Same Data, Two Models: How the Profitability Ranking Changes When ABC Gives Way to TDABC

In April 2026 we built a costing model for a post-graduate class in Lisbon with one unusual property: it was designed to be loaded twice. The same cost pools, the same ledger accounts, the same 4,322 customers and 387,339 invoice lines from the public UCI Online Retail dataset. Version A allocates fifteen activities with the simple drivers of classic activity-based costing. Version B allocates the same fifteen activities with time equations. Everything upstream is identical, so any customer that changes place in the profitability ranking between the two runs changed place for exactly one reason.

One dataset, one ledger, two fourth allocations

The business is a UK online retailer selling to wholesale and retail customers in 35 countries: £11.2 million of revenue over 21,194 invoices, 3,676 products, an average of 18.3 lines per invoice. The organisation chart has 65 people in ten departments, from warehouse receiving to finance, with practical capacity set at 1,760 hours per person per year, 80 percent of the theoretical 2,200.

Both versions of the model share the first three allocation stages: ledger accounts into cost pools, cost pools into departments, departments into activities. They diverge only at the fourth stage, where activity cost reaches customers, products and countries. In version A each activity has one driver. Goods receiving is spread by quantity received, order packing evenly per invoice, account management evenly per customer. In version B each activity has a time equation: a base time per unit of work, plus conditional terms that add minutes when the order has a characteristic that costs minutes.

The fifteen equations

This is the table the class worked from. The ABC column is what most first models do. The TDABC column is what the warehouse and the customer-service desk would tell you if you asked them how long things take.

Read down the third column and the pattern is visible before any number is computed. “International” appears in eight of the fifteen equations, and a ninth charges extra for markets outside the EU. “Wholesale” appears in three. Returns, which cost nothing at all under the ABC driver for a customer with clean orders, cost five minutes a line the moment they exist. None of this is exotic. It is the difference between asking how many and asking how long.

What a minute costs in each department

A time equation is only half of a TDABC allocation; the other half is the capacity cost rate, the cost of one minute of a department’s practical capacity. At the design stage the rates were estimated from headcount and departmental cost, before the software recomputed them from the actual allocated ledger: about £0.26 per minute for most of the operational departments, £0.29 in picking and packing, £0.38 in finance and administration, £0.49 in IT and £0.19 in HR. The spread matters. Eight minutes of customs documentation is not the same cost as eight minutes of goods receiving, and a model with a single blended rate would miss that as well.

Picking and packing is the largest department, 18 of the 65 people and 31,680 practical hours a year, and it handles all 387,339 lines. That is where the multi-line term does its work: an order with many different products is 40 to 60 percent more expensive to pick than a single-product order of the same quantity, and under a quantity driver the two look identical.

Who moves, and in which direction

Three groups of customers fall in the ranking. Small international orders, because customs, documentation, packing labels, support and finance handling add between 17 and 23 minutes of work to an order that may be tiny; in the class model international orders cost three to four times as much to process as domestic ones. Customers who return goods, because a return line costs five minutes plus three more if it crosses a border, against zero for a clean order. And customers whose orders are long lists of different products, because the picking term charges 0.3 minutes for every additional product and the packing term half a minute per line: invoices with more than 50 lines consumed five times the processing time of simple orders.

Three groups rise. Large domestic wholesale orders, whose fixed per-order minutes are spread over a big basket. Customers with low return rates, who under the ABC driver had been paying for other people’s returns. And high-volume single-product lines, which are exactly what a warehouse is efficient at. Under the volume driver these customers subsidised the first three groups; under time equations the subsidy becomes visible and reverses.

The size of the movement is the point. Individual customer margins moved by up to six percentage points between the traditional allocation and the time-driven one. The invoice-processing cost, a flat £1.66 per invoice when spread evenly, ran from £0.80 to £6.70 once line count and complexity were allowed to matter. Wholesale accounts, which show the higher gross margin, gave one to two points of it back to the processing and shipping work that big orders generate. And about 800 customers, 18.5 percent of the base, were loss-making after cost-to-serve, most of them small international accounts where even the marketing spend allocated to their country exceeded their gross margin. Under the volume allocation most of those 800 looked marginally profitable.

ABC asks how many. TDABC asks how long, and for whom. The ranking changes because the second question has a different answer for every customer.

What this does not show, and what to do with it

A teaching model is not a client engagement. The equations were written from experience of how online-retail operations work, not from timing this retailer’s warehouse, and the rates were estimates until the software recomputed them. What the exercise proves is structural: given the same data and the same first three allocation stages, the choice of driver at the last stage alone is enough to move a customer from the profitable half of the ranking into the loss-making tail. That is not a modelling curiosity. It is the difference between a discount given to a customer who earned it and one given to a customer who was already costing more than they paid.

If your model allocates by revenue, headcount or an even split per invoice, the fifteen equations above are a usable starting point. Most of them transfer to any business that receives, stores, picks, ships and invoices, and the conditional terms, international, wholesale, returns and line count, are the ones worth measuring first.

Related reading: TDABC versus ABC, how time equations work, the cost driver library and the whale curve.