The Customer Ranking Was Upside Down
A global air express carrier had its customer profitability ranking backwards. This is the causal costing method that fixed its pricing.
By Pedro San Martín, Asher & Company
A global air express carrier was convinced that its banking customers lost money. A spreadsheet said so, and the commercial team had started to act on it with tighter discounts, less account investment and harder renewals.
The spreadsheet was wrong. Nobody had made an arithmetic mistake. The model was answering the wrong question. It asked how many shipments each customer sent. The question that mattered was what it cost, activity by activity, to serve each one.
The case comes from the Armstrong Laing Group (ALG), the firm behind the HyperABC and Metify ABM platforms, and dates from the late 1990s. It predates my own years at ALG, which began in 2004. The method behind it is the one I learned there, and I still use it in every profitability engagement at Asher & Company.
The method before the case
The team worked from the CAM-I Cross, the model the Consortium for Advanced Management International developed as the conceptual base of activity-based costing. It has two axes, and they run at the same time.
The vertical axis is the cost assignment view. Money flows from resources (people, technology, facilities) to activities (the work those resources do) and then to cost objects (the products, customers and channels that consume the work). This is the chain most people mean when they say ABC.
The horizontal axis is the process view. It runs from cost drivers (what causes the work) through activities to performance measures (how well the work is done). Most implementations skip it. It is the axis that explains why something costs what it costs, and what that says about the process.
Companies that build only the vertical axis get a better allocation. With both axes they get a decision tool. That is why this case ended up changing commercial strategy instead of producing one more cost report.
What the old model got wrong
The original model had 30 cost pools and five allocation drivers. The main driver was shipment count.
It looked reasonable. In air express, though, shipment count moves with cost without causing it. Cost comes from weight, distance, customs complexity, pickup frequency, special handling and how often the customer calls the service center. One 50-kilogram industrial part that needs a customs declaration and special handling generates about ten times the activity cost of ten 100-gram document envelopes collected on a scheduled daily route.
The model treated both as shipments. So it costed them the same.
The result was a systematic inversion. Banks shipped light documents on predictable daily pickups. They rarely needed customs and almost never called service. Because they shipped often, the model made them look expensive. Heavy manufacturers shipped less often but needed complex customs work, irregular pickups and frequent service contacts. The model made them look cheap.
Management was reading a profitability ranking that mirrored reality. Every pricing decision and every renewal built on it pointed the wrong way.
That is what a failed cost model looks like in practice. No single number raises an alarm. The outputs are plausible, margins look normal, and the business keeps running while value leaks out every month.
How the activity model was built
The new model had 106 activities in 15 categories, from pickup and delivery through customs, international transport, customer service, sales, IT, HR and finance. Each activity had an operational definition, a primary driver and a way to collect the data.
The number of activities mattered less than the drivers chosen for them.
- Pickup and delivery. The driver became courier minutes per pickup. A scheduled daily pickup at a bank headquarters took about 8 minutes. An on-demand industrial pickup with a route change and a weight check took 35.
- Customs. The driver became a complexity index. A document envelope needed no customs work. An industrial component needed a full declaration, a classification lookup and sometimes an inspection, at roughly 40 times the cost.
- Customer service. The driver became inbound contacts per account. Banks averaged 0.3 contacts per shipment and manufacturers 2.8 (tracking requests, proof-of-delivery queries, exceptions, damage claims). That is a ninefold gap.
Once those costs were traced to customer segments, the ranking flipped.
| Segment | Rank on GL allocation | Rank on activity costs | What explains the gap |
|---|---|---|---|
| Banks and financial services | 6th, least profitable | 2nd | Light documents, predictable pickups, almost no customs |
| Technology | 3rd | 1st, most profitable | High value, time-sensitive, document-heavy |
| Healthcare | 5th | 4th | Regulatory complexity offset by premium pricing |
| Retail | 4th | 3rd | Moderate complexity, volume-driven |
| Heavy manufacturing | 1st, most profitable | 5th | Complex customs, special handling, high contact rate |
| Logistics intermediaries | 2nd | 6th, least profitable | High volume, extreme price sensitivity, high coordination cost |
Scaling it without losing rigor
Proving activity costing in one country is a pilot. Running it across a global network is an organizational change, and that gap is where most ABM programs die.
The team treated scale as a design problem. The software already worked. The open question was how to make each country's implementation fast, repeatable and locally owned without weakening the analysis. Their answer was to standardize the framework and localize the data.
Every country got the same chart of 106 activities, the same 15 categories, the same driver definitions and the same calculation logic, adapted to its own operations. A five-day training program came with it, along with an interface that fed local data straight into the calculation engine. The monthly cost update ran on its own.
Implementation time fell sharply.
| Stage | Duration |
|---|---|
| First pilot | 4 months |
| Second pilot | 4 months |
| Third pilot | 5 weeks |
| Rollout, larger countries | 3 to 4 weeks each |
| Rollout, smaller markets | 1 to 2 weeks each |
The software did not change between the first pilot and the third. The method did.
Not every country got the full model. Twenty-five ran all 106 activities. Fifty more ran a simplified 50-activity version sized to their market, and the smallest used a spreadsheet. That was deliberate. The depth of the model should match the complexity of the decisions it supports. We apply the same rule when we design Decision-to-Value™ architectures today.
What it paid for
The commercial output was a pricing evaluation tool. It used activity costs to set a price for any customer based on that customer's shipment profile. For the first time, a sales manager could walk into a negotiation with a number tied to the real cost of serving that account.
The tool evaluated more than USD 500 million of revenue a year. Each percentage point of margin improvement was worth about USD 5 million in profit. The whole ABM program cost USD 6 to 7 million.
The return is the headline. The reason behind it matters more. The old model used one driver to allocate costs that several drivers caused. It produced precise-looking numbers that were structurally wrong, and management trusted them because they came from the general ledger.
I still find that pattern in most of the organizations I work with. Tools have improved and data is far more plentiful. Yet costs are still allocated with proxies instead of traced with causal drivers.
Five lessons that still apply
1. A model can be precise and still be wrong. The spreadsheet gave every customer a number, and the error was structural enough to invert the whole customer base. Precision in the wrong model does more damage than honest uncertainty in a simple one.
2. Use both axes of the CAM-I Cross. The vertical axis gives you a better allocation. The horizontal axis, from drivers to activities to performance measures, links cost to how operations are run. That link turns a cost model into a decision tool.
3. The driver is the model. Moving from shipment count to weight and complexity inverted the ranking. Software, activities and cost pools barely mattered next to that choice. Driver selection deserves more senior attention than system selection.
4. Design the method for repetition. Going from four months to five weeks came from a standard activity chart, structured training and automated data feeds. None of it depended on the platform.
5. Size the model to the decision. Use 106 activities in large markets, 50 in mid-size ones and a spreadsheet in the smallest. Aim for the best decision quality per dollar of analytical effort.
The same question, new tools
ALG was acquired in 2006 and the software moved on. The question stayed. Every engagement I run at Asher & Company starts where this one did. Which customers, products, channels and operations create value, and which ones consume it without anyone noticing?
Oracle EPCM, connected to modern planning tools and AI-assisted analysis, can now deliver in weeks what took that team years. The answer is still only as good as the drivers behind it. Organizations that price, invest and prune portfolios on GL-allocated costs are repeating the shipment-count mistake, with more at stake.
We call the alternative Decision-to-Value™, a causal and traceable link between the data operations generate and the decisions that move enterprise value. It starts with one question. What is driving this cost?
Pedro San Martín is Founder & Principal of Asher & Company, a strategic finance and profitability analytics firm specializing in Enterprise Performance Management, profitability architecture and cost management.
asher.company | Insights
