Direct Customer Observation Loop
Watch real behavior, isolate friction, improve the product, and repeat
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 6
- Confidence
- 98%
Start by entering the workflow as a real participant or quietly observing customers use the product, including competing alternatives. Ask short questions about why they chose a product, abandoned a task, or reacted negatively, but prioritize behavior over stated preference. Convert recurring friction into a concrete metric such as cancellations, lateness, or price volatility. Then inspect the entire information and incentive chain behind that metric: missing information, poor presentation, unsuitable matching, or conflicting incentives may all contribute. Ship focused changes, monitor the metric, and return to the field regularly to learn whether the numerical improvement feels better in practice. The loop works because firsthand observation generates hypotheses, operational data tests them at scale, and repeated participation prevents executives from drifting away from customers.
Origin
Rischer traced this practice to observing database users at Microsoft and described continuing it by driving for Lyft roughly every six weeks.
Core principles
- 01Observed behavior reveals needs that dashboards miss
- 02Direct participation builds empathy for every side of a marketplace
- 03Specific friction metrics turn anecdotes into operating priorities
- 04Small interface details can produce large behavioral effects
- 05Repeated field exposure keeps customer knowledge current
How to run it
- 1
Enter the real workflow
Use the product under normal conditions or observe a customer completing the task. Experience constraints that reports and demonstrations remove.
Pro tip Participate in every major role when managing a multisided marketplace.
Watch out Do not let a staged executive demonstration substitute for normal usage.
- 2
Watch before interpreting
Notice where users hesitate, cancel, compare alternatives, or construct workarounds. Record what happened without immediately assigning a cause.
Pro tip Pay attention to seemingly minor details such as font size, screen duration, and information order.
Watch out One memorable interaction is a hypothesis, not proof of a general pattern.
- 3
Ask about the decision
Use concise questions to understand why the person selected the product or behaved as they did. Listen for emotional and practical consequences that aggregate data conceals.
Pro tip Ask about a decision that just occurred so the context remains concrete.
Watch out Leading questions can turn research into validation theater.
- 4
Name a friction metric
Translate the observed problem into a measurable event such as driver cancellations or late arrivals. Establish the baseline and monitor it frequently.
Pro tip Choose a metric customers directly experience rather than a convenient internal proxy.
Watch out Optimizing a metric without understanding its mechanism can create new friction elsewhere.
- 5
Trace the mechanism
Examine the information, matching logic, incentives, and interface details that produce the behavior. Generate several possible causes rather than blaming the user.
Pro tip Look on both sides of a marketplace before selecting a remedy.
Watch out Punitive policies may suppress the symptom while worsening the underlying experience.
- 6
Improve and repeat
Ship a focused change, measure its effect, and revisit the workflow firsthand. Continue grinding down the defect rather than treating one release as completion.
Pro tip Schedule recurring field participation so observation survives calendar pressure.
Watch out A better dashboard number does not guarantee that the lived experience improved.
In the wild
Lyft examined a cancellation rate of about 15 percent and considered what drivers saw before accepting rides: expected earnings, destination neighborhood, presentation size, display duration, and fit with prior preferences. The company repeatedly refined information and matching instead of treating cancellation solely as driver misconduct.
→ Rischer reported that driver cancellations fell to less than 4.5 percent.
While driving, Rischer met a commuter who checked ride prices far in advance and changed among Lyft, a competitor, driving, or staying home because daily surge pricing created stress. That firsthand account made the customer cost of price volatility tangible and informed efforts to reduce surge pricing and offer Price Lock.
→ Lyft created a predictable-price option for recurring routes and reduced its reliance on surge revenue.
Common mistakes
Relying only on dashboards
Aggregate data identifies patterns but may not reveal the anxiety, confusion, or interface detail causing them. Leaders need direct observation to form better hypotheses.
Generalizing from one encounter
A field interaction should initiate investigation rather than dictate a universal solution. Validate it with repeated observations and operating data.
Blaming users for predictable behavior
Cancellations and abandonment may result from missing information or poor matching. Trace the system before adding penalties or pressure.
Is it for you?
Best for
It is best for product leaders responsible for apps, services, or marketplaces with frequent customer interactions and measurable friction.
Not ideal for
It is not ideal when leaders cannot ethically observe the workflow or when sparse usage makes behavioral patterns impossible to validate.
From the transcript
“And really all that meant is my job was to go around and watch other people you use competing products at the time paradox debase…”
“So much, so much?”
“there's just no substitute hearing somebody you know go off about how they much, how much they love that card and how important that partnership…”
From the episode
BONUS: The Future of Ride-Hailing with Lyft CEO David Risher