Instagram's Connections Pivot: From Celebrities to Friends at First Post
When Bangaly Kaba joined Instagram's growth team in January 2016, the onboarding funnel had 8 steps with no mid-funnel logging — and the follow-recommendation algorithm treated a celebrity follow identically to a friend follow. New users loaded up on celebrities in their first session, then posted months later into a void with no friends following back. The connections pivot — deprioritizing celebrities for new users in favor of real-world mutual connections — doubled retention over the following 18 months.
When Bangaly Kaba joined Instagram's growth team in January 2016, the onboarding funnel had 8 steps with no mid-funnel logging — and the follow-recommendation algorithm treated a celebrity follow identically to a friend follow. New users loaded up on celebrities in their first session, then posted months later into a void with no friends following back. The connections pivot — deprioritizing celebrities for new users in favor of real-world mutual connections — doubled retention over the following 18 months.
The Situation: A Leaky Funnel Nobody Could See
When Bangaly Kaba arrived at Instagram as Head of Growth in January 2016, the product had 440 million monthly active users and was growing fast. But the sign-up funnel had eight steps and almost no instrumentation: only entries at the top and completions at the bottom were logged. Steps 2 through 7 were a black box.
The team's first move was emergency instrumentation — get the data before writing a roadmap. As Bangaly put it: "The first bit of understand work was we needed to do the instrumentation of that funnel as fast as possible to get the data to figure out where the drops were happening and what to fix."
The Tension: A Retention Curve That Dipped Twice
Once the cohort data was clean, a strange pattern appeared. New users would sign up, retain reasonably well, and then — at month 7, 8, or 9 — the retention curve would flatten and dip again. That anomalous second dip was the signal.
Bangaly's colleague Rob Andrews, Head of Growth Marketing, formed the hypothesis: Instagram's follow-recommendation algorithm treated all follows equally. A follow of Kim Kardashian and a follow of your actual friend were weighted the same in the ML model, which was optimized purely for clicks. So new users, shown celebrities at the top of recommended lists, would follow a stream of famous accounts in their first session. That felt great. The graph filled up fast.
Months later, when those same users went to make their first real post, none of their friends were following them — because their friends had never been prompted to connect. As Bangaly described it: "People were revving up Instagram, following a bunch of people, following a lot of celebrities, actually, because the celebrities were being shown, and then when they actually went to make their first post a few months later, none of their friends were following them. And so there was posting into an echo chamber."
The anecdotal read was stark: people felt bad. They posted into silence, received no social validation, and quietly stopped using the product.
The Move: The Connections Pivot
The insight reframed what "activation" meant for Instagram. It wasn't completing sign-up. It wasn't even following accounts. It was the moment of a user's first post — and whether real friends were on the other side to see it.
"The most important thing to do is actually get regular human-to-human connections in the first [session], when people first sign up so that when you actually go make your first post, your friends would see it and you would be validated and you would feel like, okay, this is a place for me. I have a community here."
The fix — what the team called the connections pivot — required a hard internal argument. They had to convince Kevin Systrom and Mike Krieger that recommending celebrities to brand-new users was the wrong call. The revised logic: celebrities should be recommended to users who already had an established social graph. For new users, the algorithm should surface mutual friends and real-world connections first, reserving celebrity recommendations for later in the journey.
"It doesn't mean that we shouldn't recommend celebrities, but [we should] recommend celebrities to people who are already on there. They already have their graph."
The Result: Retention Doubled
The effect was decisive. Over the following 18 months, Instagram's retention doubled. Bangaly was emphatic about the magnitude and the fact that it went largely uncredited in the press:
"There was a TechCrunch article about Instagram's growth at I think 2017, 2018 when we were going 40, 50% year over year... People think that stories was the sole reason why we grew and story brought us a lot of people, but we literally, our retention doubled over the course of a year and a half."
Instagram ended 2016 with 636 million monthly actives — roughly 47% YoY growth from the 440 million baseline when Bangaly joined. The retention improvement compounded on top of simultaneous top-of-funnel gains from celebrity partnerships, an SEO-rich web product launch, invitations, and paid media.
The Lesson
The connections pivot is a clean case study in what activation actually measures. The naive proxy — completing onboarding, following accounts — looked healthy. The lagging signal — retention curves at month 7–9 — revealed the truth: users had filled a graph with strangers and were posting into a void.
The real activation moment for Instagram was the first post landing in front of real friends. Everything before that was preamble. Once the team oriented the early experience around enabling that moment — genuine mutual connection before the first post — the retention math changed entirely. The lesson Bangaly draws applies broadly: "If you're bringing a lot of people but they're not staying around, then you just have a leaky bucket and it doesn't matter how big your top funnel is."
Challenge
Instagram's follow-recommendation algorithm treated celebrity follows and friend follows identically, causing new users to build graphs full of celebrities in their first session. Months later, when those users made their first post, no friends were following them — they posted into silence, felt no social validation, and churned at months 7–9.
Approach
Bangaly's team ran what they called the 'connections pivot': deprioritizing celebrity recommendations for brand-new users and instead surfacing mutual friends and real-world connections first. Celebrities were reserved for users who already had an established graph. The goal was ensuring that by the time a user made their first post, real friends would be there to see it.
Results
- Retention lift: Doubled over 18 months following the connections pivot
- Instagram MAU when Bangaly joined (Jan 2016): 440 million monthly actives
- Instagram MAU at end of 2016: 636 million monthly actives (~47% YoY growth)
- Reported growth rate (2017–2018): 40–50% year over year
Sources
The full record sits in the studio register.
Related
Part of the Activation growth pillar. See also Slack's 2,000-Message Activation Threshold, Notion's Template Gallery as Activation Lever, Twitter's 'Follow 30' Onboarding Optimization.