A refer a friend redesign that cut acquisition cost by 35 per cent

In 2021 most of PokerBaazi's new players came through paid channels, and the refer a friend flow was hidden so deep in the app that most players never found it. I redesigned it as a full module with its own tab, a friend tracker and a leaderboard. This study covers how jobs to be done shaped the decisions, and how the 35 per cent drop in blended acquisition cost was worked out.

The short version

• My role. Hands-on design from the first sketch to the handoff, with the research, the flows, the screens and the motion all mine. I worked day-to-day with the product and growth leads and the Android and iOS engineers. The growth lead owned the reward amounts and the referral budget, and the graphic design team made the banners to my briefs.
• The timeline. Discovery and design ran through the middle of 2021. The module shipped in October 2021, and the referral leaderboard went live in November 2021.
• The outcome. Referral grew from 5 to 50 of every 100 new players, and blended customer acquisition cost fell by 35 per cent.
• The method. Jobs to be done for framing, an effort and impact matrix for scope, two built candidates for every contested decision, and rewards released in stages so that the referrer keeps coming back.

What was the challenge?

PokerBaazi's old refer a friend flow lived several taps deep. New players, who are the most likely to still be telling friends about a new app, never came across it. So it brought in very few players, and because it brought in so few, nobody had put real effort into it.

• The business goal was to bring in recreational players at a lower cost than paid installs, and keep them long enough to earn that cost back.
• The second goal was to turn each referred player into a referrer too, so the channel keeps growing without a matching rise in the ad budget.
• Both had to work for someone who had installed the app that morning and was still learning the game.

What was happening in the market?

Before drawing anything I went through the referral programmes of fantasy sports, payments, food delivery and rummy apps, the apps our players already used every day. Those apps had already set what our players expected a referral to look like.

Two things stood out. The reward was almost always tied to what the friend did after signing up, which changes the referrer's job, because they now care whether their friend actually plays. And a programme that paid once, on signup, gave the referrer no reason to open it again. Tiers and a way to track friends did. None of the apps had a referral leaderboard, so that was PokerBaazi's own addition.

Jobs to be done

A referral involves two people, the referrer and the friend, and the old flow did not work well for either. I wrote the jobs out before drawing any screens. I spoke to 12 regular players who already invited friends to private tables, and read three months of support tickets about referrals. Most of those players shared invites on WhatsApp, and many wrote to their friends in Hindi.

The referrer. When a friend asks what I keep playing, I want to invite them in a way that does not feel like a sales pitch, so that they join because they want to.

The friend. When I am curious about poker but wary of losing money, I want a way in that someone I trust has vouched for, so that I do not lose money on my first day.

The business. When paid installs bring in most of our new players, we need a channel that costs less for each new player, so that growth does not depend only on the ad budget.

These forces shaped the design more than the competitor screens did. I turned them into three rules and checked every screen in the module against them.

The rules tie the jobs to the screens. The invite goes out from the referrer's own WhatsApp, and a toggle on the invite screen switches the message to Hindi, so the friend gets it from someone they know, in the words that person would use. The friend gets a joining bonus too, so they can try a table on the first day without putting in their own money. The referrer's own bonus stays on the referrer's screen and never appears in the message, so the invite does not read like a sales pitch.

Journey of a referred player

A referred player reaches PokerBaazi through a link a friend shared, usually on WhatsApp. They are new, and most are recreational players still learning the game. The journey below follows that player from the invite to the moment they send an invite of their own.

Install is the lowest point. The friend has to leave the chat, find the app and sign up, and what carries them through is that someone they trust sent the link. That is why the invite starts in the referrer's own WhatsApp. The first real game is the peak, and every step before it is built to get the friend there.

The matrix set the build order. The home screen entry point and the main screen were high impact and low effort, so they came first. The tabs and the friend tracker were high impact but heavier, so they followed. The nudge sheet was cheap and useful. The leaderboard and the full set of micro interactions were the expensive corner and came last.

Where to put the entry point

The live home screen had one obvious place for discovery, the bottom navigation bar. I tried two placements.

The raised button. A raised button in the centre of the bar. Fitts's law says a bigger target closer to the thumb is quicker to hit, and a raised centre button is both. But a centre button pushes every other tab down a level, and for PokerBaazi those tabs carry equal weight. It also pushes the module at people instead of letting them find it.

The inline tab. A tab in line with the others, in the slot of the route used least. It has the same priority as everything else and is still visible to every player. This is the one we took. The next problem was making a new tab noticeable to players who expected something else in that slot.

Animated or still

To make the tab hard to miss I made two icons. One was a Lottie animation, which the app already supported. The other was a still icon in a different style from its neighbours, a filled gold coin among line icons.

A looping Lottie in a bar that is always on screen dropped frames on the entry-level Android phones we tested on. We showed both versions to 8 players, and 7 of the 8 found the tab without help with either icon, so the still coin shipped.

A reason to come back

Discovery gets a player into the module. The harder problem is giving them a reason to come back after the first invite. The answer was to show the referrer how far each friend had got, and give them a way to help.

The referrer's bonus is released in three slabs as the friend plays more, 5,000, then 15,000, then 25,000 in bonus cash. The friends invited tab shows each friend's progress as a line of steps. A filled dot marks a step behind them, a red dot the step they are on, and an empty circle each step still to come. A nudge button sits on the current step only, so the referrer always knows the one thing that would move their friend forward. A bottom sheet on the main screen carries the same two prompts, nudge a friend or invite another, so the next action is never more than one tap away.

From three milestones to one

The earnings strip on the main screen originally showed three milestones, counting the starting point. The first milestone alone needed the friend to play for a fair while before any bonus landed. Shown next to two more of the same size, that first stretch looked long, and a player who sees a long road ahead is likely to give up before the first reward.

So we cut the strip to a single milestone, with what has been earned on one side and what is left on the other. The payout rules stayed the same, and only the strip changed.

The referral leaderboard

The last piece was a referral leaderboard with a prize pool, so that the players who invite the most get some recognition. The top three sit on the main screen with a link to the full board.

Showing three winners took more thought than it sounds. I tried them in a row and in a column, and tested both with 8 players on the smallest phones we supported.

1. Reading a rank top to bottom is faster than reading it left to right. Nobody has to work out which of three equal circles is first.
2. On a narrow screen the row grew cramped, and there was a real chance of losing the prize figure or a name.

The column won on both counts.

What shipped

Four calls were contested enough to need two built candidates each. Here they are in one place, with what decided them.

The module went out with an inline entry point, an invite screen with the player's code, WhatsApp share and the Hindi toggle, four tabs for referring, transactions, friends invited and how it works, the nudge sheet, the friend tracker and the referral leaderboard with its prize pool.

The outcome, and how the cost fell

Blended customer acquisition cost fell by 35 per cent. Blended means paid and referral together, all the spend on new players divided by all the new players. The comparison is new players from December 2021 to February 2022, after the leaderboard went live, against June to August 2021, before the module shipped.

Two things set the number. Referral grew from 5 in every 100 new players to 50 in every 100. And a referred player cost a quarter of what a paid player cost. The cost of a referred player counts the joining bonus for both sides, the slab bonuses and a share of the leaderboard prizes. The cost of a paid install was the same in both periods, so the drop comes from the mix of players.

Before, 95 paid players at 4 units and 5 referred players at 1 unit cost 385 units for every 100 new players. After, 50 paid players at 4 units and 50 referred players at 1 unit cost 250 units. 250 divided by 385 is 0.65, so a new player cost 35 per cent less.

The paid budget stayed the same. For every 95 players who came through paid installs, 95 more came through referral, so new players went up by 90 per cent. Total spend on new players went up too, from 385 to 475 units, which is 23 per cent more, and all of the rise was referral bonuses.

The slabs also changed when the money went out. They are released only as the friend plays, so most of a referred player's cost is paid after that player has started bringing in rake. Advert spend is paid before a player has played a single hand, so a referred player also paid back sooner.

The referral module was the only new acquisition channel in those months, so the drop comes from the module.

What held up and what did not

Held up. The inline entry point with an icon in its own style, which put the module one tap from home for every player. The friend tracker, which gave the referrer a way to help their friend through the first week. And the single milestone, which cost nothing to change and removed the most discouraging thing on the screen.

Did not. Rewards tied to a friend's play sit awkwardly with user control, because a referrer cannot earn unless their friend plays with real money.

The tracker labels each slab with the word rake and a range of numbers. A regular knows what that means. A new player who has just become a referrer does not, and that is the player this module was built for. Today I would label the slabs by games played.

The step a friend is on is shown mainly in red. Anyone who cannot see red well has only the nudge button to tell them which step is live. Today I would name the live step in words as well.

Aadhaar verification sits on the friend's path before the first slab, and it was the largest drop in the tracker. It is required before cash can leave an Indian real-money gaming account, so it has to be there before the first withdrawal. It does not have to come before the first slab.

The last point is responsible play. A nudge button that asks a friend to play more is a growth mechanic inside a real-money product. Today I would cap how often a nudge can be sent, hide it for any friend who has set a play limit, and keep money out of the nudge message.

Conclusion

The module was designed around what the referrer and the friend each needed from a referral. Referral grew from 5 to 50 of every 100 new players, and blended acquisition cost fell by 35 per cent. The work I would do next is slabs labelled by games played, a live step that does not rely on red, and Aadhaar asked for at the first withdrawal.