Decide which work stops before deciding who leaves
Most technology layoffs begin with a percentage and end with a list drawn from performance ratings. I argue that a reduction should start from the work a company will stop doing, and that how people are told matters as much as how they are chosen.
I think the most common way technology companies plan a layoff is backwards. They set a percentage, divide it among departments and ask managers to fill the numbers, often from the latest performance ratings. My position, after taking part in five reductions across four companies, is that a reduction should begin with a decision about which work the company will stop, that the roles affected should follow from that decision, and that the care taken in telling people is part of the decision rather than its aftermath.
This is not an argument that reductions are avoidable or that every company that makes one has failed. Software companies hire against forecasts, and forecasts change. It is an argument about design. The way a reduction is planned determines whether it is the only one, and the way it is carried out determines what the remaining employees believe about the company for years afterwards.
The even haircut and its second round
In November 2008 I was an HR manager at Tamberlane Systems, a telecom-software company in Ottawa, when it cut 9% of its roles evenly across departments. The logic was fairness: every leader shared the pain. In practice no project was cancelled, so the work of the people who left was redistributed to the people who stayed. Seven months later the company cut another 11%, again evenly. By then many of its strongest engineers had already decided to leave, because they had spent half a year doing two jobs and saw no reason to expect the second cut to be the last.
When Kestrelford needed to reduce costs at the end of 2022, the first proposal was a 22% cut across every department in one step. It had real merits. It was quick, it put the cash position beyond doubt and it might have spared people a second round. I argued against it because it repeated Tamberlane's mistake on a larger scale. Instead I asked the executive team to name the work we would stop. We ended a rewrite of our field app, an Australian direct-sales expansion and an internal analytics platform, and froze hiring for six months. That removed 214 roles, 14% of the company.
The cost was a slower route to positive free cash flow, about two quarters later than the deeper cut would have achieved, and an uncomfortable meeting with directors who would have preferred certainty. The benefit is that, almost three years later, we have not needed a second round. I cannot prove that the even cut would have needed one. I can say that I have seen it happen.
Why ratings make a poor list
Once the work to be stopped is chosen, most of the affected roles choose themselves, because the teams doing that work are dissolved. The harder cases are reductions within teams that continue. The usual instinct is to use performance ratings, on the grounds that they are already documented and seem objective. I think that is a mistake for three reasons.
Ratings measure past performance in a role, not fit for the work that will remain. They are calibrated for pay decisions, where small differences cost little, and become very consequential when used to decide who keeps a job. And they carry the biases of the period in which they were given. In our 2023 review, counsel found that nine of the 58 people selected within continuing teams were on parental or medical leave. In five of those cases the manager's reasoning rested on recent visible output, which someone on leave cannot have. We reversed those five decisions before anyone was told.
The better approach is to write down, before any names are discussed, what the continuing team will need: the skills, the customers it serves, the systems it maintains. Then managers explain each selection against those criteria, a second reviewer checks the reasoning and counsel reviews the whole list for leave and other protected grounds. This is slower. It is also defensible to the people affected, which ratings rarely are.
Telling people, and what they keep
The part of a reduction that employees remember is how they were told. At Tamberlane, people learnt their fate in a large meeting, followed by an email, while their system access was already being withdrawn. Nobody intended cruelty. The security team had a procedure and nobody questioned it.
At Kestrelford, every affected person heard from their own manager in a scheduled call, all within one four-hour window, followed immediately by a conversation with HR. Access stayed on until the end of the day so that people could say goodbye and save personal files. They kept their laptops. Severance was twelve weeks plus three weeks per year of service, with benefits for six months, and the window to exercise vested options rose from ninety days to three years, which mattered more to some people than the cash. Colleagues on employer-specific work permits could choose working notice instead of pay in lieu, so that they kept their status for longer while they looked for work.
I would not claim we did all of it well. The decision leaked, and nineteen days of rumour between the board's approval and the announcement cost us more trust than the reduction itself. Our internal directory helping affected colleagues find new roles started three weeks late, when many of them had already sent their first applications without it. And some managers, given a script, read it rather than speaking to the person in front of them; we coached them beforehand, but not enough.
My advice to another company facing a reduction is to refuse the percentage until the work has been named. Ask what will stop, not who will go. Write the criteria before the names. Decide who will speak to each person and what they will keep, and test that plan with someone who has been on the receiving end of a bad one. The people who leave will judge the company by the day they are told. The people who stay will judge it by whether there is a second round.