Module 02 · Cost-to-Serve & KPI
Which driver should we improve first, and which customers are losing us money?
Total cost to serve is broken into a tree: transport, warehouse, inventory and accessorial, each decomposing further into the operational drivers underneath. Move any leaf and the whole structure recomputes. The tornado then ranks the drivers by what a one percent improvement is actually worth — which is the ranking that should set an improvement agenda, and rarely matches intuition. Cost is allocated to accounts by causal driver rather than as a flat percentage of revenue, which is what makes loss-making customers visible instead of hidden.
Who uses it
Logistics performance, commercial finance, continuous improvement
When
Continuously — this is the standing scorecard, not a project tool
Time to first answer
Immediate — everything recomputes on every slider move
Quick start
From open page to an improvement agenda
Look at the tree first
The main panel shows the full decomposition with bars sized by share. Before touching anything, note which of the four branches dominates. In most freight-heavy networks transport is 70–75%, but the remaining quarter is where the uncontested savings usually sit.
Set the scale to your business
Tab 1 → annual orders and average order weight. Then the transport drivers — length of haul, rate, consolidation ratio, out-of-route. Take consolidation ratio and fill from the Consolidation Optimizer rather than guessing.
Go to tab 2 and read the tornado
Drivers are ranked by how much a ±10% move shifts total cost. The controllable only filter is on by default, which removes demand volume and fuel — they move the number but nobody in the building can act on them.
Check the owner column
Each driver carries the function that owns it. A flag underneath names whichever function owns most of the top five. An improvement programme that excludes that function will underdeliver regardless of effort.
Go to tab 3 for the customer view
The scatter plots revenue against cost to serve as a percentage of revenue. Red points are loss-making. The table underneath ranks them worst-first.
Screen map
What each tab is for
Tab 1 sets the model, tab 2 sets the agenda, tab 3 finds the money, tab 4 prices the operational discipline.
| Tab | What it controls | Leave it alone unless |
|---|---|---|
| 1. Drivers | Scale, transport drivers, warehouse and inventory, accessorial rates. Everything the tree is built from. | Never — start here and put in real values. |
| 2. Sensitivity | Perturbation size and the controllable-only filter. The ranked improvement priority table. | Never. This is the output that changes behaviour. |
| 3. Customers | Account count, selling price, gross margin, order-size dispersion. Ranking and profitability quadrant. | You do not have account-level data — then treat it as illustrative. |
| 4. Leakage | Accessorial breakdown by root cause, and the KPI chain reaction from a missed cut-off. | Never — this is the most immediately actionable page. |
Reading the output
What each output is telling you
The absolute dollars depend entirely on the scale you entered. The rankings do not, and the rankings are the point.
| Metric | What it means | What to do about it |
|---|---|---|
| Elasticity | Percentage change in total cost per one percent change in the driver. | Anything above 0.3 is a serious lever. Below 0.05, improving it is a rounding error however hard it is. |
| Impact span | Dollar range between the driver moved down and moved up. | Rank by this, then filter by whether your organisation can actually move it. |
| Loss-making accounts | Accounts where cost to serve exceeds gross profit. | Almost always small-drop, high-frequency, long-haul. Levers are minimum order quantity, delivery-day consolidation, or a price conversation. |
| Profit concentration | How many accounts produce 80% of net profit. | If it is a small number, service failures inside that group cost far more than the average OTIF figure suggests. |
| Leakage per order | Accessorial cost divided by order count. | This number never appears in a rate negotiation and accumulates every single day. It is the easiest number to put in front of an operations meeting. |
Use cases
Three questions this actually settles
Use case 1
Where do we start? We have limited improvement capacity.
| Set up | Load real drivers on tab 1. Tab 2 → keep controllable only ticked, set perturbation to 10%. |
|---|---|
| Watch | The top three levers and their owners. In a typical freight-heavy network these come out as length of haul, consolidation ratio and linehaul rate — the first two owned by planning, the third by procurement. |
| Decide | Fund the top three. Everything below rank five is inside model error; assigning analysts there is misallocation. If procurement owns most of the top five, a planning-only programme will not deliver. |
Use case 2
Which customers are we losing money on, and what do we do about it?
| Set up | Tab 3 → set account count, selling price and gross margin to your business. Sort by worst net margin. |
|---|---|
| Watch | The quadrant separates high-revenue accounts with acceptable cost from small accounts whose cost to serve exceeds their gross profit. Revenue size alone does not predict which is which. |
| Decide | For each loss-maker choose one of three: raise minimum order quantity, consolidate them onto fixed delivery days, or reprice. The cost driver shown for that account tells you which of the three will work. |
Use case 3
What is a missed dock cut-off actually worth?
| Set up | Tab 4 → set the cut-off miss rate, the share that gets escalated to expedite, and the spot premium those shipments carry. |
|---|---|
| Watch | The chain runs missed cut-off → expedited orders → spot premium paid → OTIF impact on whatever was not escalated. The final line prices a single point of improvement. |
| Decide | Take that per-point figure to the operations meeting. It converts a dock discipline conversation into a budget conversation, which is the only version of it that gets funded. |
Pitfalls
Where people misread this tool
Every one of these has actually happened during testing.
The customer book is synthetic
Accounts are generated in the browser from a fixed seed. Their shape is realistic — small accounts carry disproportionate cost — but they are not your customers. Use this page to understand the mechanism, then repeat the allocation on real data before naming anyone.
Ranking is robust, absolute dollars are not
Every figure scales with the volume and price you entered. If the scale is wrong, every dollar amount is wrong by the same factor — but the ordering of the levers survives, and the ordering is what the tool is for.
Turning off the controllable filter buries the useful answer
Order count and average order weight will dominate the tornado, because total cost scales with volume. That is arithmetically true and operationally useless. During testing they occupied the top two positions and pushed every actionable lever down the list.
Allocation basis is a choice, and it changes who looks unprofitable
Cost here follows drop size, haul length, consolidation feasibility and accessorial incidence. Allocating as a flat percentage of revenue — the common shortcut — makes every account look equally profitable and hides exactly the ones you are looking for.
Data in and out
Files this tool reads and writes
| Direction | File | Purpose |
|---|---|---|
| In | — | No file input. Drivers are entered directly; the account book is generated from a seed. |
| Out | cost_to_serve_analysis.csv | Component decomposition, ranked sensitivity with elasticity and owner, full account-level cost to serve, and the complete assumption set. |
Where it sits in the loop
This is the performance stage and it is the hub of the whole set. Transport drivers come from the Consolidation Optimizer, inventory and facility cost from the TCO simulator, accessorial rates from the Budget simulator. It is where all four other tools resolve into a single cost equation — and where an improvement agenda gets its ranking.