operational memory for frontline teams
Make what your people know last.
Carried captures hard-won frontline know-how through voice and photos, then turns it into reviewed, cited memory linked to the equipment and processes people work with.
“Let line 2 sit a full minute before you reset it, or it trips straight back.”
R. Alvarez · shift handoff · @01:12
line 2 · after an overload, wait 60s before reset
cross-checked · 2 operators, manufacturer spec
ratified by D. Chen · supervisor
in corpus · audit-trailed, queryable
the problem
The document misses the why. The fix for the once-in-a-decade failure lives in one person, and that person is leaving.
Anyone can run the operation on a normal day. The value is what to do when something breaks that last broke twenty years ago. That knowledge moves by word of mouth, and the time for it is collapsing as a generation of experts retires at once. The numbers below are from the first industry we measured.
The same retirement wave is running through utilities, oil and gas, manufacturing, the skilled trades, and rail: roughly six to seven million aging skilled workers. Water is simply where we measured first.
evidence from our first industry · water & wastewater
to bring a new operator to certified competency.
Portland publishes the program: a six-month academy, then eighteen months of on-the-job training before the state signs off. Every departure restarts that clock.
source · Portland BES operator-in-training program
replacement openings a year, every one of them backfill.
Not growth. The occupation is projected to shrink through 2034 as routine monitoring automates, so the judgment that retires is not being replaced by new hires, it is simply leaving.
source · US Bureau of Labor Statistics, Occupational Outlook
charge Boeing took after 9,000 early retirements.
The case study that documented it calls lost knowledge a major contributor to the line disruption, not the sole cause. The honest framing is the one that holds up.
source · DeLong, Diagnosing the Costs of Lost Knowledge (Babson, 2005)
the risk-priced line
$125k
median unplanned downtime, per hour
Industrial operators self-report unplanned downtime at a median near that figure, so a single eight-hour incident runs about a million dollars. One avoided knowledge-gap incident pays for the system many times over.
directional · ABB / Sapio cross-sector survey, 2023
voice from the field
Not testimonials. These are practitioners describing the problem in their own words on public industry forums, lightly trimmed for length and attributed by role. We have no customers yet; we have a problem people keep naming for us.
A firefighting department has less stress than a maintenance team in constant turnover.
Over the years they've added, modified, or even removed parts to make it work better. None of that will ever make it into the CMMS plan.
You don't know what they took with them until later, when a problem develops.
You can't write down a sound or a vibration. When you learn the machine's voice, it will usually tell you what is wrong.
The knowledge of why the changes were made and what problems they fixed gets lost. The records say what was done, without much nuance.
When I had a problem machine I couldn't figure out, I would call the veteran day shift guy and hold the phone up near the machine. He would listen to a cycle or two, and invariably he was correct.
how we answer it
When knowledge lives in people, every departure starts a fire. Carried moves it into the system, so it stays when they don't.
how it works
Nothing enters memory unchecked. Watch one claim earn its place.
the claim
“Past 80% speed, pump 2 starts to vibrate. Ease it up or it trips.”
- 01
capture
stated by R. Alvarez · 2026-03-14 · @01:12
- 02
extract
claim · pump 2, above 80% speed → vibration
- 03
verify
cross-checked · 2 operators, manufacturer spec
- 04
ratify
ratified by D. Chen · shift supervisor
- 05
corpus
in corpus · citable, audit-trailed
No single person is the source of truth. A claim is cross-checked against the operators who never worked together and against the manufacturer’s own curve, a supervisor ratifies it on the record, and only then does it become citable.
the shelf
One operating logbook per industry. We write them one at a time.
Scroll the shelf, then open a logbook to read what we captured. Two are written. The greyed ones are industries whose manual is still walking around in people’s heads.
the operating environment · pump station and clarifier
We started where the knowledge is oldest and the cost of losing it is highest.
Water and wastewater utilities run on judgment that was never written down: which clarifier needs a manual push-start, what a failing pump sounds like, why a valve from a 1980s retrofit is labeled wrong. Anyone can run the plant on a normal day. The value is what to do when something breaks that last broke twenty years ago, and the person who fixed it then is the person now retiring.
So this is the first chapter we wrote. Capture rides on the handoffs and walkthroughs operators already do out loud. Each claim is extracted, cross-checked against other operators and the manufacturer’s own data, and ratified by a supervisor before it can be cited. Nothing enters memory unchecked.
operators name it themselves
Our control manual had a typed, formatted page for the hardest mode the plant runs. The entire procedure it gave was one word. Pray.
Even if someone wrote it all down, with pictures, most of the crew still would not read it. The quirks of the plant pass by word of mouth or not at all.
Unless you are deliberate about keeping it, the knowledge just walks out the door. The ones who still pick up the phone do not remember much a year or two after they leave.
See it on one of your own recordings. We will walk a real capture session through the pipeline with you, from the spoken handoff to the citable record.
Request demo accessthe product
Ask in plain words. Get an answer that shows its work, or an honest no.
The screens below are the live product. Anyone on the team asks a question the way they would ask a colleague and gets an answer with its sources attached. The review queue keeps a wrong or unsafe claim out before anyone acts on it, and every entity stays traceable to who said it.
ask
Grounded answers, every claim cited back to who said it. Follow-ups stay in the thread.
review
Nothing enters memory unchecked. Approvals commit on a delay you can undo.
knowledge
Every entity is traceable. Sort by open contradictions, filter by what it is.
what a supervisor asks first
Built for work where being right matters more than being fast.
We know more than we can tell. The job is to draw out the part that can be told, keep it honest, and leave the person who knows it in the loop.
The leader who signs for it gets a record they can trust. The expert whose knowledge it is stays credited every time it is used, and decides what gets kept.
Transcript only. No audio stored.
Capture rides on conversations operators were already having. Your people correct what's already there, with no forms and no blank pages. We keep the transcript, not the recording, and sensitive details are redacted before anything is stored.
Role-based access, on the record.
Who can read, who can capture, and who can ratify is set by role. Every answer shows who said it and who endorsed it, so a supervisor can trust it and an auditor can trace it.
It says when it does not know.
When no captured claim covers a question, the system says so and points to the person most likely to know. It does not guess. Unknown is an answer, not a gap papered over.
It shows whose head it lives in.
A knowledge-at-risk report scores how much of what the plant knows is held by exactly one person, and what walks out the door with the next retirement — so it gets captured while they are still around.
request access
See it on one of your own recordings.
Demo access is rolling out to pilot utilities. Request a code and we will walk a real capture session through the pipeline with you.