Growth Is Addition. Scale Is Multiplication.
Growth is addition: more volume, more resources, more cost, roughly in lockstep. Double the work and you roughly double what it takes to carry it. Growth is real, often necessary, and stubbornly linear.
Resources means more than people, though people are what everyone counts first. It is also the licenses, the systems, the working capital tied up while a case sits open, the vendor capacity you reserve against a busy quarter, and the management attention consumed keeping all of it coordinated. Each of those tends to rise with volume, which is why the cost curve so rarely bends on its own.
Scale is multiplication: output rises while the resources behind it stay flat or grow far more slowly. Double the work and you add a little. Scale bends the curve.
Growth adds resources at the pace of demand. Scale adds output faster than resources.
Here’s the test I put to ops leaders: if your volume doubled next quarter, would your cost double with it? If the honest answer is yes, you’re set up to grow, not to scale.
Your Operation Runs on Judgment, and Judgment Lives in People
Software scales, because one product serves millions at almost no extra cost. Your operation doesn’t, and it is worth being honest about why. The core of operational work (approvals granted, claims processed, suppliers coordinated, exceptions resolved) runs on human judgment and coordination, and that lives in people. Every extra unit of volume needs another unit of person to carry the judgment.
So when volume rises you hire, and when you hire, coordination overhead rises with it: more handoffs, more meetings, more places for something to slip. Cost climbs, complexity climbs and risk climbs together. That isn’t a failure of your team. It is the shape of work that depends on a person for every decision.
Hiring is a growth strategy pretending to be a scaling one.
Economists named this sixty years ago. William Baumol observed that a string quartet needs the same four musicians for the same thirty minutes as it did in 1800, so as wages rise everywhere else the quartet simply becomes more expensive to put on. He called it cost disease, and the sectors he identified as structurally stuck were the labor-intensive ones: healthcare, education, care work, and anything else where a person is the unit of delivery. His conclusion was not that these industries were badly run. It was that they could not get productivity out of the work itself.
Baumol's point was not that these industries were badly run. It was that they could not get productivity out of the work itself, because the work was inseparable from the person doing it. That is the asymmetry we opened with, stated as economics.
That diagnosis has held for six decades. This is the wall, and no amount of “efficiency” gets you over it.
Efficiency Is Growth With a Discount
AI and automation projects love to promise scale, then quietly deliver something smaller: a little more efficiency. They shave a few points off a task here and speed up a step there. Useful, but the curve doesn’t change shape. You are still adding cost at roughly the same rate as volume, just slightly less of it.
You can see the wall in the data. In the AMA’s survey, physicians reported completing roughly 39 prior authorizations a week and spending about 13 hours on them, and 40% now keep staff who do nothing but prior authorizations. When the work depends on people, the only way to handle more of it is to assign more of them. That is growth, not scale.
Efficiency is growth with a discount, not scale. Scaling requires changing what carries the volume, rather than making the same linear process a bit cheaper.
What Changes the Shape Is Taking People Off the Volume
The operations that break the wall do one thing: they stop requiring a person for every unit of volume, without giving up the judgment that made the work trustworthy. The routine reading, routing, drafting, checking and coordinating gets carried by intelligence, while the judgment calls that a person must own still route to a person. Volume stops being chained to headcount, and control stays where it belongs.
That is the shift this series is about, and it has a name: scaling your operations with governed intelligence. “Governed” isn’t a footnote. It is the reason this is scale and not exposure, because intelligence without control scales your risk as fast as your output.
This is the piece Baumol's stagnant sectors never had. Agents absorb units of work on their own and return only what needs judgment. That is not a faster tool for the worker, which is what software has been for forty years. It is capital you can finally place between the volume and the person.
Agents are to knowledge services what robots were to manufacturing.
If that holds, it is the first serious challenge to cost disease in sixty years. Not for every labor-intensive trade, because the quartet still needs its four musicians. But the claims desk may not, and neither may the procurement desk, the prior-auth queue or the reconciliation team.
Picture a prior-authorization team the week demand spikes against a fixed CMS turnaround (the US prior-auth mandate): more requests, the same clock, and no way to hire fast enough to clear the queue safely. Put a governed layer over that workflow and the routine submissions, payer-rule checks and status chases run automatically, and only the clinically ambiguous or high-risk cases route to a human reviewer. The team holds the deadline without launching a hiring drive.
I watched the same shape in a procurement operation handling long-lead, high-stakes sourcing. Consider their lead buyer on two different Tuesdays.
On the first, she arrives to eleven supplier updates, three of which matter. She spends the morning working out which three: opening attachments, cross-checking promised dates against the build schedule, chasing two suppliers who replied on the wrong thread. She makes her first real decision at two in the afternoon.
On the second Tuesday, two years later, those same eleven updates have already been read, cross-checked and ranked. The three that threaten the schedule are sitting at the top with the evidence attached. She makes her first real decision before her coffee goes cold, and four more after it.
Nothing about her judgment changed. She still knows which supplier slips when the rains come, and which lead-time update actually threatens the build. What changed is that she no longer spends her morning assembling the information she needs in order to use it. Over those two years volume tripled and headcount did not. They didn’t grow three-fold, they scaled.
Three Tests You Can Run This Week
Run the doubling test. Take one workflow and ask what you would actually do if its volume doubled next quarter. If the honest answer is “add more of something” (people, a shift, a vendor, another license), you have found something built to grow rather than scale.
Find the linear core. Locate where throughput is chained to a resource you have to keep adding, which is usually people, because judgment and coordination live in them. That is your scaling frontier: not the tasks you could shave, but the decisions that currently require a person.
Separate volume from judgment. In that workflow, list what is genuinely a judgment call versus routine execution wrapped around it. The gap between the two is where scale is hiding.
Every Operation Grows. Few Learn to Scale.
Every operation will keep growing demand makes sure of that. The real question is whether it will also learn to scale: to bend the curve on the work that has always grown in a straight line. That’s the purpose underneath everything else we’ll cover in this series not doing a bit more with a bit less, but changing the shape of what your operation can carry.
So before your next capacity conversation, sit with the real question: are you building to grow, or to scale? Then tell me in the comments which of your workflows would break first if volume doubled, and could you fix it without a hiring plan?
Recent blogs
Secure your agents
We’d love to chat with you about how your team can secure and govern Ai agents everywhere
