It is 11:14 on a Tuesday and Priya, a planner at a UK 3PL with eighty carrier integrations, is trying to answer one question before lunch. Where is shipment 47C? The customer wants confirmation it will reach their Coventry line by Thursday morning. Priya has eleven browser tabs open. Tab one is the TMS. Tab two is the carrier's tracking portal. Tab three is a second carrier portal because the load swapped hands in Rotterdam. Tab four is the customs broker's status page. Tab five is the OMS. Tab six is NetSuite. Tab seven is the customer's SAP vendor portal, on a different login. Tab eight is Outlook. Tab nine is Teams. Tab ten is a spreadsheet Priya has maintained on the side for three years because none of the other ten tabs holds the columns she needs. Tab eleven is the ticketing system, because the last time she answered this question for the same customer it took two days.
Anyone watching Priya would say she is doing logistics. Anyone counting the keystrokes would say she is doing data reconciliation. The eleven tabs exist because no single system holds the answer to her question. The answer lives across all of them, and only Priya can stitch it together.
We have spent the last year watching this exact scene play out in every industry we have touched. The systems change. The role changes. The tab count moves between five and fifteen. The pattern does not. The most expensive workflow in a B2B operation is almost always the one nobody mentions when they describe the job.
Why this work is invisible
Reconciliation does not appear in any system's reporting. Each system thinks it is the source of truth. The carrier API returns a shipment status and assumes that is the shipment status. The customer's vendor portal returns a different status. The OMS returns a third. None of them carries a row in a dashboard reading, "hours your planner spent reconciling our answer with theirs."
The source of truth is the planner. The reconciliation work happens in her head and across eleven browser tabs. When the CFO asks what the logistics team does all day, the honest answer is that they plan logistics for about forty per cent of their time and reconcile data for the other sixty. The team does not give that answer because nobody on the team thinks of it that way. They think of it as their job.
This is the central reason reconciliation never makes the executive's agent shortlist. It is not on the agenda because it is not on any report. The buyer asks for a pilot of customer-facing automation, or sales copilots, or document extraction. They do not ask for a pilot of "the thing my planner does for four hours a day that I cannot name." To find this work you have to sit next to the planner and watch.
What the work actually costs
Take conservative numbers. A planner who reconciles for four hours a day, four days a week, forty-six working weeks a year, spends 736 hours a year on reconciliation. Fully loaded, a mid-level B2B operations planner in the UK or US costs the business between forty-five and sixty-five dollars an hour once benefits, overhead and management time are included.
That puts the reconciliation cost of a single planner at roughly thirty-three to forty-eight thousand dollars a year. A ten-person planning team carries a reconciliation bill of three hundred and thirty to four hundred and eighty thousand dollars a year. None of this appears in a budget line. It is buried inside the salary line of a team whose job description says "planner."
An agent that does the same reconciliation work, calling the same eleven systems through their APIs and presenting a consolidated view, runs at roughly three to twelve dollars a day in model and infrastructure cost depending on call volume. That is between one thousand and four thousand dollars a year per planner-equivalent of work. The arithmetic is not subtle. The agent is an order of magnitude cheaper than the labour it replaces, and the labour it replaces is not the planning. It is the stitching.
Where this surfaces hardest
Reconciliation lives in every operations team. Three industries have it worst, because the system count is highest, the systems are oldest, and the cost of getting the answer wrong is real.
Logistics and supply chain
The eleven-tab scene above is a real logistics planner we shadowed for a day. She was trying to confirm whether one inbound container would land in time for a customer's production line. The answer required a customs status from the broker, a vessel ETA from the carrier portal, a yard appointment from a third system, an inventory reservation from the OMS, and a confirmation in the customer's SAP portal that the receiving dock was booked. Five systems for one question about one container. She answers questions like this around fifty times a day.
Banking back-office
Regulatory-report assembly in banking is the same shape with higher stakes. A single end-of-day report draws on the core banking system, the trade-capture system, the risk engine, the corporate-actions feed and the custodian's reporting interface. Analysts spend the back half of every afternoon reconciling positions across these five sources before they can sign off the number. A break of even a few cents has to be chased back through every system to find the source. Most banks have a team whose actual job is this. The job title is not "reconciliation" but that is the work.
Manufacturing planning
A manufacturing planner managing a single production order is reading from SAP for the master schedule, the MES for actual line output, the quality system for hold status, the supplier portal for component ETAs, and often a separate Maximo instance for equipment availability. The plan on paper says the order is on track. Whether it actually is on track requires the planner to log into five systems and form a judgement. They do this for every active order, every shift.
The systems all think they are the source of truth. The real source of truth is the planner, and the work happens in her head.
What the agent's job actually is
The mistake most teams make when they build into this space is framing the agent as something that answers questions. That is a chatbot, and it puts the planner in the position of asking questions she already knows how to answer. The framing is wrong.
The agent's job is to be the single source of truth the systems refuse to be. It calls the same eleven endpoints the planner would call. It normalises the answers. It holds the consolidated view as a live picture of reality, refreshed on whatever cadence the business runs at. It surfaces only the exceptions worth the planner's attention. The container genuinely going to be late. The position that genuinely does not reconcile. The production order genuinely at risk.
The planner's role then shifts from "find what is wrong" to "decide what to do about what the agent has flagged." That is a meaningfully different job. It is the job she was hired to do. The stitching was never the job, it was the tax.
Why this is the safest first-agent territory
Reconciliation is the right place to start an agent programme for reasons that have nothing to do with how exciting it is and everything to do with how forgiving it is.
The work is read-only. The agent is not deciding anything irreversible. It is not writing to a system of record. It is not sending an email to a customer. It is reading from eleven systems and presenting what it found. The blast radius of a wrong answer is small. A planner who would have caught the discrepancy in the old workflow can still catch it in the new one, because she is still reviewing the agent's output before acting on it.
The failure mode is recoverable. If the agent misreads the carrier API and reports the wrong ETA, the planner notices the same way she would have noticed in the old workflow: by checking the source when the number looks off. The agent does not have to be perfect to be useful. It has to be right often enough that the planner trusts the consolidated view as a starting point rather than rebuilding it from scratch.
The upside is large and measurable. A planner who spends four hours a day on reconciliation today and twenty minutes a day reviewing the agent's flagged exceptions tomorrow has had four-fifths of her week handed back. Output rises. Headcount does not need to. The CFO finally has a number for the cost of the work nobody could name.
The one place this falls apart
Reconciliation agents only work if the source systems have APIs the agent can call. This is the assumption to test before anything else.
In a clean environment where every system has a documented API, the build is a few weeks of integration plumbing and a few more weeks of prompt and output tuning. In an environment where half the systems are accessed through a web portal with no API, where the customs status requires logging in with a username and password that rotates weekly, where the customer's vendor portal blocks programmatic access, the build cost spikes. The agent now needs screen-scraping or RPA, which works until the portal redesigns its login flow. The maintenance burden becomes the new hidden cost.
The first thing we do on any reconciliation engagement is an integration audit. Every system on the planner's tab list, in order, against the question of whether it can be reached programmatically and how stable that access is. If more than a third of the systems are unreachable, the conversation shifts. Sometimes the right answer is to fix the access first and the agent second. Sometimes the right answer is to build the agent against the systems that are reachable and accept the planner will still check the others by hand. The honest answer is decided before any code is written.
The boring agent that pays back fastest
The agents we are most confident recommending are not the most exciting ones. They are not the customer-facing copilots and they are not the autonomous decision-makers. They are the ones that take Priya's four-hour daily ritual of opening eleven tabs and turn it into a twelve-minute review of what the agent has flagged.
The planning stays with the planner. The judgement stays with the planner. The agent does the part of the job nobody put in the job description, the part nobody can see, the part that costs the most. That is where the payback is, and that is where the first agent in any B2B operation should be pointed.