Digital Operations: The Next Operating Model for Enterprise Logistics

“We’re being thoughtful about AI.” 

I hear that line constantly from brokerage presidents, week after week. But by this time, I’ve finally figured out what it really means. It’s a pilot that started in March, a Slack channel, a deck, a standing Thursday meeting, and zero loads anyone would trust it with. Five months of “thoughtful.” 

At some point, thoughtful is just a nicer word for stuck.

I sell the other side of this, obviously. My company Envoy builds Ellie, and she’s an execution layer that functions almost like a carrier rep. She lives in the same browser tabs your team already works in, sources the trucks, works the email and the texts and the phones all at once, negotiates inside whatever guardrails you give her, and checks MC and DOT before anything books. Your rep still approves the move and still owns the relationship. In our live accounts, Ellie books about 75% of the freight.

The machine exists, and it works, but walk any brokerage floor and try to find the person whose job is managing it. We spent a hundred years building management science for people and about 20 minutes on management for machines. 

So I’m writing the missing discipline down. Digital Operations is how you run work that humans and AI do together, the role Lean played for factories and DevOps for code. I made the argument in my substack about The Boss of Machines. But now this is the how behind it all. 

And yeah, this field manual comes from the guy selling the machine. Somebody had to do it.

Digital Operations Is a Management Job, Not a Software Purchase

I give it two quarters before every vendor in the space is wearing this phrase like a conference lanyard, so let me define it while it’s still clean. Digital Operations is how you design, run, and measure work that’s split between people and AI, and you can’t buy it anywhere. There’s no SKU, and believe me, we’d sell one if we could.

You build it out of decisions instead, none of them complicated, all of them unwritten. Somebody has to say what the machine runs from start to finish, where its authority stops, who’s watching, and who catches the load it can’t close when the TMS is down and the carrier won’t answer. Nobody ever wrote that down, because a person came attached to every task and carried the rules in her head. That held up right until the person wasn’t the one doing the task.

The stack sorts itself after that: your TMS stays the system of record, the execution layer does the work, and Digital Operations sits over both. Even HBR admitted in February that managing this is a real job rather than a hobby for IT. Four years late, but I’ll take it.

Lean and DevOps Already Ran This Play

Every AI vendor quotes Toyota now, me included, so at least hear the part they all skip. 

Toyota’s machines were never the advantage; Ford and GM had machines too. What Toyota had that they didn’t, though, was a way of thinking about work. John Shook, the first American the company ever hired, told an MIT Sloan class that treating Lean like a toolkit is exactly how you miss it. Half of Detroit proved him right, buying the kanban boards and getting a poster program. Toyota got the decade.

Software ran the same arc 20 years later while I watched from the freight side, a little jealous. The CI tooling went commodity almost immediately; the discipline around it never did, and DORA’s 2025 survey of nearly 5,000 engineers landed on the line that explains both eras: AI amplifies whatever a team already is.

Now the same movie is playing in freight, fast-forwarded. The very morning my team met about our copycat problem, a new one launched with our vocabulary in its homepage hero. 

You can’t buy timing like that, and I’m not even mad. The tools were always going to spread. The discipline is still sitting there unclaimed.

Most AI Pilots Deserved to Fail

MIT tore through 300 enterprise deployments in 2025 and found that 95% produced nothing a CFO could see. The lead author, Aditya Challapally, blamed the wiring rather than the technology. 

I’d push further, though, because half the pilots I hear about exist so somebody can put “led AI initiative” on a QBR slide. If a vendor burned you in 2024, the pitch may well have been garbage. There’s also a fair chance nobody inside your building truly owned the deployment. Both failures can show up in the same postmortem.

McKinsey’s numbers help explain why this keeps happening. AI now appears somewhere inside 88% of companies, while only about 1-in-5 has redesigned a workflow around it. Most businesses are still dropping a machine into a process built for humans in 2009, leaving every old handoff and approval in place, then wondering why they get 2009 back with faster typing.

I get to say that because I ran the human version. 

I’ve told the story about paying people to send emails so many times my team winces, but the part I tell less often is that I also sat in the meetings where we agreed to fix it later. The problem is, later is what cost me more than the emails ever did. The loads we lost came down to decision latency, not effort, and no amount of outreach fixes that.

Onboard the Machine Like You’d Onboard a Rep

The fix turns out to be familiar, because it’s the playbook you already run every time you hire. A new rep gets a job description, limits on what she can commit to, a manager who reviews her work, and somebody to grab when a load gets weird. Digital Operations is refusing to skip those four steps because the new hire happens to be software.

With Ellie, that gets concrete immediately. Her job description covers the ask through sourcing, negotiation, compliance, and track-and-trace, written down rather than assumed. Her limits are your guardrails: max pay by lane, margin floor, autonomy dialed per rep or per customer, and a quick way to test whether yours are real is to ask who’s allowed to change a max pay setting. 

If the answer takes longer than five seconds, those guardrails are vibes.

The manager who reviews her work is your observability layer, or TOAS in our world. And it’s why she shows you the rate before she sends it. The person to grab is an escalation path with a name on it. Gartner’s 2026 list finally put decision governance beside autonomous execution, so the analysts have caught up to any decent floor manager.

Put the Machine on a Scorecard

You wouldn’t let a new hire work for six months without knowing whether they were pulling their weight. Digital Operations deserves the same scrutiny. Grade the system on the work it finishes, not how clever its answers sound: loads closed without human help, exceptions cleared, time to resolution, and margin per load. Once those figures sit beside your team’s numbers on the ops dashboard, the machine’s become part of the operation.

FreightWaves made the same point at its July symposium: that roughly 80% of supply chain work is repetitive while operations get won or lost on the 20% that require judgment. Most shops staff that ratio backward, their sharpest people grinding the 80 while the 20 gets whatever’s left of the afternoon.

The margin math is what makes this urgent rather than interesting. FreightWaves also worked out in January that a brokerage needs about $210 of gross margin per load to break even on $1,912 of revenue. At spreads that thin, an unwatched machine is a liability while a watched one is the whole play. Your best reps come out of it as exception specialists, a better job than the one they have today.

The Judgment You Throw Away at 6 p.m.

One more thing a well-managed hire does that an unmanaged one never will: they remember. The same MIT team traced most stalled AI back to systems that retain nothing and repeat their mistakes, a fair description of every generic tool ever parachuted into a brokerage.

Watch your own floor for an hour. A driver calls in a flat tire, and your best rep already knows the delivery appointment just collapsed, so they’re rescheduling it before dispatch even finishes the sentence. Nobody wrote that rule anywhere. It lives in their head, and at 6 p.m. it drives home with them. Captured and structured, it compounds into the Carrier Context Graph, a memory layer your brokerage owns outright and no competitor can ever buy a copy of.

That might be the last real moat left in freight, because the second AI vendor you ever hire will underperform the first unless it inherits your context. And you’re not early anymore. Truckstop and Bloomberg Intelligence surveyed 187 brokers last fall and found 41% deploying against 48% holding out, a split that closes this year with half of you compounding first.

Envoy Built the Execution Layer. Digital Operations Is How You Run It

I sell the machine, so run whatever discount you want on everything I’ve said. What survives the discount: Lean took a decade to conquer automotive, DevOps took about that long in software, and freight isn’t going to get a decade because the hard part already runs. The execution layer is booking three of every four loads in live accounts while an operator approves each rate and keeps each relationship.

I said something on a team call recently that I’ll put in writing: no one cares about AI. What people care about is executing and growing in a nonlinear way without stacking headcount. Everything past that is vendor theater. Mine included when I’m not careful. Ellie loads as a Chrome extension over the TMS, portals, and load boards your team already lives in, with no rip-and-replace, no new tab, SOC 2 Type II, and a human on every rate.

So go find that pilot from March. Give it a job description, limits, a reviewer, and an escalation path by Labor Day. Or shut it down and keep the money. Either one counts as a decision, which is more than most of the market has made this year.

Skip the nurture sequence. Book a demo, bring a real portal and a live lane, and we’ll cover it in front of you with no sandbox and no sizzle reel. Put one test to every vendor on your list, including us: how much of your customers’ freight runs through your tool today, in production? Most of the category goes quiet right there. Our answer is 75.

And if you’d rather pick a fight with the premise before you book anything, even better. I answer my own email: robby@tryenvoy.ai.