Technology
Moving an engineering ladder off the wiki
The ladder usually already exists. The rollout is about making it the thing people are assessed against, rather than a document consulted after the decision.
By the CLEAR Talent team6 min read
The organisation in this example
- Setting
- A product engineering organisation running several squads
- Constraint
- A levelling framework in a wiki and a review cycle somewhere else entirely
- Owner
- The people team working with engineering leadership
- Starting point
- Promotion decisions argued from written cases rather than a published standard
Decisions that shape the rollout
- How many proficiency levels the ladder has, and whether IC and management tracks differ
- What a quarterly goal cycle owes the annual decision it feeds
- Who rates whom, and what the anonymity and minimum-rater rules are
- What Ava is allowed to draft, and who signs the result
Almost every engineering organisation has a levelling framework, and almost none of them assess against it directly. The ladder lives as a document that gets refined by one group of people, while the decisions it is supposed to govern get made in a different forum, on a different cadence, from different evidence.
So the rollout that matters here is not a migration of data — there is barely any. It is the decision to make the ladder operative: to hold it as the structure assessments actually run against, so a promotion case becomes a comparison with a published standard rather than a persuasive document written by whoever writes best.
The ladder exists. It is just not what people are assessed against.
Ask an engineering organisation whether it has a levelling framework and the answer is almost always yes. Ask what a promotion panel actually reads and the answer is a document written for the occasion, plus whatever the manager remembers, plus peer comments gathered in the two weeks before the deadline.
That gap is the whole problem, and it is not solved by rewriting the ladder. It is solved by making the ladder the structure the assessment runs against, so that the standard and the judgement stop being separate artefacts maintained by separate people.
A ladder nobody is assessed against is documentation. A ladder assessments run against is a standard.
The decisions worth settling first
How many levels, and how many tracks
Competency scales are configurable from three to seven levels and can differ by job family, so an individual-contributor ladder and a management ladder can sit side by side with genuinely different expectations. Deciding this first prevents the common failure where the management track is quietly the IC track with more meetings.
What a quarter owes the year
Quarterly and annual cycles run side by side and cadences can be mixed per department or level. Decide up front what proportion of the annual decision the quarterly goal scores carry, because they roll up automatically — which is either exactly what you wanted or a surprise, depending on whether you chose it.
Who rates, and under what rules
Rater groups are configurable across managers, peers, direct reports and cross-functional stakeholders. Raters are anonymised within their group and minimum-rater thresholds apply, so agree the thresholds before the first cycle rather than discovering them when a small squad produces identifiable feedback.
What the AI is allowed to do
Ava can assemble existing goal scores, competency ratings and feedback into a first-draft narrative and summarise multi-rater input into themes. Every output is labelled, editable and approved by a human, and every edit is logged — but the policy on where drafting is welcome and where it is not is yours to set, and it can be disabled per cycle, team or module.
What gets built
- The ladder as role profiles
- Each level becomes a role profile holding the competencies it requires with behavioural indicators describing what that level looks like in practice. This is the wiki page, expressed as something assessments can run against.
- Goals with enforced linkage
- Strategic Linked Goals requires every individual goal to connect to a parent objective, so orphaned goals cannot quietly accumulate and a squad cannot spend a quarter optimising for something the company stopped asking for.
- Two cadences, one record
- Run OKRs on a quarterly rhythm and keep the annual cycle for compensation and promotion. Goal scores roll up rather than being retyped at year end.
- The 360° configuration
- Rater groups, anonymity within group and minimum-rater thresholds, with input structured against the competencies in the role profile rather than collected as free text.
Run a quarter of goals before the first assessment
Starting with an assessment cycle means assessing people against a ladder they have not yet worked under, using evidence gathered before anyone knew what would be asked for. Starting with a quarter of goals gives the first assessment something real to read.
It also surfaces the linkage problems early. Completion rates by department, team and individual, at-risk flags based on pace and check-in signals, and end-of-cycle score projections mean a drifting goal becomes a conversation in month two rather than a surprise at review — and a cascade that nobody can connect to a parent objective becomes visible in week one.
Then calibrate across squads before ratings go final. The same title means different things to different managers, and nobody notices until promotion season; a calibration session working from goal scores, evidence-backed competency ratings and structured 360° input is where that gets caught.
What changes, and what happens at the next reorg
- Promotion cases become comparisons
- A case is assessed against the published standard for the level, with the evidence already in the record, rather than assembled as an argument in the weeks before a panel.
- Reorganisations stop destroying history
- Reporting lines, departments and locations are first-class concepts, and bulk import plus staff transfers keep the structure current without rebuilding it. Historical assessments stay in the employee record and the audit log captures the change.
- Distributed teams get the same process
- Web, iOS and Android access with biometric sign-in, push notifications and offline support, plus 16 languages — so a review is not something only the people in the head-office time zone complete comfortably.
- Multiple entities, one platform
- Multi-entity support covers groups that have grown by acquisition as well as by hiring, without forcing every acquired team onto an identical framework.