Data Science Playbook Pt 1 - Learnings at Instagram & Zynga
Recaps learnings at Instagram and Zynga
My career spans a number of marketplaces - gaming, social media, e-commerce. This article is an attempt to distill my key takeaways at each of the places I worked at.
Part 1 consists Zynga and Instagram.
Watch out for bulleted Playbook learnings after each section.
Instagram was a very design-led company when I joined. For a company that made almost half the revenue of Meta (the parent company), it had surprisingly low personnel to support. For instance, FB Stories (borrowed from Instagram Stories) had 6 Data Scientists for every Data Scientist that was on Instagram Stories. So the level of responsibility here was high and more focussed.
My first assignment on the team was to join the Sharing org - responsible for Stories, Feed and Reels - 3 top sharing formats on Instagram. I was excited by the challenges that lay ahead- the first of which was to ship “Add Yours” - a way to collaboratively add-on to your friend’s coffee photo by picking a coffee photo from your own gallery. It tackled 3 core principles of engagement - social proof, mimicry and users needing inspiration or a cue to sharing (a friend’s photo gave them the cue and adding on to a thread reduced the barrier to share). It engaged people to the extent that some of these threads ran into Millions of photos being shared, one for
”plant a tree for every share” that would run into legal challenges. After that viral launch I was handed, Instagram Notes (which was called “Status” then). Situation was that “sending a one-way text message” was the key signal why teens would choose Instagram over Snapchat or vice versa and we had to design a feature that allowed people to create Notes, share them with close friends and invite first time messages. The only caveat was that it was a particularly difficult to get people to send messages if they hadn’t before.
Among things that we had to navigate included
(a) audience that was not large to become “broadcast-y” and not too small that would affect cold start
(b) Design X vs Design Y: One Featuring author over note vs Note over author affected if people wanted to read any note from a close friend or just a note that they could read better even from someone in their outer social circles
(c) making the messaging tab a content consumption surface (not ad monetized and trading off against revenue),
(d) displacing “active now” friends that occupied spot,
(e) adding “Music notes” (!) - pardon the pun, and believing that it would somehow scale and work.
At this time, I created a decision tree with series of “country tests” that we would run for soft launches before our key launch in US where majority of our revenue came from - we also had tease out the revenue loss that would come from users navigating to this surface. I was well supported by the Director of Product, Design, UX Research and my data org as I designed the series of tests that would help narrow (a) through (e), and communicating detail to engineers on detail (read: character limits of notes for optimal engagement, localization challenges in Arabic vs Korean vs English and to succinctly communicate tests results to leadership, guiding towards the launch to 1B users - and Instagram’s 4th new sharing format, that on an entirely different surface - (not Home) but Messages and enabled 1B IG users to more easily access close friends through 1 way messages they had never sent before, all because they had Close Friends’ Note as a “cue”.
Key playbook learnings →
Concise Choices. Be succinct—limit to two options—when seeking leadership feedback, and stay detail‑oriented / rigorous with the engineering team.
Bundle Bets and Align early with Stakeholders. Prioritize and bundle multiple parameters into a single variant instead of running N + tests; this quickly narrows choices and early alignment sets expectations
Frame for Learning. Use “Pros & Cons” / “Learning” tests rather than “Elimination” tests when aligning large orgs and core features.
Test Through Ambiguity. Keep the bar to test low and ship high; when ad‑revenue sensitivity looked low in India and Brazil, we validated in Europe to generalize findings.
Zynga
At Zynga, I supported analytics for one of the company’s largest franchises - FarmVille 2. Back in the day, during Zynga onboarding, the team’s first assignment was simple: Play FarmVille until I hit level 20. That deep dive was formative in uncovering bugs and building empathy for the product you can’t find on a dashboard. Solutions begin in "experience".
I learned about virtual economies and Japanese Gacha “treasure chests” (“loot‑box”) that let you trade off effort in the game in exchange for payment through currency. There are two types of currency - primary (that you get in the Millions and spend in the Millions to drive spending behavior - these are gold coins in most cases and a secondary currency that is limited < 100 usually for most players that people spend actual money $$ to acquire. The secondary currency often can be exchanged for millions of gold coins (our primary currency). This drives the basic abstraction - a currency you earn and spend easily and another that is expensive to acquire that can be exchanged for the primary currency. There are loops of all sorts to get you to do a series of actions and trade off effort for currency (or paying). Games build new features to create new repetitive loops that are engaging at first, then become difficult and allow 2-step currency models to let you buy yourself some “progress”. This is at the core of gaming, when the loops don’t drive curiosity and “near win-near loss” situations - where the player felt that he could win the next time, the game usually dies. This is at the heart of engagement - keeping it “just interesting” but never going overboard.
Key playbook learnings →
Embed yourself in the loop. Playing to level 20 surfaced hidden bugs and built empathy dashboards can’t.
Two‑tier economy powers monetization. A plentiful primary currency & scarce premium currency let players trade effort for $$ while still feeling meaningful progress.
Sustain “near‑win” tension. Design loops where the player is always one step from success; too easy or too hard breaks engagement.
Refresh loops frequently. Introducing new features resets curiosity and gives veterans fresh progression paths.
Guard against currency inflation. Tune exchange rates so premium currency stays valuable without tipping into pay‑to‑win territory.