Real S&OP optimization with AI starts with routing rules nobody has re-examined in years

Thousands of shipments a year still default to premium freight when standard transit would have arrived on time, and no algorithm put them there. That gap is where S&OP optimization with AI actually begins.

Inside one large manufacturer's transportation system, a routing rule written years ago still sends ordinary orders out on premium freight lanes. Not because the orders are late, and not because anyone pays for rush service on purpose: the rule simply names the expedited lane as the default, and nobody in the current supply chain organization has reopened it since it was set. Standard ground transit would clear the delivery windows with days to spare. The orders ship the way they always have, because that is what the system tells them to do.

The default that quietly became expensive

A review of one manufacturer's own freight transaction data turned up thousands of shipments a year defaulting to premium modes when standard transit would have cleared the requirement with room to spare. None of these were emergencies. They were ordinary orders moving on an extraordinary lane, assigned when the network looked different: fewer carriers, a different plant footprint, service agreements nobody had renegotiated.

Nobody had made a bad decision so much as no decision at all, for years running. The rule sat inside a transportation management system doing exactly what it was configured to do, while the conditions behind it quietly expired. Finding the pattern took someone willing to pull lane-level data and ask, order by order, whether the mode still matched the requirement, tedious and unglamorous work that most transformation programs skip on the way to a pilot.

Every order pushed through the distribution center, even when direct is faster

The same audit found a second default underneath the first: predictable, repeat orders that could ship directly from a plant or supplier were instead routed through a regional distribution center as a matter of course, adding a handling cycle and transit days to shipments that needed neither. Direct shipment was faster and cheaper for a meaningful share of this volume. The network routed it through the DC anyway, because DC-first was the default path and nothing in the system distinguished a predictable order from an exceptional one.

That distinction used to be somebody's job. As the network scaled and the routing logic hardened into policy, the judgment call disappeared and the default took over. The cost showed up twice: once in freight, and again in the working capital tied up in a warehousing step a direct shipment never needed.

Dead stock, misplaced inventory, and the case for S&OP optimization with AI

The third pattern was dead and misplaced inventory: positions accruing carrying cost in locations where network demand had moved on, or had never existed at all. Some of it was product built against a stale forecast. The rest sat correctly counted on a balance sheet, disconnected from where the next order would actually ship from.

This is the kind of position that survives audit after audit precisely because it does not look urgent. Nothing is broken. Money is simply sitting still, in a warehouse, waiting on demand that a functioning sales and operations planning process would have flagged as unlikely to arrive.

Where S&OP optimization with AI actually adds value

Fixing the first three problems required no artificial intelligence at all. Going lane by lane and order type by order type through the transaction data, correcting the premium-freight defaults, the forced DC routing, and the dead stock recovered a mid-single-digit percentage of a multi-million-dollar annual freight-and-warehousing cost base before a single model touched the network. The largest tranche of value here was discipline, applied to data the company already owned.

The harder problem is that networks do not hold still. Carriers reprice, demand shifts by season, a distribution center closes, and rules built around old conditions go stale again within a year. "Automate a bad routing rule and you just get the wrong answer faster," says Matt Leta, founder of Future Works. "Fix the defaults against the real transaction data first, verify it with finance, then let the agents keep it fixed." A different response is emerging among enterprise operators: fix the defaults against real transaction data first, verify the fix with finance, and only then hand the ongoing maintenance to AI agents plus named experts working in 12-week cycles rather than an annual planning calendar. An AI-native transformation engine runs that sequence on outcome-staked terms, a third of the fee tied to a value target finance signs off on, so the agents are paid to keep a fixed default fixed, not to invent it.

Continuous re-optimization is not a switch flipped once and forgotten. An agent fleet re-checking allocation and routing every cycle still needs a named owner reviewing its calls, the kind of AI-powered control tower visibility that surfaces a drifting rule before it costs anything, or a model optimizing against last year's cost assumptions will happily rebuild the same premium-freight default it was deployed to remove. That kind of oversight is not glamorous either, but it is the difference between a fix that holds and one that quietly drifts back to where it started.

None of the first three fixes required a foundation model or a program measured in years. The premium-freight lanes, the DC-first routing, and the aging stock were all sitting in the network's own transaction history; what was missing was authority to re-ask the questions and a finance team willing to validate the answer. The blockers behind that kind of stalled value are rarely about model capability, and IDC's research this year points instead to data quality, integration complexity, change management, and unclear ownership. Skip that order and deploy the agents before the defaults are fixed, and an AI driven supply chain transformation just automates the mistake faster. Get the sequence right, and the same agents keep a good rule current instead of optimizing around a bad one.

For a CSCO or COO deciding where to look first, the routing rules already running the network cost nothing extra to re-examine, and Future Works stakes a third of its fee on proving the value of doing exactly that. Start with a proof-of-value pilot.

Matt Leta, Founder and CEO, Future Works.