03 · work · measurement

A common basis for performance and staffing decisions

I built DPET and SPET to give managers a common basis for performance reviews, staffing and promotion decisions, with written criteria behind each rating.

Role: VP of Engineering, Teravision

The situation

We ran dedicated teams with mixed seniority, plus a bench. The bench was a cost. It was also how we covered vacations and sick leave and how we ramped teams up fast. That only works when the people on it can add value from day one. Putting someone with performance problems on a project was the risk I wanted to remove.

Our reviews already had guidelines. Managers were trained, and a tech lead sat in on the technical side. The final rating still came down to the project manager’s judgment. The same engineer could get one rating on one project and a different one on the next. I could not tell whether that person was stalling or improving. Sometimes there was an explanation, like a new technology. Sometimes there was none.

We also knew some people did better on certain kinds of projects. We could name those patterns and had no way to check them.

What I built

Two instruments, used together.

DPET, the Dimensional Performance Evaluation Tool, runs every quarter. It splits performance into dimensions, and each dimension into attributes. Every attribute has a written explanation and a scale, so two managers reading the same line mean the same thing.

SPET, the Sprint Performance Evaluation Tool, runs every sprint. Six sprints make a quarter, and the latest ones feed one dimension of DPET. If someone’s recent sprints drop, it shows, even when their record is strong.

I split every dimension into attributes so a rating error stays inside one attribute. Arriving late to meetings can be proven with data, and it gets its own line. Missed code reviews point to code reviews as the area to improve, a finding about that one practice. Raters still make mistakes, and each mistake stays small.

What changed

The framework informed staffing, project rotation and promotion decisions. Two examples show how we used it.

Rosters. We began matching teams to the complexity of the project or the account, using ratings and seniority. When a consistently solid senior moved into a technology new to them, the data told us the risk was worth taking.

Reputation and data. A well-regarded engineer’s recent scores had dropped. That gave us a signal to investigate, rather than a cause to assume. We adjusted staffing around the recent performance evidence.

The framework also helped us build teams that were cost-efficient as well as strong. Later, the same data informed staffing decisions as the company’s model changed.

It measures what we chose to measure. Every rating decision since has rested on the same written scale, so the reason for a rating is on the page, and we have consistent data to study trends over time.