Deloitte found that 30% of retailers surveyed use AI for supply chain visibility, expected to rise to 41% within a year, with 59% of executives anticipating positive return on investment from supply chain initiatives within 12 months.
Most of that investment is described as making logistics faster. The more useful question is narrower. Which decisions actually expire if they are not made immediately, and what does it take to make those four in the time available?
What real-time decision making means here
Real-time decision making in a hub network means changing an allocation while the freight it concerns is still physically present and still able to move differently.
That definition sets a hard boundary. A decision made after the vehicle has departed is not a fast decision. It is a report.
MIXMOVE frames the test that matters commercially. A decision is real-time if acting on it is still possible. Everything else is analysis, and analysis can wait for the overnight cycle without costing anything.
Why static planning held for so long
Overnight planning suited demand that closed at a cut-off. Orders arrived through the day, were planned as a complete set, and executed the following day against known volumes.
Buffers covered whatever the plan got wrong. Stock at the spoke, slack in the schedule, and spare capacity at the hub all absorbed the difference between the plan and the day.
Both conditions have gone. Orders now arrive after the plan is fixed, and the buffers were removed for cost reasons. The gap between plan and reality now converts directly into dwell, rework, and missed commitments.
The four decisions that expire
Not everything needs to be fast. These four do, and each has a point beyond which it cannot be recovered.
Flow or store. Whether an arriving unit crosses the dock or goes into storage. Expires the moment unloading begins, because a unit put away has already consumed the handling the decision was meant to avoid.
Load composition. What goes on the outbound vehicle. Expires at departure, and every minute before departure is an opportunity to include freight that has just arrived.
Release timing. Whether to hold a load for consolidation or dispatch it now. Expires continuously, because the latest safe release time moves as the delivery window approaches.
Reallocation under disruption. Where freight goes when a spoke is congested, a vehicle fails, or a window is missed. Expires when the alternative capacity is committed to something else.
All four are made on a dock, with freight physically present, in a window measured in minutes.
Why speed alone does not help
Faster recalculation is a common response and an incomplete one.
A recalculated plan still has to reach the person moving the freight. Where the instruction arrives after the pallet has been staged, the speed of the calculation is irrelevant. The constraint moves from computation to communication, and communication is where most implementations stop.
McKinsey research into mid-mile and last-mile handovers found that waste created at blind handoffs accounts for between 6% and 13% of carrier revenue, with dwell time named as a leading driver. Dwell is what a correct decision looks like when it arrives too late.
The five conditions a fast decision requires
Item-level knowledge of what is present. Consignment-level records are too coarse. Reallocation happens at the unit, so the record must exist at the unit.
A known commitment for every unit. The latest safe decision point can only be calculated against a real delivery obligation.
Live downstream capacity. Reallocating into a site that cannot receive the freight relocates the problem.
Instruction delivery at the point of work. The decision must reach the dock team while the freight is still on the floor. This is the condition most often missing.
Reversibility inside the window. Conditions keep changing. A decision that cannot be unwound forces conservative choices, which costs fill rate every single day.
What holds adoption back
Three conditions, all addressable, and none of them about willingness.
Execution data describes plans rather than events. Where the record shows what was scheduled instead of what happened, recommendations are built on an inaccurate picture of the network and behave accordingly.
Recommendations arrive without the basis for them. A suggested reallocation with no visible reasoning cannot be checked against what the dock team can see in front of them. Systems that show the constraint driving the recommendation get acted on.
Integration is partial. Where the fast layer sees only some of the network, its recommendations are correct locally and wrong overall, which erodes confidence quickly and reasonably.
What the evidence shows
Deloitte reports that 30% of retailers surveyed use AI for supply chain visibility, expected to reach 41% within a year, with 59% anticipating positive return on investment within 12 months and 66% planning supply chain restructuring if input costs rise.
McKinsey attributes between 6% and 13% of carrier revenue to waste at handover points.
McKinsey and the World Economic Forum put the coordinated scenario at 30% lower delivery emissions and congestion by 2030, alongside a 25% reduction in delivery cost through technology.
Across MIXMOVE deployments, hub operations have recorded dwell time reductions of 40%, fill rate improvements of 10% to 20%, up to 130% higher warehouse hub throughput, up to 80% fewer errors, and up to 58% labour cost savings. The platform is in use across 35+ distribution companies in 20+ countries.
At 3M, a decade of collaboration produced a 90% truck fill rate, a 35% reduction in transport costs, and a 50% reduction in CO₂ emissions.
“By using the MIXMOVE software, 3M managed to reduce transport costs by 35% and CO₂ emissions by 50%.”
— Patrick Van De Vyver, Former Head of EMEA Logistics Operations, 3M
How MIXMOVE HUB OS closes the decision loop
MIXMOVE HUB OS is built around the four expiring decisions rather than around faster planning.
Freight is identified at item level on arrival and matched against live outbound commitments before unloading, which is where the flow-or-store decision is made rather than discovered. Load composition is built against current network state, so freight arriving late can still be included. Release timing is calculated per load against real delivery windows. Reallocation under disruption happens per shipment, leaving the rest of the network undisturbed.
The instruction reaches the dock team at the point of work, which is the condition that determines whether any of the above changes anything. Recorded dwell reductions of 40% come from that last step rather than from the calculation.
MIXMOVE HUB OS operates alongside an existing TMS, WMS, or ERP as an orchestration layer, or as a standalone platform.
MIXMOVE DI reports the outcome, including dwell by site and utilisation by leg, alongside Scope 3 transport reporting built to ISO 14083 methodology.
Four decisions expire on the dock. Everything else can wait for the overnight cycle. Networks that resolve those four in the time available stop paying for the ones they missed.
Read the MIXMOVE HUB OS overview to see how flow, load, release, and reallocation decisions are made per shipment.
Frequently asked questions
What is real-time decision making in logistics?
Changing an allocation while the freight it concerns is still physically present and still able to move differently. A decision made after departure is a report rather than a decision.
Which logistics decisions actually need to be made in real time?
Whether an arriving unit flows or is stored, what goes on the outbound vehicle, when the load is released, and where freight is reallocated when a disruption occurs. All four expire while the freight is on the dock.
Is faster planning the same as real-time decision making?
No. Faster recalculation produces a new plan sooner, but the instruction still has to reach the dock team before the freight moves. Where it does not, calculation speed changes nothing.
What prevents real-time decisions from working?
Execution records that describe plans rather than events, recommendations presented without the reasoning behind them, and partial integration that makes decisions correct locally and wrong across the network.
