Supply Chain Optimization: The Part of Cost to Serve Nobody Models

Supply chain optimization means lowering total cost to serve without breaking service levels, across demand planning, sourcing, inventory, production, warehousing, network design, transportation and final delivery. Planning suites model the first six well. The execution layer, where cost per drop, first attempt delivery rate and on time in full are actually decided, is where most optimization programmes lose their gains.

Every supply chain plan is a forecast until the day it gets delivered. The one number that survives the trip from the board deck to the loading dock is cost to serve.

Most content on this topic optimises the parts of cost to serve that model cleanly: demand signals, safety stock levels, warehouse siting, freight lane rates. Those are real levers and they matter. But the final leg, the part with the most variability and the least central control, usually gets a paragraph and a nod toward "visibility".

This article covers the whole chain properly, then spends its second half on the layer where plans meet drivers, addresses, delivery windows and locked gates.

What Supply Chain Optimization Actually Covers, End to End

Supply chain optimization is the continuous work of reducing total cost while holding or improving service, across seven connected domains. Skip one and the savings just move somewhere else on the P&L.

Demand planning. Forecasting what will be ordered, by whom, when. Everything downstream inherits the error in this number.

Sourcing and procurement. Supplier selection, order quantities, lead time reliability, dual sourcing for risk. A cheaper supplier with a volatile lead time is often a more expensive supplier.

Inventory. How much stock, in what form, held where. Safety stock is a purchased insurance policy against forecast error and supplier variability.

Production and capacity. Batch sizes, changeover time, scheduling against demand. Relevant if you make things, ignorable if you only move them.

Warehousing and handling. Slotting, pick paths, put away, cross docking to skip storage entirely for fast moving lines. Handling cost per unit is one of the easiest numbers to improve and one of the least measured.

Network design. Where the depots, hubs and stockholding points sit relative to demand. This decision fixes your transport cost floor for years.

Transportation and final delivery. Mode, carrier, load, route, dispatch, proof of delivery, returns. This is the domain that turns a plan into a completed order.

Now the scope honesty, because vendor content rarely offers it. If you run manufacturing, sourcing or multi echelon inventory, the advice in the second half of this article is not your highest priority lever. Go and fix forecast accuracy and supplier lead time variability first.

If you dispatch vehicles to customers, sites or collection points, the second half is where your money is.

Cost to Serve: The One Metric That Links Planning Decisions to Delivery Reality

Cost to serve aggregates inventory holding, warehousing, handling, freight and final delivery cost down to a single customer or a single order. That is why it is the only metric that connects a network design decision to what happened on a driver's run sheet on Tuesday.

Most operations know their gross margin per customer. Far fewer know their net margin per customer after delivery.

The gap between those two numbers is where the unprofitable accounts hide. The retail chain that orders small quantities twice a week. The site that only accepts deliveries between 7am and 9am. The rural drop that adds ninety minutes to a route for one carton.

None of those customers look expensive in a planning model. They are all expensive in a vehicle.

Planning systems handle structured variables well: rates, volumes, lead times, capacities. They handle volatile execution conditions poorly, because those conditions change hourly. Traffic, a driver calling in sick, a customer who moved their receiving window, a gate code that stopped working.

Optimization that never reaches the volatile layer produces a great model and an average operation.

Supply Chain vs Logistics vs Transportation vs Route Optimization

These four terms get used interchangeably in software marketing and they mean genuinely different things. Getting the distinction right determines what you buy and who owns the result.

Layer	Scope	Typical owner	Time horizon	Typical system	Example decision
Supply chain optimization	What you buy, make, hold, store and move, end to end	Supply chain director, COO	Quarters to years	ERP plus planning suite	Should we hold stock in a second state or pay more freight
Logistics optimization	Moving and storing goods: warehouse, freight, delivery, returns	Logistics manager	Months to quarters	WMS plus TMS	Do we cross dock this line instead of storing it
Transportation optimization	Mode, carrier, load build, linehaul and delivery cost	Transport manager	Weeks to months	TMS, carrier portals, telematics	Own fleet or third party carrier on this lane
Route optimization	Sequencing stops for individual vehicles under real constraints	Dispatcher	Today, and again this afternoon	Delivery management platform	Which of six vans takes the 11am to 1pm window drop

Where Supply Chain Optimization Breaks Down: The Execution Layer

McKinsey's research into last mile delivery economics identifies the final leg as the most cost intensive part of the parcel journey, with stop density and failed first attempts as the dominant cost drivers. That finding is specific to parcel and delivery networks, not to every supply chain, and it should be read that way.

But if you run vehicles to customers, it describes your business exactly.

Three things reliably break at the plan to execution handoff.

First, data quality. Addresses that geocode to the wrong side of a dual carriageway. Service times set at a flat ten minutes for every stop when a pallet drop at a construction site takes forty. Garbage constraints produce a route that looks efficient in the office and fails by 11am.

Second, integration. If orders arrive in the dispatch system by spreadsheet or retyping, the plan is already stale when it is built. Every manual re-entry point is a place where the plan and reality separate.

Third, measurement cadence. If cost per drop is reviewed monthly, a bad routing pattern gets thirty days to run before anyone sees it in the numbers.

The World Bank Logistics Performance Index, in its 2023 edition, scores economies across six dimensions including timeliness and tracking and tracing. Traceability and on time performance are internationally benchmarked measures of logistics quality, not vendor talking points. If your operation cannot report on either, the optimization conversation has not started.

Six Transportation and Delivery Execution Levers, Including Reverse Flows

1. Mode selection. Road, rail, air, sea, or a mix. Set by lead time promise against unit cost. Revisit whenever a lane's volume profile changes, not just at tender time.

2. Carrier mix. Own fleet for dense, repeatable, service sensitive work. Third party carriers for peaks, long tail postcodes and volatile volume. Most operations get this backwards and run their own vehicles into the sparse work.

3. Load and capacity planning. Building loads by weight and volume together, not by order count. A van at 100% volume and 40% weight is a full van.

4. Route optimization as a constraint problem. This is a computational instance of the Vehicle Routing Problem, solved under delivery time windows, vehicle capacity, driver shift length, service duration at each stop, and required skills or vehicle type. A tool that only sorts stops by proximity is not solving that problem. Constraint handling is the whole game.

5. Dynamic dispatch and re-sequencing. The morning plan degrades from the first delay. Being able to reassign a stop mid shift, or re-sequence the back half of a run, is the difference between one late delivery and nine.

6. Real time visibility and proof of delivery. Live status, honest ETAs to the customer, and geo-stamped photo or signature capture at the door. Control tower dashboards are useful once the underlying event data is trustworthy, and worthless before that. Digital twins and scenario planning belong in the same category: real capability, wrong starting point for most operators.

Reverse logistics and scheduled collections

Returns, collections and recovery flows are a standard part of any complete end to end supply chain definition. They are also almost invisible in the leading definitional pages on this topic.

That is a strange omission, because collection routing is harder than delivery routing. Volumes are less predictable, the item is often not the reason for the visit, and a missed collection is rarely rebookable at zero cost.

Treat reverse flows as first class stops in the same optimization run, not as an afterthought handled by whoever has spare capacity.

Supply Chain Optimization KPIs, Formulas and Review Cadence

Concepts without a measurement plan produce meeting language, not change. Here is the scorecard, with where each number actually comes from.

Metric	Definition and formula	Data source	Review cadence
On time in full	Orders delivered on time and complete divided by total orders	ERP or OMS promise date against POD timestamp	Weekly
Perfect order rate	Orders on time, complete, undamaged and correctly documented divided by total orders	ERP, WMS, POD records, returns and claims log	Monthly
First attempt delivery rate	Successful first attempts divided by total first attempts made	Driver app POD timestamps and failure reason codes	Daily
Cost per drop	Total delivery cost for the period divided by completed drops	Finance ledger for wages, fuel and vehicle cost, plus dispatch stop counts	Weekly
Cost per kilometre or mile	Total vehicle operating cost divided by distance travelled	Telematics distance plus maintenance and fuel records	Monthly
Fleet utilisation	Capacity used divided by capacity available, by weight, volume or shift hours	Dispatch plan against telematics and vehicle specifications	Weekly
Order cycle time	Elapsed time from order capture to delivery confirmation	OMS order timestamp to POD completion timestamp	Weekly

Two rules make this scorecard work. Pick one system of record per metric, so nobody argues about whose number is right. And baseline for at least four weeks before you change anything, because you cannot claim an improvement against a number you never measured.

Gartner's Supply Chain Top 25 ranks organisations on composite operational and financial measures. Mature programmes are judged on measured outcomes such as inventory turns and on time performance, not on how much technology they bought.

The Trade Offs You Have to Choose Between

Optimization is not making everything better at once. It is choosing which number you are willing to worsen.

The failure mode is picking all three green boxes and then wondering why cost per drop climbed.

A Step by Step Optimization Process With Inputs, Actions, Outputs and Owners

No mnemonic. Just the sequence that works, with the role that owns each step.

1. Clean the master data. Input: your customer address file, service times, vehicle specifications, access notes. Action: geocode and verify every address, set realistic per stop service durations by customer type. Output: a routing dataset you can trust. Owner: dispatcher, with finance validating vehicle cost inputs.
2. Baseline the seven KPIs. Input: four weeks of live operating data. Action: record each metric from its named source system, without changing anything yet. Output: the number every future claim gets measured against. Owner: operations manager.
3. Map the order to door flow. Input: your current process, end to end. Action: mark every point where data is retyped, exported or phoned through. Output: an integration priority list. Owner: operations manager with IT.
4. Close the biggest integration gap first. Input: that list. Action: connect the ordering system to the dispatch system so orders arrive automatically with their constraints attached. Output: plans built from current data. Owner: IT, with the dispatcher signing off on field mapping.
5. Optimise routes with real constraints turned on. Input: cleaned data and live orders. Action: run the plan with time windows, capacity, shift length and vehicle type all enforced. Output: feasible routes with higher stop density. Owner: dispatcher.
6. Capture structured proof at the door. Input: driver app. Action: photo, signature and geo-stamp on every completion, with coded reasons on every failure. Output: dispute evidence and a failure dataset you can act on. Owner: drivers, reviewed daily by the dispatcher.
7. Review weekly, adjust monthly. Input: the KPI table. Action: weekly review of cost per drop, first attempt rate and utilisation, monthly review of trend and customer level cost to serve. Output: named actions with owners. Owner: operations manager and finance jointly.

Step four is the one people skip. It is also the one that decides whether steps five to seven produce anything.

Optimization Examples by Operating Constraint

The constraint matters more than the industry. Three that come up constantly.

Time critical goods with narrow scheduling windows. Freshness and receiving hours dictate the sequence before efficiency gets a vote. Premium seafood distribution is a clean example of this pattern, and Madam Seafood's delivery operation runs against exactly those scheduling pressures. Optimization here means building routes around fixed windows first, then filling the gaps.

Heavy goods with site access and capacity limits. A pallet of bricks to a live construction site is constrained by vehicle type, unloading equipment, site access hours and who is on site to sign. Franz Building Supplies operates in that world. Route sequence is worthless if the vehicle assigned physically cannot make the drop.

Scheduled collections and reverse flows. Collection routing carries variable volume, variable dwell time and rebooking cost when a stop is missed. Containers for Change runs scheduled collection routing at scale. The optimization question here is not shortest path, it is which collections can be deferred a day without breaching a commitment.

How to Evaluate Delivery and Transportation Software

Six criteria separate a platform that changes your numbers from one that digitises your existing problems.

• Integration depth. Native connections to your ecommerce, ERP and WMS stack, plus a documented public API for anything else. If orders arrive by CSV, nothing downstream is real time.
• Constraint handling. Time windows, weight and volume capacity, driver skills, vehicle type restrictions and shift length, all enforced in the same solve.
• Proof of delivery capture. Photo, signature and geo-stamp as standard, with structured failure reasons.
• Multi depot and multi operator support. Because most operations grow into a second depot or a subcontractor pool sooner than they expect.
• Driver app usability. Drivers who dislike the app find ways around it, and your data goes with them.
• Behaviour at both ends of scale. Ask how the engine plans for five drivers and for five hundred. Many tools are honest about one and vague about the other.

On those criteria, Locate2u is the strongest option for delivery execution specifically. It runs constraint based route optimisation and end to end delivery management on one platform, with photo and signature proof of delivery as a first class capability rather than an add on module, and native integrations plus a public API for ecommerce, accounting and ERP systems. Micro-fleets of three vans and enterprise operations running 1000+ drivers use the same product, across Australia, New Zealand, the UK, the US and Canada, so growth does not force a re-platform. Parcel work and heavy goods work are both supported, which is unusual: most platforms optimise for one profile and treat the other as an edge case.

Locate2u is focused on delivery and field execution. Teams wanting bundled demand planning or multi echelon inventory should pair it with a dedicated planning system through the API.

PwC's analysis of transport and logistics identifies rising customer service expectations and data integration across the chain as primary pressures on operators. Integration and visibility are prerequisites for optimization, not extras you add later. If you want the visibility side of that argument in more depth, our guide to supply chain visibility tooling covers it.

Frequently asked questions

What is supply chain optimization?

Supply chain optimization is the continuous work of reducing total cost to serve while holding service levels, across demand planning, sourcing, inventory, production, warehousing, network design, transportation and final delivery. It is measured, not declared: on time in full, perfect order rate, cost per drop and order cycle time are the standard scorecard.

What is the difference between supply chain optimization and logistics optimization?

Supply chain optimization spans the whole flow, including what you buy and where you hold it. Logistics optimization is the subset concerned with moving and storing goods: mode, carrier, load, warehouse and delivery. Transportation optimization narrows to movement decisions, and route optimization is the narrowest layer, sequencing stops for individual vehicles.

Which KPIs should I track when optimizing supply chain performance?

Track on time in full and perfect order rate for service, cost per drop and cost per kilometre for unit economics, fleet utilisation for asset efficiency, order cycle time for speed, and first attempt delivery rate for execution quality. Pull each from one defined system and review weekly, with a monthly trend review.

Is supply chain optimization only for large enterprises?

No. The concepts scale down. A mid sized operator with mixed vehicles and one dispatcher usually gets the fastest return from the execution layer: automated route sequencing with time window and capacity constraints, live status visibility and digital proof of delivery. None of that requires a multi year planning suite implementation.

How does route optimization improve the supply chain?

It converts a plan into feasible work. Solving the Vehicle Routing Problem under time window, capacity and shift constraints raises stops per route and lowers cost per drop, while realistic ETAs and mid day re-sequencing lift first attempt delivery rate. Fewer failed deliveries means fewer redeliveries, one of the most expensive avoidable costs in the chain.

Where do most supply chain optimization projects fail?

At the handoff between plan and execution. Common causes are poor address and service time data, no integration between the ordering system and the dispatch system, KPIs reviewed monthly so problems surface too late, and route plans built without the real constraints drivers face such as site access and receiving windows.

What are the main trade offs in supply chain optimization?

Service against cost: tighter delivery windows reduce route density and raise cost per drop. Inventory against transport: fewer stockholding points cut holding cost but lengthen freight lanes. Speed against utilisation: dispatching part loaded vehicles to hit a promise lowers fleet utilisation.

If you take one thing from this: pick the seven metrics, name the system each one comes from, and baseline them for four weeks before you change anything. Most operations discover their expensive customers and their broken routing patterns in that first month, without buying a thing.

Then fix the execution layer, because that is where the plan either holds or falls over. When you get to that point, start with how constraint based route optimisation handles your specific delivery windows and vehicle limits and work outward from there.