New products collect events the way anxious sailors collect instruments. Soon the dashboard glows and nobody knows which dial changes a decision.
In the first week, measure the path to value and the path back. The startup traction signal guide shows how those early events become stronger evidence through return, payment, expansion, and referral.
A metric is useful when a bad number tells you what to inspect.
Choose one completed job
Define the smallest result that proves the product did what it promised. Not “created account.” Not “viewed dashboard.” A completed scan, sent invoice, published page, exported file, or resolved request.
Call this the key action. Instrument the steps immediately before it and the return after it. A simple funnel is enough:
- Arrived with intent.
- Began the core job.
- Completed the core job.
- Returned to use or inspect the result.
Every event should have a sentence explaining why it exists.
Keep acquisition attached to behaviour
Record how a visitor arrived, but do not celebrate a channel until its visitors do something useful.
One hundred visits that produce ten completed jobs are different from a thousand visits that produce two. Keep source data simple and consistent. Campaign naming should be boring enough that two founders spell it the same way.
Add failure events
Success events tell you where people arrived. Failure events tell you where to work.
Track rejected input, failed provider calls, abandoned setup, expired links, payment declines, and empty results. Include a safe reason code, not private content. Pair product analytics with server errors so you can distinguish confusion from malfunction.
Read sessions as stories
For the first users, numbers are not enough. Watch them or speak to them. A person may complete the funnel only after fighting the interface for twenty minutes. Another may fail because the product correctly rejected the wrong use case.
Write a short note beside each early account: who they are, what job they attempted, where they hesitated, whether they returned. This does not scale. It does not need to yet.
Avoid false precision
Seven users do not produce a conversion benchmark. They produce seven cases.
Use counts and plain ratios. Keep raw events. Do not redraw the roadmap because a tiny percentage moved after lunch. Look for repeated behaviour, strong disappointment, and changes large enough to see without statistical theatre.
Review on a fixed rhythm
Each week, answer five questions:
- Who reached the promised result?
- Who tried and failed?
- Who returned without prompting?
- Which acquisition source brought them?
- What will we change because of this?
Delete or ignore metrics that never enter an answer. Your dashboard should become smaller as your understanding improves.