Most startups confuse agile with speed, but iteration without direction is chaos. This misconception leads teams to sprint aimlessly, shipping features with no learning objective. Studies show many early-stage startups adopt agile practices “without following any specific methodology,” which amplifies this chaos [3].
My years in Meta’s Trust & Safety and Product Growth teams taught me that the most resilient products aren’t built through raw velocity—they emerge from disciplined learning. True agility is a constant cycle: Understand → Identify → Execute. What if the key to better products wasn’t launching faster, but learning faster?
Product discipline begins when “I think” becomes “the data shows.” This marks the first phase—Understand. Early teams often obsess over output (shipping features) instead of outcomes (creating value). Product expert Marty Cagan reminds teams that “our job is not to build features; it’s to solve problems” [4].
A single perfect metric rarely exists in early-stage products. Instead, effective teams build a suite of imperfect metrics. Drawing from Spotify’s experimentation framework, use both success metrics (what you aim to improve) and guardrail metrics (what you must protect) [5]. For instance, a feature that boosts engagement shouldn’t harm user retention.
This approach transforms metrics from vanity numbers into navigational instruments. It ensures that speed points toward meaningful learning, not noise.
Once the problem is clear, hypotheses must be tested—enter Identify and Execute. A/B testing isn’t about proving you’re right; it’s about finding what is right. As Spotify engineers note, “Every product change carries two risks: false positives and false negatives,” and experimentation exists to manage both [6].
At Calm, behavioral data revealed that users who set “Daily Reminders” had higher retention. The team hypothesized that prompting new users to set reminders might improve engagement. The experiment worked—40% of users acted on the prompt, and retention tripled among that cohort [7].
LinkedIn saw similar success, finding that iterative experimentation yielded “an additional 20% improvement in one of the firm’s primary metrics” [1]. These cases illustrate a deeper truth: experimentation isn’t a phase—it’s a mindset. Each test reduces uncertainty and compounds learning velocity.
At Meta, the product growth teams operated on a relentless loop: Understand the problem, Identify the biggest lever, Execute flawlessly. The structure scaled from feature tweaks to company-wide initiatives [2].
Understand: Gather quantitative and qualitative data—why guess when you can know?
Identify: Isolate the biggest opportunity by separating signal from noise.
Execute: Channel all energy into solving that single, validated problem.
When Facebook analyzed its account-confirmation funnel, a waterfall analysis showed one-third of users didn’t complete sign-up. The team identified a 23% drop-off among users who never attempted confirmation. The solution—an SMS reminder—raised confirmations by nearly 10%. One insight, one precise fix, massive impact.
This process exemplifies agile discipline: iterate not for speed’s sake, but for clarity.
A sound Understand–Identify–Execute loop depends on diverse perspectives. Too many startups over-index on success metrics while overlooking potential harm. Product innovation thrives on cross-functional tension—the constructive friction between product managers, designers, and engineers.
Marty Cagan defines an “empowered product team” as one where all functions share accountability for outcomes [4]. A designer might defend user satisfaction, an engineer might protect latency, and a PM might prioritize business goals. Together, these perspectives create balance.
Research on agile teams shows that diversity in problem framing directly improves product-market fit and long-term innovation [8]. In essence, inclusion isn’t just ethical—it’s an operational advantage.
For startups to accelerate learning, data must flow freely. As Spotify’s experimentation team puts it, “Data-driven decision-making works only when experimentation is a team sport” [6].
Empowering every team member—not just analysts—to explore data independently fosters curiosity and autonomy. This is where Self-Service Business Intelligence (SSBI) becomes a growth multiplier. When designers and PMs can analyze user flows without waiting for a data scientist, the entire feedback loop compresses.
Research by Mäntylä et al. (2022) on continuous experimentation in startups confirms that decentralized access to analytics “shortens iteration cycles and increases learning throughput” [9]. Democratized data isn’t a luxury—it’s a necessity for sustained agility.
Agility isn’t about moving fast—it’s about compounding insight. The Understand, Identify, Execute loop transforms speed into focus, data into narrative, and iteration into growth.
Startups often see structure as bureaucracy, but in reality, discipline liberates creativity. As Cagan argues, “The best teams learn fast because they’re organized to learn” [4].
So before your next sprint, ask: Are we chasing features—or understanding our users better? The velocity that matters most is not launch speed—it’s learning speed.
Yusuke Kawano is a product leader and data analyst with multiple years of developing world-class trust and safety systems. He focuses on building measurement frameworks and data infrastructure for agile product development. His work also spans across product sales and marketing for digital ads solutions.
1. Mao, J., & Bojinov, I. (2021). Quantifying the Value of Iterative Experimentation. arXiv:2111.02334 [stat.AP].
2. Gleit, N. (2022, April 21). Understand, Identify, Execute. Medium.
3. Mäntylä, M. V., et al. (2022). The Viability of Continuous Experimentation in Early-Stage Software Startups. arXiv:2212.05750.
4. Cagan, M. (2020). Inspired: How to Create Tech Products Customers Love. Wiley.
5. Spotify Engineering (2024, March). Risk-Aware Product Decisions in A/B Tests with Multiple Metrics.engineering.atspotify.com.
6. Confidence by Spotify. (2024). A/B Tests and Rollouts.confidence.spotify.com.
7. Dataversity (2023). 7 Best Practices for Data Collection in 2023.dataversity.net.
8. Huss, M., et al. (2023). Comparing Measured Agile Software Development Metrics.MDPI Journal of Software Engineering, 2(3), 15.
9. Mäntylä, M. V., et al. (2022). Continuous Experimentation in Early-Stage Startups. arXiv:2212.05750.