The country launch is not run in your ops system. It is run in threads and tabs.
The approval pack rebuilt from scratch for the next market. The partner who has gone quiet. The driver document re-attached by hand for the fifth time. The weekly numbers pulled across six countries. We recorded exactly this team — eight people, eight days, 461 tasks — and below is what that work looked like, what it looks like after, and what it gives back.
How the work looks today. How it looks after.
Seven processes, in the order we would take them. The left-hand column is not a caricature of a marketplace operation — most of it is what one was recorded doing, on its own machines, in its own tools. Where a row carries a measured number, the tag says it came from a real recording, and the row itself says which capture.
Every new market starts by rebuilding the same object: the approval pack, the local requirements, the partner checklist, the launch plan — assembled by hand from the last three markets, none of which is quite the right template. The work that made the previous launch possible exists, but it exists as attachments in someone’s thread.
The pack assembles itself from the launches you have already done: the documents, the checklist, the local variations, the gaps listed. What reaches the launch owner is a filled pack with the open items flagged, not an empty folder and a memory of what went wrong last time.
A partner, a fleet, a courier, a driver — each arrives as a packet of documents that someone collects, renames, files and chases. In the operations team we recorded, a browser took about 31% of screen time and a messenger about 24% — more than every internal system combined. The onboarding does not live in the portal. It lives in a thread with a file attached to it.
The onboarding packet is assembled from what already exists across the tools, named to your convention, filed, and the missing items listed. The chase message is drafted in the language your own team already uses. The human reviews and sends.
Who owes what. Who is blocked. What has gone quiet in a country nobody has looked at this week. Coordination is one of the categories that came straight out of the report on the operations team we recorded, and it is performed by hand, in threads, by people holding several markets in their head.
The chase becomes a queue with a shape: what is waiting, on whom, since when, and what would unblock it. Standard follow-ups arrive as drafts. Nothing goes quiet without someone being told it has.
Measured across recorded teams, including a 9-person go-to-market capture: work that needs a second human — a review, an approval, a sign-off, an external reply — waits a median of 7 days. Work one person finishes alone waits 1.1. Two thirds of all waiting, 67.4%, sits across a handoff. And it is not time zones, which for a company spread across continents is the finding that matters most: the least-overlapping pair still shares a quarter of the working day, and the median overlap is 55%. There is always an hour to hand over in. Work stops because the handoff was never defined.
Every handoff gets a shape: what is attached, what "done" means, who is waiting and since when. This is the largest recoverable block of elapsed time in the whole operation and the cheapest one to change — and it is not a tool you can buy, because nobody sells the waiting.
The recurring pull-assemble-format cycle: numbers out of the internal portal, numbers out of the dashboards, into a spreadsheet, into a deck, by a person senior enough to know which number looks wrong. In the capture the internal portal, a spreadsheet and the file manager all sat below the browser and the messenger in screen time. The system of record is downstream of the work.
The pull, the assembly and the format run themselves. What lands on the human’s desk is the part that needs judgement — what moved, what broke, what is worth explaining to the country lead.
Measured on the 9-person go-to-market capture: the two most connected pairs of people both crossed the boundaries of their declared teams, while the declared pair inside one team shared exactly one topic. The structure on the slide and the structure doing the work are not the same structure — which is why a reorganisation aimed at the chart moves nothing.
The real working structure, drawn from who actually works with whom on what. Not a ranking of people and not an input to anyone’s review — a map of where coordination genuinely happens, so a process change lands on the pair that is actually carrying the work.
The company had built its own internal AI assistant — able to read tickets, move statuses, write comments, run queries, build dashboards, assemble presentations — and recommended it to the team. Recorded usage across eight operators: seven at zero percent, one at one percent. The licence is paid every month. The work never changed.
Adoption measured from the work itself rather than from seats and logins — and, more usefully, the reason. An assistant nobody opens is almost never a model problem. It is that nobody mapped it onto a procedure anyone actually performs, in the tool they actually perform it in.
The operations team we recorded is this industry: eight people running cross-border operations at an international mobility and delivery platform — country launches, partners, drivers and couriers, support, process and analytics. Eight days, about 95 hours of recorded work, 461 reconstructed tasks across 44 distinct work directions. Anonymized. Flow and handoff numbers quoted where noted were measured across recorded teams, including a 9-person go-to-market team over five weeks — supporting the same work shape this page describes. Every row marked Measured comes from real recorded work. Every figure was computed before a model was allowed to write a sentence about it, and all thirty supporting quotes in the procedure report were verified against the raw scenes rather than against a summary.
What an agent may touch, and what it may not.
A marketplace operation runs on judgement calls with money and regulators on the other side of them. So we draw the line before anything is built, rather than discovering it in production.
- The launch and approval pack, rebuilt from scratch for every new market.
- The partner and driver onboarding packet: documents collected, named, filed, and chased when missing.
- The chase: who owes what, who is blocked, what has gone quiet.
- The recurring metrics pull across markets, and the assembly and formatting that follow it.
- The case packet: a support, partner or admin case assembled into one reviewable object.
- The meeting pre-read, and the actions drafted from what was actually said in it.
- The commercial negotiation with a partner. Silow drafts nothing that is sent to one unreviewed.
- Pricing, incentives and any unit-economics decision.
- Anything a local regulator expects a named, accountable human to own.
- The exception nobody has seen before — which is most of what you hired an operator for.
- Any decision about an employee. No ranking, no scoring, no automated judgement about a person.
Silow removes the assembly work around a decision. It does not take the decision, and it never takes a decision about a person. The named human the market, the partner or the regulator expects to find stays exactly where they are.
Time, money, and the people you already have.
The rule for this section: not one invented percentage. Every figure below is either measured in a recording or supplied by you — hours multiplied by your own loaded cost. A saving you cannot defend in a board meeting is not a saving, it is a slide.
We do not quote you an industry percentage. Every candidate above is scored in hours against your own recording and ranked by payback. The anchors we can already point at were measured across recorded teams, including a 9-person go-to-market capture: a median flow efficiency of 3.4% across 105 reconstructed tasks. A median seven-day wait every time work crosses a desk, against 1.1 days when it does not. One document re-located in the file manager and re-attached across two messengers on five separate occasions in three weeks, because no shared link for it existed.
The first is hours × your loaded cost — your rate, not a benchmark we invented. The second is the one nobody counts: the AI you already pay for and nobody opens. In the operations team we recorded, the company had built its own internal assistant, capable of running the queries and assembling the decks, and recorded usage across eight operators was zero to one percent. That licence line is being paid this month.
No new system to adopt, no data to move, no process change while we look — Silow runs on the tools the operation already uses, inside the perimeter you choose. The capacity comes back into the same team, running the same markets with the assembly work removed. What you do with it — one more country, or the same countries with the operators no longer assembling packs — is a decision for the people running the operation.
A task is moving 3.4% of the time it exists. Making the operator faster optimises that 3.4%. The launch date lives in the other 96%.
From a launch nobody has ever seen performed to a layer you own.
The agents are the visible part. The thing that makes them possible — and that keeps paying after they ship — is the record of how the operation actually runs a market.
There is a launch playbook somewhere, and the real launch does not follow it. The procedure performed at 09:00 on a Tuesday, across a browser, a messenger and a portal, has never been observed — which is why every automation attempt begins with a workshop and a guess.
The real steps, from the recording, with the evidence attached: in the operations team we recorded, 461 tasks across 44 distinct work directions, and every supporting quote checked back against the raw scene it came from. Ranked by impact, effort, risk and payback.
The launch pack, the onboarding packet, the chase, the recurring pull. Each arrives as a specification an engineer can build from — the observed steps, the evidence, and the lines it must not cross. Human review stays in the loop by design.
The agents are consumers of something more valuable: a private, structured record of how your operation actually launches a market and runs it. Onboarding, search, the next country, and whatever you build after that — on data only you own.
The office work nobody has a line item for.
These six categories are not our taxonomy — they came out of the report on the operations team we recorded, ranked from what that team was observed doing all day across its markets.
The recurring pull-assemble-format cycle across markets, dashboards and sheets.
The chase: who owes what, who is blocked, which country has gone quiet.
Support, partner and admin cases assembled into one reviewable object.
The pre-read and the post-meeting actions, drafted from what was actually said.
Approval and launch packs, rebuilt from scratch for every new market.
Onboarding, access requests, and the tool blockers that stall a whole day.
Where AI actually works in marketplace operations — and where it stalls.
Everyone selling AI into logistics has a case study with a number in it. We are not going to quote you someone else’s, because we cannot verify it. Here are the patterns instead, including the failures.
The part of the business that already runs on models, and has for years. It is also the part your engineers own, it is measured, and it is not where your operators spend their day. The money left on the table is upstream of it.
The most valuable repeated object in the company and the least industrialised. Every market is treated as new because the previous one was never written down in a form anything could reuse. Retrieval, not generation, is the differentiator.
High volume, low judgement, document-shaped — which is exactly the profile that pays back in the same quarter. It stalls when the automation is built against the portal instead of against the messenger thread where the documents actually arrive.
Deflection works on the repetitive tail and stops working the moment the case is genuinely novel — which is most of what escalates to an operator. Assembling the case packet in front of a human beats trying to replace the human.
The highest-frequency, lowest-judgement work in an ops team, and the one most companies believe they solved with a brittle sheet. It looks solved. It is a senior person’s recurring half-day, every week, in every region.
Each market wants its own pack, its own evidence, its own named signatory. The assembly is mechanical and the signature is not — and the companies that get this wrong automate the wrong half.
Pilots do not die because the models are bad. They die because nobody could say which of forty candidate processes to do first, or what "correct" looked like when it was done. That ranking is what Silow produces; the agents are what you build on top of it.
Nobody automates the waiting. It is not a tool category, so no vendor sells it — yet on the go-to-market team we recorded, two thirds of the elapsed time sat in a queue between two people, and not one hour of it was a time-zone problem. It is the largest number on this page and the cheapest one to move.
The part other vendors leave out.
Two of the hypotheses we most wanted to be true were killed by our own data. You should hear that from us here rather than discover it in month three — and in an operation this distributed, a vendor who tells you what their product cannot do is the one worth the second meeting.
The recording sees activity, never outcome. Whether the market launched on time, whether the partner signed — that is not in the data. Join one column of your outcomes to our trace and it becomes provable. Until then, any ROI figure quoted at you is a guess, including one quoted by us.
The pitch was "N people doing the same thing, here is the saving." The real answer across nine people over five weeks was 44 recoverable minutes — 0.22% of working time, with a median duplicate of fifteen seconds. There is no product there, so we do not sell one.
Three quarters of what our naive metric called "meeting time" had no room attached — one person’s 666 "meeting minutes" turned out to be a parked video-call landing page. That is a capture problem, not a metric problem, and we say so instead of shipping the chart.
In the operations team we recorded, we captured about two hours per person per working day. The report says so on its own coverage page rather than quietly extrapolating to a full week. If a number rests on thin data, you will see that in the report itself, next to the number.
Weeks, not quarters. Across every market at once.
Before any recording, we agree where it runs — including fully air-gapped, on-prem or inside your own VPC. No keystroke logging, no microphone, no message content, no scoring of anyone. Those are product constraints, not policy promises.
Silow runs on the team’s existing machines and tools, in whichever countries they sit in. No migration, no new system, no process change while we look — the point is to see the operation as it actually is.
Scenes become tasks; tasks become the real procedure — the launch, the onboarding, the chase. Every opportunity scored by impact, effort, risk and payback against your own hours.
A specification, not a suggestion: the observed procedure, the evidence, the scenes behind it, and the boundaries it must not cross.
What you do with the capacity is yours.
Silow maps work, not workers. No productivity ranking, no performance evaluation, no automated decision about anyone’s job — not as a policy written afterwards, but because the unit of analysis is the process, not the operator. Two weeks of recording, and the first ranked roadmap lands with the hours attached.