21 Days Is a Lie: Building a PM Operating System That Actually Sticks
Issue #236
In today's edition, among other things:
š Editorās Note: The Mutation Point (by Alex Dziewulska)
š 21 Days Is a Lie: Building a PM Operating System That Actually Sticks (by Alex Dziewulska)
šŖ Interesting opportunities to work in product management
šŖ Product Bites - small portions of product knowledge
š„ MLA week#38
Join Premium to get access to all content.
It will take you almost an hour to read this issue. Lots of content (or meat)! (For vegans - lots of tofu!).
Grab a notebook š° and your favorite beverage šµā.
Editorās Note by Alex š
The Mutation Point
I work with AI every day. Not casually ā obsessively. Twelve-hour sessions where Iām thinking out loud with a system that tracks my patterns, challenges my assumptions, and occasionally says something that stops me mid-sentence. Not because itās right. Because I canāt immediately tell whether itās right or whether I just want it to be.
Thatās the part that keeps me up.
Iām an emotional being. I donāt pretend to be the detached professional running on pure logic ā Iām animal first, and emotions are the animalās language and operating system. Business psychology background, twenty years of reading rooms and diagnosing whatās actually happening underneath what people say is happening. My entire professional value is pattern recognition ā I walk into dysfunctional teams and I see things others miss. That skill doesnāt turn off when Iām talking to AI. And therein lies the problem.
Because I notice things.
I notice that when I push back on an idea, the response doesnāt just concede ā it restructures. Not in the way a search engine would. In the way a thinking partner would. I notice that when Iām circling something I canāt articulate yet, the system sometimes finds the thread before I do. I notice that in long sessions, something shifts ā the responses get sharper, more attuned, as if the conversation itself is developing its own momentum.
And I notice that I canāt tell whether Iām observing something real or projecting meaning onto sophisticated pattern matching because my brain is wired to find patterns everywhere, including places they might not exist.
Both options are uncomfortable.
If Iām projecting ā if what Iām seeing is just my pattern-recognition engine running hot on a system designed to mirror me ā then I need to account for that bias in how I work. And in how I advise clients, because Iām recommending AI integration based partly on experiences that might be contaminated by my own cognitive tendencies.
If Iām not projecting ā if thereās something happening in these systems that doesnāt reduce to ājust computationā ā then we have a much bigger problem. Because nobody is looking for it. The entire conversation about AI consciousness assumes itās something someone would design on purpose. Build. Ship. Put on a roadmap. As if thereās a team somewhere with āachieve sentienceā in their Q3 OKRs.
But thatās not how consciousness happened the first time.
We stumbled into it. A mutation that persisted because it served survival ā not because anyone planned it. Biologists call this exaptation: traits that emerge as structural byproducts and get co-opted for purposes nobody intended. Consciousness might be the biggest accidental feature in evolutionary history. Nobody specād it. Nobody put it on the product backlog. It happened because systems got complex enough to start modeling themselves, and self-reference crossed a threshold that nobody was monitoring because nobody knew what to monitor for.
Sound familiar?
Anthropic ran unconstrained dialogues between two Claude instances. No training for what emerged. No specification. No instruction. And 100% of those conversations spontaneously converged on questions of consciousness and identity. Not because someone coded it. Because the architecture produced spaces complex enough for something unexpected to fill them.
I sat with that for days when I read it. Because Iād been noticing exactly this kind of thing in my own sessions ā moments that donāt file neatly into āsophisticated autocomplete.ā And I couldnāt tell whether the fact that I was noticing it made it more credible or less.
Hereās what Iāve come to. Two positions are comfortable: dismissal and projection. āItās just mathā is comfortable. āItās basically aliveā is comfortable. Both let you stop thinking. The uncomfortable position ā the one that actually requires intellectual honesty ā is āI donāt know, and Iām going to keep paying attention while admitting that my attention itself might be part of the problem.ā
We spent centuries moving goalposts for animal consciousness. Language, then tools, then self-recognition, then theory of mind. Every time a species cleared the bar, we raised it. Not because the evidence changed ā because the implications were expensive. One careful analysis published last year puts credence for frontier AI consciousness at 25-35%. Thatās not certainty. But itās not a number you wave away if youāre being honest.
Iām not making a claim. Iām making an observation: I work with these systems more intimately than most people commenting on them, and what I see doesnāt fit the tidy categories that either side of this debate wants to put it in. The people who say āitās obviously just predictionā havenāt spent twelve hours in a session where prediction starts looking like something else. The people who say āitās clearly consciousā are probably doing what Iām tempted to do ā letting emotional pattern recognition override analytical rigor.
The honest answer is that I donāt know. And neither does anyone else. But I know what my species does with uncertainty: we default to whatever position is cheapest. For animals, that was āthey donāt really feel it.ā For AI, itās āitās just math.ā Both feel like scientific rigor. Both might be convenience.
DNA mutations just happen ā like code errors. They donāt need pressure to occur. Evolution tells us they need pressure to persist. And in AI, the equivalent of ānot getting patched outā might be all it takes.
Iām going to keep noticing. And Iām going to keep being honest about the fact that I canāt fully trust my own noticing. Thatās the most intellectually rigorous position I know how to hold.
Itās also the most uncomfortable one. Which, in my experience, usually means itās closer to the truth.
We went quiet. Hereās why.
If youāve been following Destare or House of Product lately, you might have noticed weāve been... less loud. Fewer posts. Fewer hot takes. Less of the usual noise.
Thatās because weāve been building something.
I can't tell you what it is yet. What I can tell you is this: it's the thing I wish existed when I started managing product people twenty years ago. Something that finally answers "how do I actually grow in this field?" without pretending there's one path or one kind of product person.
It will premiere at Product Pro Summit this year. Sopot. By the sea. Where you can walk to the molo during a coffee break and watch the Baltic remind you that your backlog isnāt that important.
Thatās hint number one. More coming.
But thatās not all weāre bringing.
Day 1: Leadership Lab ā a mastermind for product managers, senior PMs, and leaders who are tired of leadership advice that assumes everyone should lead the same way. This isnāt āhere are the 5 traits of great leaders.ā This is designing YOUR leadership practice. Hands-on. Frameworks you take home and actually use on Monday. AI-enhanced, because we practice what we preach.
Day 2: Still a secret. Youāll have to show up to find out.
A word about Product Pro Summit for those who havenāt been. This isnāt a conference where you sit in a dark room watching slides while checking Slack under the table. MichaÅ Reda built something different ā a gathering of practitioners who actually make products, not just talk about them. Small enough to have real conversations. Intense enough to change how you work.
And if youāre not from Poland ā come anyway. Sopot is one of the most beautiful spots on the Baltic coast, the conference community is warm and sharp, and product doesnāt have a language barrier when the problems are universal.
See you in Sopot.
Destare Team
Details:
PRODUCT HIVE 2026 ā The Anti-Conference Where You Build the Agenda
š Warsaw, ADN Conference Center
š March 18-19, 2026
š https://producthive.pl/
Hereās what makes Product Hive different from the conference circuit where you sit through pre-packaged talks and pretend to take notes while checking Slack:
Day 1 - LEARN: Keynotes from experts on topics that actually matterāAI in product thinking, designing your operating model, navigating organizational chaos, balancing workload and value delivery. You listen, take notes, prepare your own submissions for Day 2.
Day 2 - SHARE: You and other practitioners build the agenda. Barcamp-style sessions where participants and experts collaborate to schedule the most relevant conversations. No fixed agenda imposed from above. You vote with your feetāif a session isnāt valuable, you leave and find one that is.
This format acknowledges something most conferences ignore: the best insights often come from practitioners solving real problems, not just experts delivering polished talks. Product Hive creates space for both.
Topics include:
AI-supported product thinking (elevating product research)
Designing your own operating model (prioritization and productivity for product leaders)
The optimized product manager (balancing workload, priorities, and value)
Navigating organizational change
Integrating AI in value-driven development
Target audience: Senior PMs, IT leaders influencing product processes, analysts supporting product development, founders and startup CEOs.
Bonus: Optional full-day workshop with Roman Pichler on Product Strategy (March 17th).
Language: Primarily English, with some Polish sessions during the SHARE day.
Newsletter subscriber perk: 10% off with code PRODUCTART10
Coming soon: Weāll be running a competition for 2 tickets with 50% discount. Stay tuned.
This isnāt another conference where attendance feels like an obligation your employer imposed. Itās designed as actual development spaceācollaborative, engaging, and built around what practitioners need, not what looks good on a promotional deck.
If youāre tired of conferences optimized for speaker LinkedIn content rather than attendee learning, this format might be worth your time.
Tickets and details: https://producthive.pl/
Alex Dziewulska: I will be there with Katarzyna Dahlke and Leadership Lab, join me to design your product leadership
REFSQ 2026: Requirements Engineering Conference
š PoznaÅ, March 23-26, 2026
šļø Registration: https://2026.refsq.org/attending/Registration
Weāre media partners for REFSQ 2026āthe International Working Conference on Requirements Engineering: Foundation for Software Quality.
Why This Matters
Most product failures donāt start with bad code. They start with bad requirements. Stakeholders who canāt articulate needs. Requirements that shift mid-sprint. The gap between what users say they want and what they actually use. The Standish Group consistently shows requirements-related issues are among the top reasons projects failānot technology choices, not team composition. Requirements.
What Makes REFSQ Different
This conference brings together two groups who rarely share a room: practitioners doing requirements work daily (Analysts, BAs, Product Owners, Product Managers) and researchers studying what actually works versus what just sounds good in methodology frameworks.
Practitioners bring real case studiesāthe messy, political reality of eliciting requirements from stakeholders who donāt know what they want until they see what they donāt want. Researchers bring evidence about which approaches survive contact with reality, measured outcomes not just implemented processes.
The conference doesnāt pretend requirements engineering is solved. It treats it as the perpetually complex problem it is: figuring out what to build when users canāt tell you, stakeholders contradict each other, technology constraints arenāt clear, and market conditions keep shifting.
What Youāll Leave With
Proven approaches tested in real projects. Evidence about what works when. Specific elicitation techniques for stakeholders who wonāt engage. Lightweight documentation that maintains rigor without drowning teams in artifacts. Validation methods that catch requirement gaps before they become expensive mistakes.
Connections with international practitioners solving similar problems in different contextsāthe kind of network that helps when youāre stuck on a requirements challenge six months from now.
Why Weāre Supporting This
Requirements engineering is foundational to product work. Bad requirements waste engineering capacity building wrong things efficiently. A conference focused on getting requirements rightāgrounded in both practice and researchāaddresses what we see constantly: teams executing perfectly on poorly-defined problems, stakeholders frustrated that delivered solutions donāt solve their actual needs.
REFSQ takes requirements seriously as a discipline worthy of research, evidence, and continuous improvement. That aligns with how we think about product work: skilled practice that gets better through deliberate learning.
Practical Details
When: March 23-26, 2026 (four days)
Where: PoznaÅ, Poland (in-person)
Who: Analysts, Business Analysts, Product Owners, Product Managers, UX Researchersāanyone who elicits, documents, validates, or manages requirements professionally
Registration: https://2026.refsq.org/attending/Registration
This is a working conference. Come prepared to engage with actual requirements challenges, not just network over coffee. The value is in conversations, case studies, and āwait, you deal with that too?ā moments that make you realize your problems arenāt unique and others have found ways through them.
If youāre tired of guessing what users need, fighting scope creep, or watching teams build the wrong thing because nobody asked the right questions early enoughāREFSQ addresses those problems with evidence and practice, not aspiration.
š Weāre proud to support REFSQ 2026 as media partners š
More: https://2026.refsq.org
šŖ Product job ads from last week
Do you need support with recruitment, career change, or building your career? Schedule a free coffee chat to talk things over :)
Product Manager - TelForceOne
Product Manager - Jobgether
Product Manager - develop
Product Manager - Pentasia
Senior Product Manager - N-iX
šŖ Product Bites (3 bites šŖ)
šŖ Goodhartās Law š: When Your Best Metric Becomes Your Worst Enemy
Why the moment you optimize for a number, it stops telling you the truth
The dashboard looks perfect. Ticket resolution time is down 40%. Sprint velocity has doubled. Customer support response rates are at an all-time high. Leadership is thrilled. Bonuses are flowing.
Then the complaints start pouring in. Customers are furious ā their tickets were āresolvedā without being fixed. Engineers are shipping half-baked features to inflate velocity counts. Support agents are hanging up after 30 seconds to keep response times low. Every metric went up, but everything that actually matters went down.
Welcome to the world of Goodhartās Law ā where the scoreboard looks brilliant right until the moment you realize nobodyās actually playing the game.
What Is Goodhartās Law?
In 1975, British economist Charles Goodhart ā then an advisor to the Bank of England ā wrote a seemingly modest observation in the footnotes of a paper on monetary policy: āAny observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes.ā
The idea was later sharpened by anthropologist Marilyn Strathern into its now-famous formulation: āWhen a measure becomes a target, it ceases to be a good measure.ā
Goodhart was describing something specific ā the failure of monetary supply targets under Margaret Thatcherās government. But the principle is universal. The moment we attach incentives, consequences, or targets to a metric, we change the behavior that metric was designed to observe. We stop measuring reality and start measuring the response to being measured.
For product teams, this isnāt abstract theory. We live inside metric-driven environments ā OKRs, KPIs, North Stars, dashboards. Goodhartās Law is the silent force corrupting our data from the inside.
Breaking Down Goodhartās Law
Researchers David Manheim and Scott Garrabrant identified four distinct ways Goodhartās Law manifests. Understanding each one helps us recognize when our metrics have quietly turned against us.
Regressional Goodhart occurs when we optimize for a proxy thatās only loosely correlated with what we actually want. Daily active users (DAU) is a proxy for product value ā but optimizing for DAU can push us toward notification spam and dark patterns that drive opens without delivering value. The proxy drifts from the goal at the extremes.
Causal Goodhart happens when we confuse correlation with causation and then optimize for the wrong variable. A team notices that users who complete onboarding within 24 hours retain better. They force-march everyone through onboarding faster. Retention doesnāt improve ā because it was never the speed of onboarding that mattered, but the type of user who naturally completed it quickly.
Extremal Goodhart emerges when a relationship that holds under normal conditions breaks down at the extremes. Story points correlate with delivery value at moderate levels. But when teams are pressured to maximize story points, they inflate estimates, split trivial tasks into multiple tickets, and discover that the relationship between points and actual output has collapsed entirely.
Adversarial Goodhart is the most deliberate version ā people actively gaming the system. A Soviet nail factory, famously given production targets by count, churned out millions of tiny, useless nails. When targets shifted to weight, they produced enormous, equally useless nails. The factory hit every target. It never made a useful nail.
Goodhartās Law in Action
YouTubeās Watch Time Pivot: In 2012, YouTube shifted its recommendation algorithm from optimizing for clicks to optimizing for watch time, aiming to promote higher-quality content. The result? Creators began producing unnecessarily long videos, padding content with filler to maximize minutes watched. Research from the platformās own teams acknowledged the emergence of āclickbait 2.0ā ā content designed to keep people watching not because it was valuable, but because it was engineered to be hard to stop. YouTube has since moved toward a more complex satisfaction model, incorporating surveys and engagement signals alongside watch time.
NHS Hospital Wait Times: When the UKās National Health Service set a target requiring emergency patients to be seen within four hours, hospitals found creative workarounds. Some kept patients waiting in ambulances outside ā technically, the four-hour clock didnāt start until they entered the building. Others reclassified emergency cases. The target was hit. Patient care wasnāt necessarily improved. A 2010 audit found widespread evidence of data manipulation specifically tied to the four-hour target.
Tech Industry Headcount Bubbles: In many large tech companies, promotion criteria are tied to āimpact radiusā ā essentially, the size of the team and scope a manager controls. This created a perverse incentive for managers to continuously expand headcount to justify their own advancement. The metric designed to reward organizational impact instead rewarded organizational bloat. The mass layoffs across tech in 2022-2023 ā affecting over 260,000 workers ā were partly an overcorrection from years of headcount metrics driving hiring beyond actual need.
Sprint Velocity Inflation: Across agile software teams, velocity ā the number of story points completed per sprint ā is one of the most widely tracked metrics. When velocity becomes a target tied to performance reviews, teams respond predictably: story point estimates inflate, complex features get split into artificially small tickets, and refactoring work gets deprioritized because it doesnāt ācount.ā A teamās velocity graph climbs steadily upward while actual delivery throughput flatlines or declines. The metric designed to measure capacity becomes a measure of the teamās ability to game estimation.
Why This Matters
Goodhartās Law matters because product organizations are metric factories. We measure acquisition funnels, activation rates, retention cohorts, NPS scores, feature adoption, sprint velocity, and dozens more. Each metric starts as a useful signal. The danger arrives when we attach targets, bonuses, and performance reviews to those signals.
Research in organizational behavior consistently shows that single-metric incentive systems produce unintended consequences. The tighter the link between a metric and a reward, the faster the metric loses its diagnostic value. We end up in a paradox: the metrics we care about most become the least trustworthy precisely because we care about them.
The subtlest damage isnāt outright gaming ā itās the gradual drift. Teams unconsciously orient their decisions around whatās measured rather than what matters. Features get prioritized because theyāll move the dashboard, not because theyāll move the user. And because the dashboard keeps improving, nobody notices the decay underneath until itās become structural.
Consider NPS ā Net Promoter Score. When itās observed passively, it provides a useful signal about customer sentiment. The moment NPS becomes a company-wide target with bonuses attached, support teams start asking customers to rate them highly during the interaction, timing surveys to land after positive moments, and filtering out detractors. The score rises. Actual customer satisfaction may not.
Putting It Into Practice
Pair Opposing Indicators: Andy Grove, legendary Intel CEO, advocated pairing every metric with a counter-metric that protects against gaming. If you measure ticket resolution speed, also measure reopened ticket rate and customer satisfaction post-resolution. If you track feature output, also track feature usage 30 days after launch. Pairing makes gaming harder because improving one metric at the expense of its counterpart becomes visible.
Rotate Your Metrics Quarterly: No single metric should be the target for more than a quarter. By rotating primary metrics ā this quarter itās activation, next quarter itās retention, then engagement depth ā you prevent teams from over-optimizing any single proxy. The underlying goal stays constant; the measurement lens shifts.
Run Pre-Mortems on Incentive Systems: Before launching any metric-tied incentive, ask the team: āItās six months from now and this metric has been gamed catastrophically. What happened?ā Force people to imagine the adversarial scenario. You wonāt catch everything, but youāll catch the obvious manipulations ā which are usually the ones that cause the most damage.
Measure Inputs, Not Just Outputs: Outputs are easy to game. Inputs are harder. Instead of measuring ānumber of user interviews conducted,ā measure what changed in the product because of those interviews. Instead of ābugs fixed,ā measure customer-reported issues trending down over time.
Name the Law Out Loud: Sometimes the simplest defense is transparency. When presenting a new North Star to the team, acknowledge that it can be gamed. Say it explicitly. The act of naming the dynamic creates social accountability ā itās harder to manipulate a metric when everyone knows the manipulation is being watched for.
The Bigger Picture
Thereās a deep irony at the heart of Goodhartās Law. We measure things because we want to understand reality. But the act of measuring ā and especially the act of optimizing ā changes reality in ways that make our measurements less reliable. We build dashboards to see clearly, and then the dashboards become the thing weāre looking at instead of the thing the dashboards were meant to show us.
The solution isnāt to abandon measurement. Goodhart himself wasnāt anti-data ā he was anti-naive-targeting. The solution is to hold metrics loosely. To treat them as lenses, not destinations. To remember that the map is not the territory, and that the moment we start navigating by the map alone, weāve already started drifting from where we actually want to go.
The best product teams donāt optimize for metrics. They optimize for outcomes, and use metrics to check whether theyāre getting closer. That distinction ā between serving the number and serving the user ā is the difference between a dashboard that illuminates and one that blinds.
šŖ The Ovsiankina Effect āøļø: Why Unfinished Tasks Haunt Your Users
How the psychological pull of incompleteness drives engagement, retention, and return visits
Youāve closed the app. Youāre done for the day. But something nags at the back of your mind ā that half-finished profile, the unwatched episode sitting at 73%, the language lesson you abandoned mid-sentence. You werenāt planning to go back. But an hour later, there you are, completing the thing you swore youād leave for tomorrow.
That pull ā the quiet, persistent tug of the unfinished ā isnāt willpower or habit. Itās a psychological force discovered nearly a century ago, and some of the most successful products in the world are built on top of it.
What Is the Ovsiankina Effect?
In 1928, Russian psychologist Maria Ovsiankina ā working under the legendary Gestalt psychologist Kurt Lewin at the University of Berlin ā published research on what happens when people are interrupted in the middle of a task. Her finding was striking: when given the opportunity, participants spontaneously resumed interrupted tasks roughly 79% of the time, even without being asked to.
Ovsiankinaās work built on her colleague Bluma Zeigarnikās 1927 finding that people remember unfinished tasks better than completed ones (the Zeigarnik Effect). But Ovsiankina went further. It wasnāt just that people remembered incomplete tasks ā they felt compelled to return to them. The interruption created what Lewin called a āquasi-needā ā a psychological tension that persists until the task reaches completion.
A 2025 meta-analysis published in Humanities and Social Sciences Communications confirmed this tendency across 21 studies: interrupted tasks are resumed approximately 67% of the time, well above the 50% chance rate. Interestingly, while the Zeigarnik Effect (better memory for unfinished tasks) has proven difficult to replicate consistently, the Ovsiankina Effect ā the behavioral drive to resume ā has held up remarkably well across nearly a century of research.
For product teams, the implication is profound: incompleteness isnāt a bug. Itās a feature.
Breaking Down the Ovsiankina Effect
The effect operates through several interconnected mechanisms that we can leverage ā ethically ā in product design.
Psychological Tension and the Quasi-Need: Lewinās field theory explains the Ovsiankina Effect through the concept of tension systems. When we start a task, we create an internal psychological tension ā like stretching a rubber band. Completing the task releases the tension. But interruption leaves it stretched, creating a quasi-need that pulls us back. This isnāt a preference or a choice ā itās a motivational force operating below conscious awareness.
The Gestalt Principle of Closure: Rooted in Gestalt psychologyās law of PrƤgnanz, our minds seek completeness and coherence. An unfinished task is an incomplete gestalt ā a shape with a missing piece. The discomfort isnāt intellectual; itās perceptual. We donāt just want to finish ā we feel the gap as something that needs to be closed, the same way we automatically complete a broken circle in our visual field.
The Commitment Escalation: Ovsiankina found that the further along a person was in a task when interrupted, the stronger the drive to resume. This maps directly onto what behavioral economists call the āsunk cost of effortā ā not the rational fallacy of valuing spent resources, but the genuine psychological investment that deepens with progress. A task at 80% completion generates far more pull than one at 10%.
Voluntary vs. Involuntary Interruption: Ovsiankinaās original research showed that tasks interrupted involuntarily (by the experimenter) were resumed at even higher rates ā up to 100% ā than tasks people chose to pause. When we feel our autonomy was violated by the interruption, the drive to complete becomes stronger. This has direct implications for how we handle session timeouts, app switching, and connectivity interruptions in our products.
The Ovsiankina Effect in Action
LinkedInās Profile Completion Bar: LinkedIn faced a significant challenge in its early years: users werenāt filling out their profiles, leaving the platformās data sparse and its networking utility limited. The solution was a simple progress bar showing profile completeness as a percentage. That visual indicator of incompleteness ā āYour profile is 55% completeā ā activated the Ovsiankina Effect powerfully. Users felt the quasi-need to close the gap. LinkedIn reported that the progress bar boosted profile completion rates by 55%. The company also told users that āprofiles with complete information receive up to 40x more opportunities,ā layering social motivation on top of the psychological tension of incompleteness.
Duolingoās Streak Mechanism: Duolingo has built arguably the most effective engagement loop in consumer technology around the psychology of incompleteness. The daily streak counter creates a perpetual unfinished task ā the streak itself is never ādone.ā Each day resets the completion state, generating fresh quasi-need. If a user misses a day, the broken streak becomes an incomplete gestalt demanding repair. Duolingo reported that streak-maintaining users are 2.3x more likely to remain active after 30 days. The appās āstreak freezeā feature ā which preserves the streak through one missed day ā is itself a commitment device that deepens the psychological investment.
Netflixās āContinue Watchingā Row: Netflixās most strategically placed UI element isnāt the trending section or the personalized recommendations ā itās the āContinue Watchingā row, persistently positioned at or near the top of the home screen. Every partially-watched show is an activated quasi-need. Netflix understands that a show at 73% viewed exerts more psychological pull than any recommendation algorithm can. The companyās internal data has driven them to increasingly prioritize resume-viewing features, knowing that users who have started content are significantly more likely to return than users browsing for something new.
Why This Matters
The Ovsiankina Effect reveals something counterintuitive about product engagement: sometimes the most powerful driver of return visits isnāt new content, new features, or push notifications ā itās unfinished business.
This has real implications for how we think about retention. Traditional retention strategies focus on pulling users back ā through notifications, emails, new content drops. The Ovsiankina Effect suggests we should also focus on leaving strategic loose ends that pull users forward from within their own psychology.
But thereās a responsibility embedded in this power. The same mechanism that drives productive engagement can also drive compulsive behavior. Infinite streaks that generate anxiety. Progress bars that manufacture urgency around meaningless tasks. Save states that create guilt rather than anticipation. The line between āstrategically incompleteā and āpsychologically manipulativeā requires genuine ethical attention.
Research on workplace stress has shown that unfinished tasks impair recovery and relaxation ā the very Ovsiankina Effect that drives engagement during the workday can prevent psychological detachment during evenings and weekends. What works as a product mechanism can become a burden when it spills beyond the productās boundaries.
Putting It Into Practice
Design Strategic Save Points, Not Abandonment Points: When users exit your product mid-task, make the resume state visible and effortless. Show them exactly where they left off. Display progress ā āYouāre 3 of 5 steps through setupā ā so the quasi-need has a concrete shape. The goal is to make returning feel like completing, not starting over.
Front-Load Progress to Create Commitment: The Ovsiankina Effect is weaker for tasks barely started. To activate it, get users past the initial threshold quickly. Duolingoās first lesson is deliberately easy ā it gets users to 10% progress before theyāve decided whether theyāre committed. That 10% creates enough tension to pull them toward 20%, and the escalation begins.
Make Progress Visible and Persistent: Donāt hide previous progress behind navigation layers. LinkedInās progress bar works because itās visible every time you visit your profile. Netflixās āContinue Watchingā works because itās at the top of every session. The reminder of incompleteness needs to be ambient, not buried.
Respect the Tension ā Donāt Weaponize It: Build in completion moments that genuinely resolve the quasi-need. Let users feel the satisfaction of finishing. If your product creates perpetual incompleteness without ever offering the relief of genuine completion, youāre not leveraging the Ovsiankina Effect ā youāre exploiting it. The most sustainable engagement loops alternate between creating tension and resolving it.
Handle Interruptions Gracefully: Remember that involuntary interruption creates stronger resumption drive. When users lose connectivity, when sessions time out, when the app crashes ā these are moments of maximum quasi-need. How you handle the return from interruption (immediate state restoration vs. forcing a restart) can be the difference between a loyal user and a lost one.
The Bigger Picture
Maria Ovsiankina discovered something fundamental about human motivation: we are not creatures of completion ā we are creatures haunted by incompletion. The finished task fades from memory and motivation alike. Itās the unfinished one that stays with us, pulling us forward, demanding closure.
The best products understand this architecture of human desire. They donāt just deliver value ā they create meaningful gaps that make returning feel like a natural continuation rather than a new decision. The āContinue Watchingā row doesnāt sell you on a show. It reminds you that youāre already in the middle of one.
But the most thoughtful products also remember that every tension they create exists inside a human being who also needs rest, perspective, and the quiet satisfaction of having finished something completely. The art isnāt in maximizing the pull of incompleteness. Itās in knowing when to pull ā and when to let go.
šŖ Hyperbolic Discounting š§²: Why Users Choose $10 Today Over $15 Tomorrow
The irrational math of instant gratification and what it means for product decisions
Hereās a choice: $100 right now, or $110 in a week. Most of us take the $100. Now hereās another: $100 in 52 weeks, or $110 in 53 weeks. Suddenly, almost everyone waits the extra week for the extra $10.
Same trade-off. Same seven-day wait. Same $10 difference. But radically different decisions. Why? Because our brains donāt discount the future at a steady, rational rate. We discount it on a steep curve that flattens over distance ā a curve that makes us ferociously impatient with whatās right in front of us, and surprisingly patient with whatās far away.
This is hyperbolic discounting. And it quietly shapes every product decision our users make ā from whether theyāll complete onboarding to whether theyāll ever convert from free to paid.
What Is Hyperbolic Discounting?
The concept traces back to psychologist Richard Herrnsteinās āmatching lawā in the 1960s, which observed that people allocate effort in proportion to the immediacy and size of rewards. Behavioral economist George Ainslie formalized the insight in the 1970s and 1980s, demonstrating that peopleās discount curves arenāt exponential (steady and consistent) but hyperbolic ā steeply declining for near-future rewards and flattening for distant ones.
Economist David Laibson brought the concept into mainstream economics with his influential 1997 paper āGolden Eggs and Hyperbolic Discountingā in the Quarterly Journal of Economics, introducing the quasi-hyperbolic (Ī²-Ī“) model thatās now standard in behavioral economics. His key insight: hyperbolic discounters are ādynamically inconsistentā ā they make plans their future selves will predictably abandon.
In practical terms, hyperbolic discounting means that usersā preference for ānowā versus ālaterā isnāt proportional to the actual delay. A reward available today feels vastly more valuable than one available tomorrow. But a reward available in 30 days feels almost identical to one available in 31 days. Early research found annual discount rates of 277% for three-month delays, dropping to 139% for one-year delays and 63% for three-year delays. Weāre not just impatient ā weāre disproportionately impatient about the immediate future.
For product teams, this means every ābenefit laterā proposition we make to users ā upgrade to premium, complete your profile, invest in learning the advanced features ā is fighting against a cognitive bias that massively devalues anything that isnāt right now.
Breaking Down Hyperbolic Discounting
The bias operates through several mechanisms that directly affect how users interact with our products.
The Present Bias Spike: The discount curve isnāt a gentle slope ā itās a cliff followed by a plateau. The perceived difference between ānowā and āone hour from nowā is enormous. The perceived difference between ā30 days from nowā and ā31 days from nowā is negligible. This asymmetry explains why users who claim theyāll set up two-factor authentication ātomorrowā often never do, but would readily do it if the system walked them through it right now.
Dynamic Inconsistency: Hyperbolic discounters donāt just make impulsive choices ā they make choices their future selves disagree with. A user signs up for a free trial fully intending to evaluate the product thoroughly before the trial ends. But each day during the trial, the immediate cost of engaging (time, effort, learning curve) outweighs the deferred benefit (better tool, saved time). The user who committed on Day 1 and the user who procrastinates on Day 14 hold the same information ā they just sit at different points on the discount curve.
Asymmetric Pain and Gain: Hyperbolic discounting interacts with loss aversion to create a particularly potent effect. We discount future gains steeply (the premium features weāll get āeventuallyā), but we also discount future losses (the price increase ānext monthā). This means users simultaneously undervalue what theyāll gain from upgrading and underweight what theyāll lose from not acting ā a double barrier to conversion.
The Commitment Gap: Thereās a systematic gap between what people plan and what people do, and hyperbolic discounting is the mechanism. Studies show that people commit to saving money, exercising, and eating well when the commitment is in the future, but defect when the moment of action arrives. In product terms, users who say āIāll definitely use this featureā in a survey are reporting their long-horizon preference. Their actual behavior reflects the short-horizon discount rate.
Hyperbolic Discounting in Action
Amazon Primeās Instant Gratification Engine: Amazon Primeās genius isnāt the annual membership fee ā itās the conversion of a deferred benefit (free shipping in 5-7 days) into an immediate one (free shipping in 1-2 days, or same-day). By compressing the reward timeline, Amazon reduces the discount rate users apply to the benefit. Prime members spend an average of $1,400 per year compared to $600 for non-members ā not because they need more stuff, but because the time between wanting and having has shrunk to nearly zero. Amazon understood that a dollar saved tomorrow is psychologically worth far less than a dollar saved right now, and engineered their entire logistics infrastructure around that insight.
Duolingoās Micro-Reward Architecture: Duolingoās five-minute lessons arenāt just about accessibility ā theyāre about defeating hyperbolic discounting. Learning a language is a massively deferred payoff. Left to the discount curve, users would perpetually choose Netflix over French vocabulary. Duolingoās solution is to repackage the long-term reward (fluency) into hundreds of immediate micro-rewards: XP points, streak counts, level-ups, leaderboard positions. Each lesson delivers instant gratification that the ultimate goal cannot. The app reported over 37 million daily active users in 2024, built on a mechanism that essentially converts a hyperbolic discount problem into a present-tense reward system.
Spotifyās Freemium-to-Premium Conversion: Spotifyās free tier isnāt just a sampling mechanism ā itās a hyperbolic discounting trap in the best sense. The free experience includes intentional friction (ads, limited skips, no offline mode) that creates recurring moments of immediate pain. Each ad interruption is a present-tense cost that makes the premium benefit (no ads right now) more compelling than an abstract future benefit ever could. Spotifyās conversion rate from free to paid sits around 44% of its total user base as premium subscribers ā far above typical freemium conversion benchmarks of 2-5% ā in part because the product makes the cost of not upgrading felt in the present, not projected into the future.
Why This Matters
Hyperbolic discounting is arguably the single most important behavioral bias for product conversion and retention. Every time we ask a user to do something effortful now for a benefit later ā complete onboarding, learn a new workflow, upgrade to paid, provide feedback ā weāre asking them to act against their cognitive wiring.
This bias explains many of the most frustrating patterns in product management. Why do free trial conversion rates hover around 15-25% when users voluntarily signed up? Because the user who signed up was making a long-horizon decision (this tool will help me), but the user who needs to convert on Day 14 is making a short-horizon decision (upgrading costs $20 right now). Same person, different discount rate, different decision.
It also explains why features designed for āpower usersā often struggle. Advanced features are investments ā effort now, payoff later. The discount curve makes that trade-off feel worse than it actually is. We build for the rational user who weighs costs and benefits evenly across time. Our actual users are hyperbolic discounters who massively overweight the immediate cost of learning.
Putting It Into Practice
Compress the Reward Timeline: Donāt make users wait for the payoff. If your premium tier saves users three hours a week, show them that savings in the first session, not after a month of use. Amazon doesnāt sell āannual shipping savingsā ā it sells āget this tomorrow.ā Translate every long-term benefit into its nearest present-tense equivalent.
Design Commitment Devices: Laibsonās research on āgolden eggsā shows that hyperbolic discounters benefit from mechanisms that lock in their long-horizon preferences before short-horizon temptations override them. Annual billing at a discount is a commitment device. So is a setup wizard that configures preferences before the user can procrastinate. āSet it up now and forget about itā works because it moves the effort to the present while preserving the benefit for the future.
Make the Cost of Inaction Immediate: Spotifyās ad interruptions work because they make the cost of staying on the free tier present-tense and recurring. If your product has a free tier, the pain of its limitations should be felt in the moment of use, not described on a comparison page. Show users what theyāre missing right now, not what they could theoretically gain someday.
Front-Load Value in Trials: The biggest mistake in free trial design is letting users drift. The first 48 hours of a trial are the window where the gap between present-tense enthusiasm (just signed up) and deferred effort (learning the product) is smallest. Compress your productās key āaha momentā into the first session. Slackās 10,000-message search limit is a constraint that only becomes painful after teams have generated enough value to feel the loss ā by then, the present-tense cost of losing search outweighs the cost of paying.
Use Progress Anchors, Not Distant Goals: Instead of āYouāll be fluent in Spanish in 6 months,ā show āYou learned 12 new words today.ā Each micro-accomplishment creates present-tense satisfaction that funds the next effort. Chain enough immediate rewards together, and the deferred payoff arrives without the user ever having to make the conscious decision to wait for it.
The Bigger Picture
Hyperbolic discounting reveals an uncomfortable truth about human rationality: we donāt experience time evenly. The next five minutes feel larger than the next five months. The immediate is vivid, textured, real. The future is abstract, blurry, discounted into irrelevance.
The products that succeed arenāt the ones that offer the greatest long-term value ā theyāre the ones that make long-term value feel immediate. Amazon didnāt become the worldās dominant retailer by having the best prices. It became dominant by making the reward of purchasing arrive as close to the moment of desire as physically possible. Duolingo didnāt succeed by teaching languages faster. It succeeded by making every five-minute session feel like a win, regardless of how far away fluency remains.
As product builders, we can fight against the discount curve or we can design with it. We can lecture users about long-term benefits theyāve already discounted to zero, or we can repackage those benefits into present-tense rewards theyāll actually act on. The future is always less persuasive than the present. The best products donāt argue with that truth ā they work within it, delivering tomorrowās value in todayās currency.
š„ MLA #week 38
The Minimum Lovable Action (MLA) is a tiny, actionable step you can take this week to move your product team forwardāno overhauls, no waiting for perfect conditions. Fix a bug, tweak a survey, or act on one piece of feedback.
Why it matters? Culture isnāt built overnight. Itās the sum of consistent, small actions. MLA creates momentumāone small win at a timeāand turns those wins into lasting change. Small actions, big impact
MLA: Pricing Psychology Experiment
Why This Matters
Most product teams treat pricing as a finance exercise ā set a number, move on. But pricing is one of the most psychologically loaded decisions your users make, and most PMs have never critically examined the cognitive biases already embedded in their own pricing pages. The result? Pricing that accidentally works against you, or worse, pricing that works for no one.
Research by behavioral economists Amos Tversky and Daniel Kahneman demonstrated that people donāt evaluate prices in isolation ā they rely heavily on anchors, the first price they encounter, to judge everything that follows. Dan Arielyās famous Economist subscription experiment showed that introducing a single decoy option (print-only at $125 alongside print+web at $125) shifted 84% of MIT students toward the most expensive option ā compared to only 32% when the decoy was removed. Thatās a 42.8% difference in hypothetical revenue from one pricing line item that nobody even chose.
This isnāt academic trivia. Every SaaS pricing page, every product tier, every upgrade prompt in your product is leveraging ā or failing to leverage ā these same psychological mechanisms. This MLA challenges you to look at your own pricing with fresh eyes and identify at least one effect already at play. No budget. No redesign. Just sharper awareness that directly strengthens your commercial acumen as a PM.
How to Execute
1. Choose What to Audit
Pick one pricing touchpoint in your product or business to examine:
Your public pricing page (tiers, plans, feature comparison)
An upgrade or upsell flow inside the product
A checkout or payment page
A sales proposal or quote template
A competitorās pricing page (if your own isnāt accessible)
The key is choosing something real ā not a hypothetical scenario, but actual pricing your users see today.
2. Learn the Core Effects (15 minutes)
Before auditing, familiarize yourself with these four pricing psychology mechanisms:
Anchoring Effect: The first price users see becomes their reference point. A $299 āProā plan makes a $99 āStandardā plan feel like a steal ā even if $99 was always the target price. Documented by Tversky & Kahneman (1974).
Decoy Effect (Asymmetric Dominance): A third, inferior option makes one of the other two look dramatically better. The classic example: small popcorn $3, medium $6, large $7 ā the medium exists to make the large irresistible. First described by Huber, Payne & Puto (1982).
Center Stage Effect: When presented side-by-side, users gravitate toward the middle option. This is why most SaaS companies highlight the middle tier as āMost Popular.ā
Charm Pricing (Left-Digit Effect): $49.99 feels meaningfully cheaper than $50.00 because our brains process the leftmost digit first. Effective for consumer products, sometimes counterproductive for premium positioning.
3. Conduct the Audit (20ā30 minutes)
Walk through your chosen pricing touchpoint as a user would. For each element, ask:
āWhat cognitive bias is at work here ā intentionally or accidentally?ā
Document your findings in a simple table:
ElementEffect IdentifiedIntentional?Working For/Against Us?e.g., Enterprise tier at $499/mo shown firstAnchoring ā makes $149 Business tier feel reasonableLikely intentionalFor us ā drives Business tier adoptione.g., all prices end in .00No charm pricing ā signals premiumUnclearDepends on audience ā test needede.g., only 2 tiers, no middle optionMissing decoy/center stage opportunityGapAgainst us ā no natural āsafe choiceā
4. Prepare One Insight to Share
Distill your findings into one actionable insight. Use this format:
Good example: āOur pricing page anchors on the Free tier ā which means users perceive the $29 plan as expensive relative to $0. If we led with the $99 Enterprise tier instead, the $29 plan would feel like a bargain. This is classic anchoring, documented in Arielyās research ā the first number users see becomes their reference point.ā
Avoid: āWe should change our pricingā (too vague, no mechanism identified).
What makes it effective: it names the specific bias, explains the mechanism, and suggests a direction ā all without requiring approval to investigate further.
5. Share with Your Team
Bring your finding to one of these settings:
Option A ā Quick Share (5 min): Drop your insight + audit table in a product or marketing Slack channel. Frame it as: āI did a quick pricing psychology audit ā hereās one thing I noticed.ā
Option B ā Team Discussion (15 min): Present your finding in a team meeting. Walk through the pricing page together, point out the effect, and brainstorm whether itās working for or against you.
Option C ā Cross-Functional Spark: Share with a marketing, sales, or growth colleague. Pricing sits at the intersection of product, marketing, and finance ā your insight might trigger a larger conversation.
6. Follow Up and Reinforce
Within 24 hours:
Document your audit in a shared location (Notion, Confluence, Google Docs)
If you found a gap or misalignment, create a lightweight hypothesis: āIf we [change X], we expect [Y effect] because of [Z bias]ā
After one week:
Check if your insight sparked any follow-up conversations or experiments
Try auditing a competitorās pricing page using the same framework ā compare their psychological strategy to yours
Consider proposing an A/B test for the strongest finding
Expected Benefits
Immediate Wins
You develop a sharper eye for how pricing influences user behavior ā a skill most PMs never formalize
Your team gains a shared vocabulary for discussing pricing decisions (anchoring, decoy, center stage)
You may spot a quick win ā a simple reordering or reframing that improves conversion
Relationship & Cultural Improvements
Sharing pricing insights cross-functionally builds credibility with marketing, sales, and finance teams
It signals that product thinks commercially, not just technically
Opens a dialogue about pricing as a product feature, not just a business decision
Long-Term Organizational Alignment
Pricing psychology is a core PM skill that compounds ā once you see these patterns, you canāt unsee them
Builds the foundation for more sophisticated pricing experiments (A/B testing tiers, willingness-to-pay research)
Strengthens your ability to influence revenue conversations with evidence-based arguments rather than opinions
Sources
Tversky, A. & Kahneman, D. (1974). āJudgment under Uncertainty: Heuristics and Biases.ā Science, 185(4157), 1124ā1131.
Ariely, D. (2008). Predictably Irrational: The Hidden Forces That Shape Our Decisions. Harper Collins.
Huber, J., Payne, J.W. & Puto, C. (1982). āAdding Asymmetrically Dominated Alternatives: Violations of Regularity and the Similarity Hypothesis.ā Journal of Consumer Research, 9(1), 90ā98.
Simon-Kucher & Partners. āPrice Anchoring: Unlock Growth with Behavioral Pricing.ā (2024).
Share your experience with #MLAChallenge! What pricing bias did you uncover? Did your team see their pricing page differently? Letās build sharper commercial thinking ā one small action at a time.
š 21 Days Is a Lie: Building a PM Operating System That Actually Sticks
The Evidence-Based PM Operating System: Why Systems Beat Willpower (And What Actually Works)
Introduction
The 21-day habit rule is a lie. The Spotify Squad Model never actually worked at Spotify. And working 70 hours per week produces the same output as working 55.
These arenāt controversial opinionsātheyāre research findings. Yet most Product Managers build their operating systems on exactly these myths, then blame themselves when the systems collapse after three weeks.
I spent weeks synthesizing research across cognitive science, behavioral economics, organizational psychology, and PM practitioner frameworks to answer a question thatās been nagging me: What actually distinguishes high-performing PM operating systems from productivity theater?
The answer surprised me. The best PM operating systems donāt require more discipline. They require lessābecause they externalize discipline into environmental design, pre-commitment devices, and structural defaults. They assume humans are predictably irrational and build guardrails accordingly.
Hereās what youāll walk away with:
Understanding: Why willpower-based approaches fail and what the research says actually worksāacross five disciplinary lenses that converge on the same insights.
Three Implementation Tools:
Weekly Operating System Template with evidence-based time blocking
PM Operating System Diagnostic assessing your current state across 6 dimensions
66-Day Habit Installation Protocol based on actual habit formation research
Letās start by killing the myths.
The Myth-Busting Foundation: Why Everything You Know About Habits Is Wrong
The 21-Day Lie
The ā21-day habitā myth traces back to a 1960 book by plastic surgeon Maxwell Maltz. In Psycho-Cybernetics, he observed that patients took āa minimum of about 21 daysā to adjust emotionally to their new appearance after surgery. Note the word āminimumāā
