Why self-assessment goes wrong
Level two looks a lot like level four from the inside. Both have dashboards, both have people who can answer questions, and both feel reasonably under control until somebody asks where a figure came from.
The difference is not capability. It is whether the answers survive scrutiny without a specific person being available.
The five levels
Level 1 — Spreadsheets. Numbers assembled by hand each cycle. The reporting depends on two people and their files.
Level 2 — Reports, no agreement. Dashboards exist, but each team built its own. The same metric is calculated three ways and nobody owns the definition.
Level 3 — One source of truth. A warehouse with modelled, tested data and a shared metric layer. Analysts answer questions without asking engineering first.
Level 4 — Decisions automated. Models run in production against live data, monitored for drift, each with an owner and a rollback path.
Level 5 — Governed and auditable. Lineage, access and model documentation are current by default, so an audit question is answered the same day it is asked.
Three questions that place you honestly
- If the person who built the main report left tomorrow, what breaks?
- When two teams disagree, what settles it — a conversation or the platform?
- How long does it take to answer “where did this number come from”?
Teams at level two usually have good dashboards. That is exactly why they overestimate themselves.
Read the level below you, not above
The next gain almost always comes from finishing the level you are standing on. Teams that jump from level two to a machine learning programme end up with models trained on numbers nobody agrees with.
Erin Castellanos
Chief Executive