Decisions from memory
Q1 2027Walk into most operating businesses and ask how a specific decision got made, why that discount, why that supplier, why that staffing level on Tuesdays, and the honest answer, once you get past the slideware, is memory. Someone remembers that this worked before. Someone remembers a bad experience with the alternative. The organization runs on recollection, and recollection is a database with no timestamps, no audit trail, and a well-documented tendency to edit itself in favor of whoever is recalling.
I do not say this with contempt. Memory-based operating is the natural state of every company, and in the early years it is even efficient; the founder's recall is fast, free, and mostly right. The trouble arrives with scale, and it arrives quietly, in a specific and recognizable pattern: the same decision gets made three different ways depending on who is in the room. That sentence, more than any dashboard metric, is the diagnostic. When I hear it inside a client, I know what we are going to find.
What we found inside one manufacturer makes the pattern concrete. A factory chasing five points of line efficiency, each point worth roughly a million dollars a year. The data existed, in the sense that hourly production numbers were written on paper sheets, but a bottleneck took an hour and a walk of the floor to find, downtime repair math was unknowable because start times were flagged from memory, and management steered off a weekly email. Every decision on that floor was made against information somewhere between an hour and a week old. When data is late, money is late. The five missing points were not an effort problem. They were the compound interest on latency.
The prescription in these situations has a name that makes executives reach for their wallets prematurely: a data layer. So let me define it at its real size. A data layer is whatever makes decisions observable instead of remembered. Sometimes that is cameras and edge computing feeding a supervisor an alert in under sixty seconds. Just as often it is embarrassingly smaller: a pipeline where every deal has a stage, an owner, and an exit criterion, so "where is this deal and why" is a lookup rather than a meeting. The technology varies. The property being purchased is the same: shared, timestamped, contestable facts.
Contestable is the important word. The deep cost of memory-based operating is not the individual bad call; it is that the calls cannot be audited, and therefore cannot be improved. When decisions leave a trail, what was known, what was chosen, what happened next, the organization gets access to its own history for the first time. Arguments change species: instead of whose recollection wins, the fight is over what the record shows, and that fight, unlike the other one, produces learning. A digital audit trail sounds like compliance furniture. It is actually how an operation acquires the ability to notice its own patterns.
For anyone wondering where to start, the sequencing rule I use is simple: instrument the decision that is made most often, by the most different people, with the most money per occurrence riding on it. Not the biggest decision; the most repeated one. Repetition is where variance lives, variance is where memory does its damage, and one well-instrumented decision that the team actually consults will outteach a warehouse of reports nobody opens.
Companies love to describe themselves as data-driven, and mostly they are recollection-driven with dashboards. The distinction is not the volume of data. It is whether, at the moment of choosing, the chooser is looking at the world or remembering it.