Plandek launches Work Units to measure engineering output as companies struggle to prove AI returns

Plandek press release: introducing Work Units, with 1.1M+ PRs scored

A new model sizes every pull request by the code change it contains, giving engineering leaders a consistent unit of output to set against rising AI spend and a way to find where AI gains stall.

LONDON, 18th November – Plandek, the engineering performance intelligence platform for agentic engineering, today launched Work Units, a model that sizes each pull request (PR) by the scale of its code change. Work Units is rolling out to Plandek customers from today, Wednesday 18th November. It gives engineering leaders a consistent measure of output that can be compared across teams and over time and set against rising AI spend. Read alongside delivery and quality data, it also shows where AI gains are getting stuck so that engineering leaders know where to intervene.

Engineering leaders are under growing pressure to show that AI is paying off. Measuring what AI is delivering is not easy. In the State of Tokenomics survey of 472 organizations (September 2026) 43% named proving value or return on investment as their top challenge, and nearly three in four could not confidently connect AI spend to an outcome their CFO would accept.

Answering that takes more than a cost report. AI amplifies the engineering system it lands in, and faster coding does not automatically mean faster delivery. If output rises but lead time or quality does not improve, the gains are being absorbed elsewhere, for example in review, testing or release. Leaders need a consistent measure of output to spot that gap and know where to act.

The usual unit of measurement does not help. Many organizations track engineering output by counting merged pull requests, which treats a one-line configuration change and a multi-file refactor as the same amount of work. A team that ships 50 one-line tweaks in a month looks ten times more active than one that ships five complex architectural overhauls, despite doing a fraction of the work. As AI changes the size and shape of code changes, raw PR counts become a less reliable guide to what teams are producing.

Work Units address that by scoring the change itself. The model accounts for whether code is new or reworked, how many files and folders a change touches, and how the effort per line tapers off as a change grows. As a result, the same work scores higher split into smaller PRs than bundled into one large one, reflecting the extra effort of deciding where to split it and of raising and merging each change, so the model does not reward the oversized changes that carry the most risk. It excludes auto-generated code, dependency lockfiles and vendored libraries, so machine-generated churn does not inflate scores. Scores sit on a readable scale, from 1 for the smallest changes to 50 and above for the largest, and measure the scale of a change whether a person or an AI agent wrote the code. Plandek’s data science team scored more than 1.1 million pull requests while developing the model.

“Every board is now asking its technology leaders the same question: what are we getting for all this AI spend? It’s the question every AI investment decision now hangs on, and counting pull requests won’t answer it,” said Charlie Ponsonby, CEO of Plandek. “Work Units give leaders a consistent way to measure what their teams are producing. And when output goes up but delivery doesn’t, you know the gains are getting stuck somewhere. That’s where you act.”

“The goal was a better measure than PR count, one that reflects the implementation effort behind a change,” said Iona Frost, Data Scientist at Plandek. “There’s no objective measure of effort for an ML model to learn from, so instead we worked through the factors that influence how much work is in a change. The rules are explicit and versioned, and they’ll keep evolving as we talk to more engineers. Because the model is explainable and the results are consistent, it can be used in ROI calculations.”

Because the rules are explicit, the same change always receives the same score, and when the rules are refined, historical data is rescored so trends compare like with like.

Work Units also have practical uses beyond ROI. Teams can set PR size targets and flag oversized, riskier changes early, judge how much review time a change will need, and estimate how much implementation work went into an initiative or epic. Because scoring is consistent, Work Units also build into a history of completed work that Plandek sees as the basis for forecasting, such as sizing a backlog or a body of tech debt before committing to a plan.

The model was developed with Plandek’s own engineers, tuned against PR authors’ and team leads’ assessments of their own work, and reviewed with customers against PRs from their codebases. Plandek recommends using Work Units in aggregate, alongside quality signals such as change failure rate and rework, and never to rank individual developers.

Plandek’s full methodology, “A normalized measure of engineering output: Work Units”, is available on the Plandek blog.

About Work Units

Work Units are Plandek’s normalized measure of the implementation work in a pull request. They size the code change, not the time, value or complexity behind it, and are designed to be analyzed in aggregate over time. The methodology is described in full in A normalized measure of engineering output: Work Units.

About Plandek

Plandek is an outcomes-focused change partner for engineering performance and AI transformation. Combining an engineering intelligence platform, proven frameworks including the 4 Pillars of Engineering Productivity and RACER™, and expert guidance, Plandek helps engineering leaders manage the AI transition, drive improvements in engineering performance and delivery, and maximize the return on human and AI investment. Plandek is used by more than 2,500 engineering teams worldwide across the private and public sectors, including Trainline, Sage Publishing and the UK Ministry of Justice, and is included in the 2026 Gartner® Magic Quadrant™ for Developer Productivity Insight Platforms.

Media contact

Inés Mitsou, Eleven | ines.mitsou@11content.com

See how your engineering efforts translate into measurable business impact

Measure delivery performance, AI impact, and engineering productivity with hundreds of metrics, OOTB dashboards and custom configurations.

  • G2 Fall 2026 Easiest To Do Business With Enterprise
  • G2 Fall 2026 Europe High Performer
  • G2 Fall 2026 High Performer
  • 4.5 Rating from software advice
  • 4.5 rating from  capterra
  • AICPA SOC badge — SOC for Service Organizations compliance certification
  • Crozdesk Happiest Users badge — High User Satisfaction 2023
  • Software Advice Best Customer Support award badge 2022
  • G2 Fall 2025 High Performer Enterprise badge
  • SD Times 100 award badge 2023
  • G2 Fall 2025 Easiest To Do Business With Enterprise badge
  • G2 Fall 2025 Best Support Enterprise badge
  • G2 Fall 2026 Easiest To Do Business With Enterprise
  • G2 Fall 2026 Europe High Performer
  • G2 Fall 2026 High Performer
  • 4.5 Rating from software advice
  • 4.5 rating from  capterra
  • AICPA SOC badge — SOC for Service Organizations compliance certification
  • Crozdesk Happiest Users badge — High User Satisfaction 2023
  • Software Advice Best Customer Support award badge 2022
  • G2 Fall 2025 High Performer Enterprise badge
  • SD Times 100 award badge 2023
  • G2 Fall 2025 Easiest To Do Business With Enterprise badge
  • G2 Fall 2025 Best Support Enterprise badge
  • G2 Fall 2026 Easiest To Do Business With Enterprise
  • G2 Fall 2026 Europe High Performer
  • G2 Fall 2026 High Performer
  • 4.5 Rating from software advice
  • 4.5 rating from  capterra
  • AICPA SOC badge — SOC for Service Organizations compliance certification
  • Crozdesk Happiest Users badge — High User Satisfaction 2023
  • Software Advice Best Customer Support award badge 2022
  • G2 Fall 2025 High Performer Enterprise badge
  • SD Times 100 award badge 2023
  • G2 Fall 2025 Easiest To Do Business With Enterprise badge
  • G2 Fall 2025 Best Support Enterprise badge