Making Big, New Feature Bets in Mature Products
It’s a feature-level Innovator’s Dilemma at play
Why is it so hard to incubate entirely new capabilities within mature products? A few strategy lessons distilled from the product histories of Instagram, Twitter, and Foursquare:
Original Tweetstorm:
Product Challenges at Scale
In the early days of a startup, you build as fast as possible to try to find product market fit before you run out of time and capital. By the time you get to scale with millions of users and hundreds of millions in revenue, the product development tradeoffs seem to invert.
At scale, even small product tweaks seem at risk of aggravating paying customers; big new features feel almost impossibly hard to bet on. The struggle to keep building innovative features come down to challenges like:
- How can we make high upside, riskier bets while keeping stable a core product with hundreds of millions of users?
- How do we optimize for learning with incremental experiments without frustrating all our users?
- How do we justify building features that only a subset of customers will love without worrying whether everyone will like it — at least to start?
The Strategy: Incubate, Iterate, Integrate
There’s a three-part strategy dozens of tech companies have used to safely incubate new capabilities, iterate on them with real usage and, if successful, deeply integrate into the core experience to scale impact.
This feature laddering up approach sounds deceptively straightforward; in reality, it can take years of challenging work. When pulled off well, with hindsight it looks like the product evolution was inevitable.
- Incubate new features in a separate product space where they can provide value to a small subset of customers. The space needs to be inconspicuous to the majority but discoverable to the relevant minority.
- Iterate on the underlying functionality, algorithms, data, curation, etc. — powered by real usage — until it’s proven useful to a large portion of users. Disclaimer: this can take many years.
- Integrate the now high quality and de-risked feature prominently throughout the core product experience.
Learning from History: Instagram, Twitter, and Foursquare
These examples focus on building ML-driven features into existing products, because this is an even more complex challenge given uncertainty around data/algorithm quality.
Instagram evolved from a “Popular” tab to a personalized photo feed:
- Incubate: Launches “Popular” tab in 2012 to surface photos from people you don’t follow for the first time, powered by a simple popularity algorithm.
- Iterate: Upgrades to “Explore tab” in 2014, powered by a collaborative filtering model to show photos from your extended network. Launches trending in 2015, personalized video collections in 2016, and so on.
- Integrate: After 4+ years of building ranking models, Instagram took the plunge to move to a ranked feed model powered by the Explore tab models. Later they continued to launch refinements like following hashtags to inject non-follower content into the feed.
Twitter evolved from a “Discover” tab with globally popular tweets to a ranked timeline with personalized alerts:
- Incubate: Launches Discover tab in 2011 to surface popular tweets in your network.
- Iterate: For 5 years years, it refined the Discover personalization algorithms (2012), launched Moments (2015), and then merged search with the new Explore tab (2017). In parallel, it worked on personalized push notifications starting in 2013 with @MagicRecs.
- Integrate: Twitter launched its ranked timeline in 2016. Later that year, @MagicRecs graduated into a 100% feature for all-users. It continued to refine ranking into 2017.
Foursquare
Foursquare evolved from an “Explore” local search tab to dedicated discovery app with personalized, contextual notifications:
- Incubate: Launches Explore tab in 2011 to surface place recommendations based on where you and your friends have checked in.
- Iterate: For 4 years Foursquare refines its background location snapping technology to deliver personalized recommendations without checking in. Late in 2011 it launches Radar to do check-in reminders, Pilgrim in 2013 to alert people to interesting nearby places, and contextual travel recs in 2014.
- Integrate: Foursquare relaunches as a personalized recommendations app in 2014 — betting the whole company on tech where check-ins were no longer required. After continuous quality improvements, opens Pilgrim as an SDK (2017) to build a new core business line.
Summary
Beyond these examples, companies from Google to Spotify to Facebook have taken similar approach. It’s a reliable playbook for for building an entirely new feature (set) into a mature product: Incubate, Iterate, then Integrate.
Carve out the product space — and time — to take big swings and avoid the siren song of incrementalism.
