Humanizing product development: algorithms, growth, and team leadership

Executive overview

Algorithms can't understand long-term effects, user responses, or product intent — humans must stay in the loop. The PM's job on algorithmic products is deciding what the machine handles and what people decide.

Adriel Frederick draws on Facebook, Lyft, and Reddit to cover three areas: when and how to override algorithms with human judgment, how real growth happens (hard work, not hacks), and what skills matter most as PMs move into leadership.

The core job of a PM on an algorithmic product is designing the human-machine interface — deciding where to amplify human intent and where to let the algorithm run.

Keeping humans in the algorithmic loop

• Algorithms can't see external context: competitor moves, weather events, new regulations, or seasonal shifts.
• At Lyft, a pricing model built with PhDs in revenue management couldn't handle basic operational requirements like changing prices city-by-city — it had to be rebuilt.
• "Techno utopians" assume feeding more data solves everything; it doesn't — machines lack strategic judgment and intent.
• The right model: give people information they're uniquely good at interpreting, then let the algorithm amplify their intent across millions of individual decisions.
• Think of ML as a screwdriver — the tool designer decides how much to put in the tool and how much to leave to the person.
• Zuckerberg's choice to exclude the angry-emoji signal from ranking is an example of a human overriding algorithmic logic for product-intent reasons.
• Operational control is a first-order product design requirement, not an afterthought.

How growth actually works at scale

• Growth hacking — small tactical changes — provides initial traction but won't sustain a product without fundamental value.
• "No silver bullets, just many lead bullets, and a few cannonballs every now and then."
• Cannonballs at Facebook: phone-number-based sign-up, reliable SMS delivery globally, breakthrough friend recommendations.
• The "10 friends in 14 days" activation goal wasn't about a magic number — it was a galvanizing rallying cry that made an abstract goal concrete and organizational.
• Most growth at Facebook came from grinding on three things: make the product easy to find, easy to enter, and stupidly easy to find friends.
• Avoid experiment portfolios that drift toward many small tests — lazy incrementalism. Allocate roughly 80% of effort to cannonballs.
• At early stages, skip the experiments entirely; just build the big things.

Finding and fixing the marginal user

• The marginal user is the person just on the cusp of taking the action you want — high intent, not converting.
• Find them by looking for markets with high traffic but poor conversion rates.
• Go to the worst-case user (slowest device, worst connection, furthest from a data center) — they reveal every problem at once.
• Data tells you how bad it is; watching real users tells you why.
• Example: users in India entering their full legal name at sign-up — nobody would recognise them, killing friend-request acceptance rates upstream.
• Quantitative funnels show drop-off points but miss orthogonal problems that only observation uncovers.

Running R&D teams inside larger companies

• R&D teams face "organ rejection" — the rest of the org sees them as unrelated and resents their resources.
• Three rules: (1) the work must feel core to the company mission; (2) success must feel shared, not just the R&D team's win; (3) innovation must visibly remain possible on other teams too.

Diversity as a product advantage

• Growing up in Trinidad — 35% Indian, 35% African, 25% mixed, multiple religions, shared schools across income levels — embedded a natural awareness of how diverse users actually behave.
• Insight applied at Facebook: the assumption of one phone + one number + one person is wrong in most of the world; designing around that assumption drove real growth.
• Diverse teams resolve product debates faster: instead of a two-week user research cycle, the team argues it out in 15 minutes drawing on lived experience.
• Diversity becomes self-sustaining when leadership recognises concrete business value from it — retention follows naturally.

Navigating controversy as a PM

• Controversy is partly inherent: changing patterns of behaviour creates enemies of those invested in the status quo.
• Separate valid criticism from noise — some complaints reflect real product failures (e.g., Lyft not paying drivers for dead-head miles), others are just power-structure conflict.
• Stay close to users: driving for Lyft revealed the real driver-compensation problem behind the PR storm and shaped the approach to Prop 22.
• The PM's job during a PR crisis: buffer the team from both euphoria and panic, keep executing, solve the real user problem.

Skills that matter most as seniority increases

• The skills that got you here — technical depth, individual execution — become less important.
• Organisation design matters most: clear goals, psychological safety, removing friction from teammates' work.
• Empathy at the leadership level means getting out of your own shoes before putting on someone else's — understanding their motivations, fears, and goals before trying to influence them.
• Nothing meaningful gets done by a single person; the job becomes building the environment where great work happens.