The robots came for the clipboard — the back-office AI wave
Two companies turned the freight back office into software in the same week. The automation you can touch this quarter isn't a driverless truck — it's quoting, vetting, invoicing. The operator map.
Two stories landed in the same week, in different industries, and they turned out to be the same story.
On June 17, C.H. Robinson — the largest third-party logistics company in North America, sitting on a network of 450,000 carriers, 75,000 customers, and roughly $23 billion of freight a year — launched BidBoardX. It’s a self-serve board where certified carriers bid directly on committed freight: the planned, repeatable lanes, “400 loads between two cities on set days,” not the one-off spot loads. Carrier selection — the work that used to live in a broker rep’s inbox and on the phone — became a product you log into.
A few days earlier, Supply Chain Dive published the inside account of how Bristol Myers Squibb rebuilt its procurement around an AI sourcing platform. The numbers are the kind vendors usually inflate, except these came from the customer: RFP cycles that used to take six to nine months now close in under 30 days. Ten times the RFPs, with about half the people. More than a billion dollars of sourcing ran through the platform in the first year.
Different industries. Same frontier. The back office — the quoting, vetting, matching, scheduling, and invoicing that runs on rules and repetition — is becoming software. And here’s the part that matters for an operator: in both cases, a human still signs off. BidBoardX reviews every bid before award. BMS didn’t fire its sourcing team; it gave them ten times the throughput.
That’s the honest version of “AI in freight” for someone who is not buying a driverless truck this year. Last week’s issue was about automation on the road — Gatik and Aurora pulling drivers out of the cab on narrow, mapped lanes. This is automation off the road, and unlike the driverless lanes, it’s available to a five-truck carrier or a two-desk brokerage right now. The trick is knowing where to point it.
”You don’t need your data to be perfect to start.”
— Bristol Myers Squibb’s sourcing team, on why they soft-launched on one workflow and cleaned as they went, rather than waiting for a tidy database that never arrives.
The map
Pick the workflow, not the tool. The mistake is shopping for “an AI” and then looking for somewhere to use it. Start from the work. The tasks that automate first share a profile: repeatable, high-volume, rules-driven, and tolerant of a quick human check. In a freight operation that’s quoting, carrier vetting, invoice and AP matching, check calls and tracking updates, and document classification. The tasks that stay human have the opposite profile: judgment, relationships, exceptions, the load that goes sideways at 4 p.m. on a Friday. This is the same line last week’s issue drew on the road — fixed and repeatable automates, irregular and complex stays human. The back office has its own version of that line, and your job is to find it in your own shop.
Start before your data is clean. This is the Bristol Myers lesson, and it’s the one most operators get wrong. The instinct is to wait — clean up the TMS, standardize the lane names, fix the customer records, then automate. BMS didn’t. They soft-launched on one workflow in late 2024, cleaned as they went, and were company-wide by February. Perfect data is a project that never finishes. A single workflow running on imperfect data is a result you have this quarter.
Keep a human on the tender. Notice that BidBoardX, the most automated carrier-selection product a major 3PL has ever shipped, still puts a rep on every bid before award. That’s not a limitation; it’s the design. Automating outreach and award without a human check is exactly the double-brokering and identity-fraud surface we wrote about in the vetting issue. The right pattern is assist, not autonomy: let the software do the gathering, the matching, the first pass — and let a person make the call that moves money or hands over a load.
Measure one number. Before you turn anything on, write down the metric you expect to move: RFP cycle time, quotes sent per day, days from delivery to invoice, hours spent on check calls. Then check it in 30 days. If it didn’t move, the tool didn’t work, and you’ve spent a month learning that cheaply instead of a year learning it expensively.
What to ignore
“Digital worker.” “Zero-touch.” “Agentic.” That’s vendor language, and it’s describing a future, not a Tuesday. Today’s reality across every honest deployment — including the two anchoring this issue — is assisted work with a person in the loop. Which is the good news, not the disappointment: assisted work is adoptable now, it fails safely, and it doesn’t require you to bet your authority on a model you can’t audit. Buy the assist. Let the autonomy prove itself on someone else’s freight first.
It’s also worth being clear-eyed about who ships these wins first. A pharma giant running a billion dollars through a sourcing platform is not a five-truck carrier, and “$23 billion of freight under management” is C.H. Robinson’s scale, not yours. Don’t read the headline numbers as the bar to clear. Read the method: one workflow, imperfect data, a human on the decision, one metric. That part transfers down to any size operation.
The move
You don’t need a strategy. You need one task. Look at your week and find the slowest, most repetitive thing your back office does — the one that eats an afternoon and runs on rules. Trial one tool against it before July. Keep yourself on the decisions that matter. Measure the one number.
The robots came for the clipboard, not the cab. That’s the frontier you can actually touch this quarter — and the operators who get leverage out of it won’t be the ones who bought the flashiest “autonomous back office.” They’ll be the ones who automated one boring, repeatable task this month, kept a hand on the tender, and moved on to the next one.
Sources
Read this in full each Tuesday