch cost about $200, so the new rules did not make it more expensive. First home-recommendation impressions usually take a week of ads. If you cannot afford it, take commissions to fund it.
Run ads to get data, read your benchmark stats, find the single worst one, ship a meaningful update aimed at it, then run ads again to feed Roblox fresh data. The same day your stats populate, start updating — every idle day costs momentum.
Ads on an unimproved game buy traffic that bounces and teaches Roblox that your game is bad. Only spend after you have actually made the game better.
The algorithm cannot rank a game with no players, so you must bring traffic yourself — ads, TikTok and YouTube Shorts, or free UGC. The faster you feed it data, the faster it decides your fate, which is why studios dump ad budget in a single day.
Roblox ads optimise for the cheapest player, not the right one, so early stats look bad and some 'players' just stand in the lobby without moving. Do not panic-redesign your game off ad-traffic numbers alone.
Players from home recommendations already play games like yours, so they have higher D1, higher D7, longer sessions and they spend more. Once you are in, Roblox keeps learning from that organic traffic without you restarting ads.
Buying ads puts you in the sponsored market only; it does not buy home placement, which is earned with stats. Think of each genre as a farmers market where Roblox sends a fixed pool of players and the best-ranked games sit at the front.
Roblox first compares you to every game in your genre — including adjacent ones with similar mechanics — and turns your stats green or red. Once it has enough data it re-ranks you against games that share your exact audience, and that is usually when your numbers look worse.
Acquisition > Home Recommendations > Benchmark Experiences with Similar Players. This is the page that shows how you compare to your real competition instead of to the whole platform — do not judge yourself off the overview.
The algorithm judges you on a rolling 7-day window, and the dashboard itself lags — 24 hours for most updates, up to 3 days early on. Do not panic the morning after a big update; the ranking speed did not change with the June 2026 update.
Since the Kids/Select split, your game needs 500 highly engaged players before it shows to under-16s — a highly engaged player is anyone who spent money anywhere on Roblox in the last 60 days, not in your game. Roblox hides the exact thresholds so bot farms cannot game it. You can hit the algorithm before you hit 500, and home traffic makes the climb far faster: one dev needed 210k visits for 190 players, then got 160 more from just 90k visits in 5 days.
Play-through rate (what QPTR became), bounce rate, 7-day playtime per user, 7-day play days per user, D1, D7, and now D28 — plus monetization and conversion. Model your game as a funnel — impressions, clicks, plays, D1, D7, D28 — and fix the biggest leak.
Bounce rate is the share of players who leave within 60 seconds, and it is the one stat where a high number actively reduces your home recommendations. Roblox also tracks a second window, 61-180 seconds, so you can see exactly when they give up.
Roblox's stated priority is play-through, bounce and retention; money is secondary. The June 2026 changes exist to push high-quality games and bury clickbait, so a fun game that people come back to wins the ranking.
The moment a fantasy proves demand, dozens of devs ship Untitled Fishing Game, Fish Go, Catch a Fish — all competing on identical metadata, all dying. The mutated copies usually destroy the game that was first to market too.
One rule he gives: stop the moment home-recommendation impressions pass your sponsored impressions, because the algorithm has taken over and ads are wasted money. The other: keep running them after every meaningful update to feed fresh data, and if you can afford it just keep spending until you hit algo. In practice it depends on whether your last update was big enough to be worth re-measuring.
Roblox publishes how discovery works because it needs the next hit as badly as you do. Roblox Twitter is an echo chamber of misinformation — check the official FAQ before you believe a viral thread, especially about the 500-player safety rules.
Nine out of ten top games let players show off — plots, pets, gear, leaderboards, speed cosmetics. Seeing someone way ahead makes a kid want to grind, so do not hide progress behind walls or force them to zoom out to see it.
Leave a question hanging: what can I unlock tomorrow, what pet is next, how big can this get. Test it on yourself by coming back the next day and asking whether you actually want to play again.
Kids get emotionally attached to rare things they rolled and want to fill the index. Combine that with random events so there is always something to log in and catch.
Twenty minutes of building or a purchased rare pet turns into sunk cost — players return to protect and grow what they made, exactly like Minecraft. Rare cosmetic blocks they can flex keep bases alive after progression plateaus.
In PvE the tsunami is always the same tsunami; with other players no two sessions repeat, which is why Rivals earns what it does. Pair it with a safe build mode so new players can invest an hour before anyone wrecks them.
Free private servers turn the game into a sleepover activity: friends build together, then tell other kids at school. That word-of-mouth flywheel costs you nothing.
Every genre has one thing all the winners include. In survival it is a class system — Alien Ranch had none and had nothing to bring players back, so Farmer Die added one. Find yours by asking what these games do to earn D1 and D7.
D1 is almost entirely tutorial and onboarding quality. D7 needs enough content plus weekly events, admin abuse and social flex. D28 needs deeper progression, not more models — Hack a Business has about 2.5 hours of initial content. Farmer Digger was destroyed by the new algorithm purely on D7/D28.
Two different rhythms: patches for bugs, exploits and balance go out daily; real updates are weekly, then bi-weekly. Tizzy's cycle is plan Saturday, assign Sunday, build Monday to Wednesday, ready Thursday, playtest Friday with the core, launch Saturday. Never ship an untested last-second update, and hire live ops internally — a studio treats your game as one line in a portfolio.
Decide the number first — play days per user, playtime, D7 — then design the update to move it while still making players happy. One new skin moves nothing; content and systems that extend play do.
Your power players grind hardest, spend most and understand the game best — treat them as co-designers. Plan updates with them, have them playtest before launch, look at the server tags they wear to find what else they love, and ask what games they used to play.
Being visibly in chat is not beneath you no matter how much the game makes — it builds loyalty and it is where ideas come from. Run polls as a side source of data, and be at the top of your own game's playtime leaderboard.
The player is usually right, but players are not designers — they ask for the rarest pet for free without seeing that it kills the economy. Filter every suggestion through design judgement, and accept that someone will always be angry.
A simple building mechanic with infinite depth, where every session is different and the community makes the content. One 'how to build an F-22' video pulled 143K likes. Clippable design is what keeps a game alive for years.
One side: in tycoon and farming games kids AFK while resources tick up and it pumps your playtime metric — Tizzy says it works for his genre. Other side: auto-collect and auto-roll conveniences broke his economy and got kids to end-game in 3 hours. It depends on your genre, and he says so explicitly.
Games that monetise hard die fast — studios that make a game pay-to-win have killed games overnight. Players expect to earn things by playing; a fair economy keeps them around long enough to spend more in total.
Every one of Tizzy's best sellers removes a specific frustration: a 24-hour base lock so AFK players do not get robbed, +30 speed to skip the grind, limited-stock items so you do not look like a filthy casual, 2x on offline earnings so you do not miss out. A product that feels like a nice-to-have does not sell.
Before you build, look at what the game you copied sells — which gamepasses, which dev products, at which prices. When you see something like 2x offline earnings in Sell Lemons, treat it as an experiment in your game, not a permanent feature. Then sort your own products by sales and build more of whatever already wins.
A gamepass is bought once; a consumable is bought like toothpaste — today, tomorrow, the day after — which pumps your 7-day spend days per user. The 24-hour base lock became the top seller two days after launch.
The easiest win on payer conversion rate: one very cheap entry gamepass that turns a free player into someone who has paid you once.
Small, medium and large fries. A cheap starter that converts, a mid-tier that most people buy, and an overpowered expensive one for whales.
Give them a free base lock that refreshes every 60 seconds so they learn what it is worth, then sell the 24-hour version. Put the paid one where the feet are and move the free one somewhere quieter.
Monetization is psychology, not mechanics. Show the Robux button exactly when the kid wants the pet and cannot afford it, and pop the 2x prompt at the moment he walks over to collect offline earnings — not in a menu, not earlier.
Do not hide the shop in a menu. Add a plus icon beside the cash display, auto-open the shop after the third insufficient-funds error in a session, and put physical purchase objects at points of interest where players actually walk.
Currency packs sell, but a billion coins for nine Robux destroys progression for everyone. Convenience features like auto-collect look harmless and wreck the game — some friction is the product. Handing out the rarest pet because Discord asked would kill the whole economy.
Success is not just more Robux. After every monetization change, verify D1, D7, session time and the new-player experience did not get worse — a revenue spike that kills retention loses long term.
Tizzy's real figures. Changing only how the roller teased rare pets took PCR from 0.3% to 0.75% blended. USA players convert at about 2% while the blend is dragged down by lower-income regions. Better placement and consumables tripled one game's revenue while CCU only doubled. He still calls his $25K/month game badly under-monetized.
A visibly vibe-coded game that peaked at 10K CCU, settled near 500 players and earns 40k Robux a day — about $140, or $4K a month. Age Test, a quiz game built in a couple of hours with a single dev product, still makes about $25 a day. Simple monetization on a simple game is a livable income.
Roblox's data pipeline is not Shopify — numbers take about 3 days to be real, sometimes 24 hours, sometimes longer at launch. And your early stats will look terrible unless you are already a top dev; that is normal, it just means room to improve.
Qualitative (what players say in feedback, Discord, surveys), quantitative (everything countable on the dashboard) and competitor research. Qualitative is the one almost every dev ignores, and it is the only stream that tells you why.
Click 'see full report' and Roblox gives you an AI summary of what your players are saying about mechanics, bugs and balance. Most devs never open it, and the complaints in there are exactly what a paid product could solve.
Search your game, sort by newest, and find the nine-year-old who screen-recorded it on an iPad. Where he gets stuck or confused is your UX bug list. Do the same with big creators to see which moments they clip, and play your own game to see the playstyles you never designed for.
Analytics > Funnels, with every tiny step tracked separately — 'opened gear shop', 'bought bat', 'closed shop', not one lumped step. Slice each by device, because that is how they found console tutorial completion at 30% against a 77% average.
Analytics > Engagement > New User First Session Retention. The targets Tizzy runs against: 95% still there at 30 seconds, 87% at 1 minute, 65% at 5 minutes. If you see 20-30%, your onboarding is broken.
Read the new-user retention graph over a 7-day window and find the steep drop — say 2:30. Then set a 5-minute timer, play your own game, and watch what is happening at that exact moment.
A 20% bounce rate can be fine in one genre and terrible in another, so judge every number against similar games rather than the platform average. Reading competitor metrics is also how you find which of your stats is really the weak one.
Pick the single number furthest below your competitors and obsess over it until it moves — stop scrolling Discord and social media. Taking Hacker Business's bounce rate from 21% to 15% doubled its impressions and took CCU from 2.5K to 4.5K.
Every update is an educated guess in one of three shapes: if I add X, if I remove X, if I improve X, then stat Y goes up. Ship it, check the stat, and if it did not move, revert and go back to research. That loop is the whole job.
Roblox's experiment system ships an update or a thumbnail to only a percentage of players so you can measure the difference. Their VP of Engineering recommends it, and you can aim a test at one metric — content updates at D7, progression at D28.
If average play time is low, the core loop is not satisfying — every action needs sound and visual feedback, like camera shake when the pickaxe hits rock. For contrast, Build a Base and Steal averages 42 minutes against a typical 3-8, which is top-percentile.
Play-through rate is almost always a thumbnail and title problem, not a game problem — and a bad CTR can also mean you are in a red ocean rather than that the game is bad. Expect CTR to decay as impressions pile up; Hacker Business started at 2.1 QPTR and slid, which is why you keep testing.
PvP layered onto the evergreen Build to Survive genre. It went 1K to 13.2K CCU in a week on ads, then to 25K normal and 38K peak, with 42-minute sessions and traffic split search 31% / home 20% / friends 27%. Revenue climbed from $25,070/month to $131,230, roughly $13K of Robux every 3 days, held there by weekly live ops.
Not programming, not modelling, not UI. Game design is the blueprint — what does the player do, how do they win, what rules hold it together — and it is what separates the top 1% of developers.
Tizzy refuses to gatekeep ideas because they are everywhere; finding the gap is only half the job and building the quality game that fills it is the other half. He puts about 80% of a game's outcome on idea plus execution and 20% on data optimisation.
Know the business behind the business. On Roblox the constraint is not ideas, it is finding developers who ship on time in a space full of teenagers who ghost you.
Post what you are learning on Twitter and YouTube even if it sounds stupid — that public record is what makes good developers answer your hiring posts, and a shipped game like Farmer D with 3.5M visits proves you are serious. When you network, offer something: pay for the call, report bugs in their game, never just ask for free time.
Most young devs are talented enough and simply will not lock in — the winners do 10x the work, not 2x. Turn down the other opportunities in your DMs and stay on one game with one team; small mission-aligned teams doing 12-hour days and daily updates beat big diffused ones.
Your first game should not be a breakout hit. Target 80-150 CCU making $500-800 a month passively — a niche game like that can hold for 5-7 months with no updates, while viral trends spike and crash.
One dev gave up on a game; a competitor copied it, changed only the map, and hit 10K CCU. You can be one update away, so do not lose faith at the first flat week.
Devs who decide the game is 'done' because it makes money are the reason it stops making money. Also accept the ending: every game eventually dies, even the good ones, and then you go make the next one.
Nobody predicts every game right; the learning is in forming the hypothesis and checking reality, not in being correct. And theory is not enough — you have to launch your own games, not just study other people's.
Ask yourself: with billions in the bank, would you still be making games? If the answer is no, do something else — money alone burns out. Tizzy's childhood gaming memories are what make the 12-hour days feel effortless.
Ten years of e-commerce UX work — heatmaps, session recordings, funnels, A/B tests for 8-9 figure brands — transfers directly to game stats. Treat games as products entering markets, and Roblox is easy mode compared to e-commerce because the competition is undisciplined teenagers.
Matia played Roblox games 12 hours a day to internalise what works, then made $250K in two months with Spin a Brain Rot and later hit 100K CCU. Study time converts into design intuition, and intuition converts into money.
One side: Tizzy does not code at all — he is a producer who validates ideas, partners with developers and invests, and he says talent is the bottleneck worth paying for. Other side: he also says AI can now build an entire game in an easy genre, so a solo dev can prototype and ship without hiring anyone. Which one you pick depends on your budget and your reputation, and he has done both.
The same playbook as a board: laws on top, one column per stage, arrows for what leads to what (dashed amber = a gate you must pass first, dotted green = what one step produces for the next). Drag to move, scroll to zoom, hover a box or an arrow for its text.