Impact

Rank work by headcount alone and the same kind of item wins every time. Rewording the home page is seen by 20,000 people a quarter; recovering a failed payment affects 300. On reach alone the copy edit leads by a factor of sixty.

What the two leave behind with one person is nothing alike. Most of those 20,000 never notice the new wording. The 300 tried to pay and left without paying. A table that multiplies headcount alone can't see that difference, so work on high-traffic screens held the top slots quarter after quarter.

Impact is the column that forces the per-person part to be written down separately. How many people a change reaches and how much it changes for them split apart here.

Fix the scale in advance and pick one of five values for each candidate: 3 is massive, 2 is high, 1 is medium, 0.5 is low, 0.25 is minimal.

There's a reason the scale isn't 1 to 10. This value gets multiplied, so narrow gaps stop separating the candidates, and nobody can defend the difference between a 7 and an 8.

Writing a sentence for each value keeps the score steady when the person filling it in changes. Something like: 3 if people can now do what they couldn't, 2 if a task they already did gets far easier, 1 if it just gets a little faster.

In the calculation, impact is multiplied by reach and confidence, then divided by effort. A candidate that 2,000 people pass in a quarter, with impact 2, confidence 0.8 and three person-months of work, scores 1,067. Drop that impact to 1 and the score halves to 533. One step on the scale can flip the ranking, so when this column is contested the team settles it before running any arithmetic.

"Reach is big but impact is 0.25, which is why the score is low."

Plenty of people see it, but almost nothing changes for any of them.

"Is this really a 3? We said 3 means people can do something they couldn't before."

Challenging a score by pointing at the written definition.

"We're keeping last quarter's impact scale as is."

Change the scale and this quarter's scores can't be compared with the previous ones.

In Pagedeck's quarterly planning, Mia fills the reach and impact columns for all 24 candidates first, and impact is almost always the column where people disagree. Sofia puts a 3 on the permission controls three enterprise accounts asked for; Theo puts a 1, arguing there's a workaround. Rather than argue about the number, the two work out what customers actually do without the feature. The workaround turned out to be an admin fixing permissions by hand for 30 minutes every week, so the column settled at 2.