Dashboards are sugar. They taste good at standup and give you a rush in front of the team. Then the crash hits and the questions return. If you build products, the win is not a pretty chart. The win is knowing which question matters right now and making the screen better for a real person.
Start with questions that tie to outcomes
Before you open Google Analytics or Mixpanel, write the question you want answered. Not ten questions. One sharp question. If your app is early, a good starter is How many new users reach first value within one day. If you have steady traffic, try What weekly cohort comes back three times within two weeks. Tie every question to a change you can ship. If you cannot imagine a pull request after seeing the answer, the question is not ready.
Teams today are wiring Segment, Kissmetrics, Flurry on mobile and a warehouse like Redshift in the back. That is fine. The danger is tracking everything and answering nothing. So anchor the work in a small set of outcomes:
- Acquisition: which channels bring people who actually use the core feature
- Activation: what event marks first value for a new user
- Retention: do they come back and do the core action again
- Revenue: do they pay and do they stay
Notice each word hides a decision. Activation is not sign up. Activation is the first time a person completes the job your product promises. For Git hosting it might be the first push. For a note app it might be the second saved note. Name the moment. Then measure that moment.
Design events and entities like you design a feature
Events are verbs. Entities are nouns. Keep both simple. A clean plan has a short list of event names that match the way a user thinks. Examples: Signed Up, Created Project, Invited Teammate, Completed Task, Upgraded Plan. No need for cute names. You want words the team will say out loud in a meeting.
Each event should carry the minimum set of properties you need for your question. If you are measuring funnel drop between Created Project and Invited Teammate, you likely need plan type, device, and referrer. You do not need twenty flags. Small payloads lead to clean answers.
Now the entities. In most products you will care about User, Account or Team, and sometimes Device. Define your primary key for each and stick to it. If you are piping to Redshift for deeper SQL, keep the same keys there. Tool changes are common right now. Consistent keys keep you sane when you move from Flurry to something else or when you add a data viz tool like Chartio or Looker.
Keep an eye on two traps:
- Vanity metrics: total pageviews, total signups, total tweets. These rise with any marketing push and hide product truth.
- Event inflation: firing five events for one action. That makes funnels fuzzy and makes product questions slow to answer.
Move from question to plan to dashboard
Once the question is clear and your events match the user story, write a tiny plan. One page. What will we track. Where will it go. Who will look at it. When. Keep the plan close to the code. A shared doc in the repo works well.
Pick tools that fit the question. Need funnels and cohorts for a new feature launch. Mixpanel or Kissmetrics will get you there fast. Need mobile session views and crashes. Flurry or Localytics. Need marketing view across sources. Google Analytics with proper goals. Want one tracking call and many outputs. Segment can fan out events to many tools. Want deep questions like How many accounts with three or more seats churned after a price change. Send events to Redshift and use SQL or a thin BI layer.
Run small tests. Tie your question to a change and put it behind an experiment in Optimizely or your own switch. Set guardrails like error rate and signup conversion so you do not trade speed for a broken flow. Then read results with the same question you wrote at the start. Did more new users reach first value. If yes, ship. If not, roll back and ask a better question.
Quality beats volume. A few clean events mapped to a sharp question will outplay a giant screen of tiny charts. Teams that win run this loop every week.
Questions first vs dashboards first
Dashboards first: You start by connecting everything. You pull in GA, Mixpanel, ad sources, app store numbers. You add ten widgets. The screen looks full. The team feels busy. Two weeks later you are still asking why activation is flat and no one can agree on what activation means.
Questions first: You write Which action marks first value. You pick Created Project and Invite Teammate. You track those two with care and you build a small look showing new users who hit both within a day. You see a drop on small screens. You fix the invite modal on mobile web. Next week the line moves. The team high fives. Not because of the chart. Because the change helped a person get value faster.
Practical checklist for product analytics
- Write the question in one sentence. Tie it to a decision you can ship this week.
- Define the moment of activation. Name the exact event. Make sure everyone repeats the same words.
- List three core events: the first value event, the repeat value event, and the money event.
- Decide on entities: User, Account, Device. Pick stable ids and keep them the same across tools and SQL.
- Map properties you truly need for funnels and cohorts. Keep payloads small and clear.
- Pick the tool for the job: GA for traffic and goals, Mixpanel or Kissmetrics for behavior, Flurry for mobile, Redshift for deep queries.
- Instrument with care: add timestamps in UTC, add version of app, add source for acquisition.
- Test events in a staging project first. Compare counts against server logs to catch missing fires.
- Set up a single view per question. One chart or table that answers it. No extra widgets.
- Review weekly with the team. If the chart does not drive a change, drop it or rewrite the question.
One more pointer for this moment in tech. With Universal Analytics rolling out and event tools shipping features every month, it is tempting to swap tools. Do not rebuild your tracking plan each time. Keep your questions stable and your event names steady. Tools are the radio. Your questions are the song.
Start with the question and the right dashboard will show up.