The coordination tax is real for mid-market organizations.

If you run a mid-market firm, let’s say 200 people, the fastest way to map a broken handoff is to put a rough agent on it in the first few weeks and let the flags show you where it breaks. McKinsey’s new coordination tax piece is right about the problem, and they measured it well: in a seven-step planning workflow at a Fortune 500 manufacturer, the work inside the steps took 12 to 24 hours while the waiting between them took 9 to 18 days. That diagnosis holds at mid-market size, and it holds across industries, because an association’s renewal handoff and a services firm’s deal handoff wait on the same thing: a person checking by hand. Where I part ways is the cure, because it’s written for their client, with a COO-sponsored cross-functional team, six to 12 months of redesign, and in their own example four months of mapping and classifying before anyone configured any technology. That’s the right order for that reader. At 200 people you don’t have that team or that runway, and you don’t need them, so the order flips.

Why the order flips at this size

Their order makes sense for their client, with more than 50 planners and 75 spreadsheets and legacy tools in one workflow. Nobody in that building can hold the whole thing in their head, so you map first.

At 200 people, two or three people own the handoff and they already know where it breaks. What’s missing is a written record of the rules they apply by hand, because those rules live in one person’s head, and a workshop asks that person to remember them. Put a first-pass agent on the checks the receiving side already does manually, send every failure to a shared channel with a one-line reason, and inside two weeks you can see whether the mismatches cluster. That cluster is the map you were going to spend four months drawing. The owner usually goes in assuming the problem is data entry, and the flags usually point at a policy nobody wrote down.

What it looked like when we did it

We did this at a professional services firm, on the handoff from sales to delivery. A partner closes a deal, and the signed SOW has to become a project with a budget, a kickoff date, and a team that knows what was promised. Before, the delivery lead rebuilt all of that by hand in the project tool, from the SOW and whatever the seller remembered.

When a deal flips to closed-won in the CRM, the agent reads the SOW, and it also reads the call transcripts and the email thread, because the commitments that cause trouble in month three are the ones a client heard on a sales call that never reached paper. It compares all of that to the project setup and flags anything missing to the delivery lead and the seller together.

Time to the first flag was about seven business days. One builder spent 15 to 30 hours. The delivery lead spent two to four. The slowest part of the whole thing was getting access approved, which is a handoff, which I find pretty funny.

The first two weeks produced a handful of flags, and the ones that mattered were things said on a call that the delivery team didn’t know about. The delivery lead slowed down, went back to the seller, and got the SOW to be the source of truth before kickoff instead of a document both sides reinterpret later. The agent is still running.

The objection, and the part of it that’s right

A McKinsey partner would say I’m automating a broken process, and BCG argues the same thing: redesign the whole process before agents go in. Ethan Mollick is closer to me. He wrote “build it and see where it fails” last year, but he assumes a dedicated AI lab doing it across all the work, and a 200-person firm has neither.

The real risk is that a prototype that works well enough quietly becomes the process, and nobody ever does the redesign. For a while my answer was that the prototype costs less than the mapping exercise and you can throw it away. That’s an argument, not an experience, and it leaves out the real cost. Build hours are cheap. Trust isn’t. Flood a delivery lead with flags that turn out to be noise and the second agent gets ignored before it starts. Hospitals have measured this for years: across 112 clinicians in one study, the chance a reminder was acted on fell 30 percent for every additional reminder in the same visit, and how busy the clinician was made no difference at all.

So “you can throw it away” has to mean deciding what would make you throw it away before you build. Write two or three kill criteria on day one with the two people who own the handoff, and tell them this is a test with an end date. After two weeks: are most flags real problems, did they cluster, did either owner change how they work because of one. If it’s no across the board, you stop. Then you work out which way it failed, because the two failures need opposite fixes. Either the checks were wrong, so you fix the checks and keep the handoff. Or the checks were fine and almost nothing fired, which means the pain is somewhere else. A failed first try costs a couple of weeks. It only costs trust if nobody agreed up front that it might fail.

The redesign still happens. It happens in week three, when you sit the two owners down with the cluster of flags and ask what the rule should be. You can’t have that conversation in week one, because nobody knows what the rule is yet.

Who owns it

The two people who live on each side of the handoff, not their bosses. They know the unwritten rules, and each department head protects their own side when the handoff is the part in between. One builder who knows the business, and whoever owns operations as sponsor, to clear access and decide early what happens to the hours that come back. IT stays close on access and security without owning the project, and if the agent reads call transcripts, the people on those calls need to know, since that question is in court right now. Your software vendor’s AI add-on won’t do this for you. It reaches other systems only through whatever the vendor chose to connect, mostly read-only and behind an admin switch, and it has no idea what your handoff’s rules are, which is the whole reason you’re building a first pass of your own.

At this size, coordinating a handoff is a slice of someone’s job, usually the nagging slice. The hours that come back go to the work that was getting squeezed. The job doesn’t go away. It gets done the way it was supposed to.

McKinsey’s own priorities put roughly 70 percent of the effort in these redesigns on people and process rather than technology, and that number is right. At 200 people you get there by watching two weeks of flags, not by running four months of workshops. Pick one handoff that happens weekly, put the people who live in it on it, write down what would make you stop, and let the exceptions draw the map.