Time-driven activity-based costing, worked out in full

Two numbers, and everything follows

Every costing method has to answer one question: this pile of cost paid for work, so which work, and for whom. Traditional absorption costing answers it with a single rate over a single base, usually revenue, units or labour hours. Activity-based costing answered it by breaking the business into activities and asking people what share of their time each activity took. The answers were better. The machinery was so heavy that most models were abandoned within a few years.

TDABC answers the same question with two parameters per group of resources that does broadly similar work.

The first replaces the survey. You never ask anyone what percentage of their year went to a task. You cost the group once and divide by the time it can genuinely deliver.

The second replaces the activity list. Instead of a separate activity, cost pool and survey for every variant of a job, one equation absorbs the variants as terms. Picking is one equation whether the order has two lines or two hundred, needs refrigeration or does not, leaves the country or stays at home.

And because the rate is anchored to what the group could do rather than to what it did, the difference between the two does not disappear. It becomes a number with a name: the cost of unused capacity. Section 08 is about where that number goes, and it is the reason the method exists.

Where the method actually came from

TDABC is usually dated to 2004, and the date is right, but the idea it rests on is older and the credit is worth splitting properly.

1992, the concept. Robin Cooper and Robert Kaplan published "Activity-Based Systems: Measuring the Costs of Resource Usage" in Accounting Horizons. It set out the identity that TDABC later operationalised: resources supplied equals resources used plus unused capacity. That sentence is the whole argument for practical capacity, and it predates the method named after it by twelve years.

1996 to 1998, the practice. Steven Anderson founded Acorn Systems and built activity models around measured time rather than surveyed percentages. In parallel, Kaplan and Cooper's Cost & Effect (1998) gave capacity costing a book-length treatment.

November 2004, the naming. Kaplan and Anderson published "Time-Driven Activity-Based Costing" in the Harvard Business Review, volume 82, number 11, pages 131 to 138. The same argument had circulated as Harvard Business School working paper 04-045, which is free to read and is the version most people actually cite. The article is a correction rather than a new theory of cost. It keeps the ambition of activity-based costing and throws away the survey.

2007, the manual. The book Time-Driven Activity-Based Costing: A Simpler and More Powerful Path to Higher Profits (Harvard Business School Press) is where the implementation detail lives.

What TDABC inherits from ABC is the whole diagnosis: averages hide the differences that matter, and cost follows the work rather than the invoice. What it changes is the measurement instrument. ABC asked people to divide their time. TDABC estimates how long a task takes and lets the transaction data do the counting. The comparison, feature by feature, is in TDABC vs ABC, and the encyclopedia entry for the method sits in our methods library.

TDABC did not invent unused capacity. It made it impossible to hide.

The cost of capacity supplied, line by line

The numerator is not "the department's cost in the ledger". It is everything you must pay to have that group standing ready and able to do the work the equation prices, for the period the rate covers. That definition does real work, because it decides what goes in and what stays out.

Here is the group we will use for the rest of this page: the pick-and-pack team of a mid-size distributor. Eight pickers, one shift supervisor, one pick face. All figures are illustrative and chosen so that the arithmetic is checkable.

Four things that do not belong in it

Anything already traced to the object. The goods being picked, the freight bought per shipment, a subcontractor invoiced per job. These reach the product or the customer directly and do not need to travel through minutes. Putting them in the numerator charges them twice: once as themselves and once as time.

The cost of a different resource group. The most common version is a quiet transfer from a group with poor utilisation into one with good utilisation, because the second one "can carry it". It cannot. Each group gets its own rate and its own unused-capacity line, and the comparison between them is diagnostic information you have just destroyed.

Corporate cost with no causal link to this capacity. Group finance, the executive team, the brand budget. There are respectable reasons to allocate those somewhere, but a capacity cost rate is not the place, and here is what happens if you do it anyway. Push a 112,200 corporate recharge into this pool and the numerator becomes 561,000. Divide by the same 748,000 practical minutes and the rate goes from EUR 0.60 to EUR 0.75 a minute. Every order in the business is now 25% more expensive to pick and not one of them got any slower. The model has stopped measuring the operation and started measuring head office.

One-off items. A restructuring charge, a legal settlement, an insurance recovery. A capacity cost rate is a run rate. If a one-off sits in it, next year's rate moves for a reason nobody can explain, and the first person who spots it stops trusting the model.

One deliberate subtlety in the table above. The supervisor's cost is in the numerator and the supervisor's minutes are not in the denominator. Supervision is part of what it costs to make picking capacity available; it is not itself picking capacity. Get this backwards and you inflate the denominator with minutes that no time equation will ever consume, which understates the rate and manufactures unused capacity that was never there.

Practical capacity, derived from a calendar

Almost every account of this method, ours included until now, says practical capacity is "about 80 to 85 per cent of theoretical" and moves on. That is a sanity check, not a method. The number should come out of the calendar, and then you compare it with the rule of thumb to see whether you have missed something.

So here is one picker's year, built downwards.

Why practical and not theoretical

Theoretical capacity is the time you could work if nobody ever ate, trained, handed over a shift or went home ill. Nobody supplies it, so nothing should be priced against it. Divide by theoretical capacity and the rate comes out too low, every transaction looks cheaper than it is, and the shortfall reappears later as an unexplained gap between the model and the ledger.

The opposite error is more tempting and more damaging: dividing by the minutes actually worked. Do that and the model balances perfectly every month, because by construction there is nothing left over. It also tells you nothing. Unused capacity has been distributed across the transactions that happened to occur, which means a quiet quarter automatically makes every customer look expensive. That is absorption costing with a stopwatch, and it is the single most common way a TDABC model is quietly turned back into the thing it replaced.

Practical capacity sits between the two, and it is the only one of the three that keeps the idle time both priced and visible. Kaplan and Anderson suggest roughly 80% of theoretical for people and around 85% for machines as a starting point. Treat those as a check on your own arithmetic. Our figure above landed at 85.1% because of a specific holiday entitlement and a specific shift pattern; a two-shift operation with a longer handover would land lower, and a plant with a summer shutdown lower again. There is more on the mechanics in capacity costing.

One division, and what it is sensitive to

That is the whole of parameter one. The interesting question is how much the answer moves when the denominator moves, because the denominator is the part people argue about in meetings.

Three things worth reading off that table. First, the choice matters: from theoretical to actual-use the same order swings by a third. Second, the argument between 85% and 90% is worth 5.8% on every price in the business, which is enough to change a tender and not enough to justify a three-week debate. Third, and this is the one to remember, only the middle rows leave anything on the table. The bottom row makes the unused-capacity line vanish, and the person proposing it is usually the person who does not want to see it.

A practical rule we use: fix the practical-capacity derivation once per resource group, write down the deductions by name, and change it only when the calendar or the shift pattern changes. Never when the answer is uncomfortable.

Built in three passes, term by term

A time equation estimates how many minutes one execution of a process consumes, as a base time plus a term for each characteristic of the transaction that changes it. Here is the one for our pick face, written the way it is actually written: in passes, each one added because the previous version was visibly wrong about something.

Pass one. The version everybody writes first.

Four minutes of base time, and 1.2 minutes for each order line. This is already better than an average per order, and for a warehouse that ships one product it might be enough. Ours does not.

Pass two. Watch the team for a morning and the missing driver is obvious: cartons. A single line of forty units fills three cartons and takes three lots of building, labelling and sealing. Forty single-unit lines can go in one carton. Lines and cartons are different work and they move independently.

Pass three. Two conditions remain that do not scale with anything. A refrigerated line means a detour to the chill room and a cold-chain check, once, however many lines are involved. An export order means a packing list, pallet labelling and customs paperwork handed to the driver, once. Conditions enter as flags worth zero or one.

What the equation does to two orders

Order A, four lines, two cartons, ambient, domestic:
T = 4.0 + 1.2 × 4 + 0.9 × 2 = 4.0 + 4.8 + 1.8 = 10.6 minutes, costing 10.6 × 0.60 = EUR 6.36.

Order B, twenty-two lines, nine cartons, refrigerated, export:
T = 4.0 + 1.2 × 22 + 0.9 × 9 + 7 + 12 = 4.0 + 26.4 + 8.1 + 19 = 57.5 minutes, costing 57.5 × 0.60 = EUR 34.50.

The same activity, on the same shift, by the same people. Five and a half times the work. A classic ABC model with one "picking" pool and a surveyed percentage would have given these two orders the same cost, and so would any rate based on revenue, units or weight.

How to extend it, and when to stop

Extension is the easy part and the dangerous part. A fourth quantity term is available immediately: units picked beyond the first on a line, at perhaps 0.1 minutes each, which matters if you ship both singles and full cases from the same face. A term can also be conditional on another term, so that E fires only for shipments leaving the customs union rather than for every foreign address. Both are legitimate.

The test for adding one is the same test as for the first three. Does it move the minutes materially, so that doubling it visibly changes how long the job takes. Does it already exist as a populated field for every transaction, not just for the interesting ones. Would the supervisor look at the term and say yes, that is what makes it longer. A term that fails any of the three is decoration, and decoration is what a controller stops maintaining in month nine.

Two things that look like terms and are not. A learning-curve adjustment for new staff is a variance, not a driver: it belongs in the analysis of why actual minutes differed from modelled minutes, not in the equation. And a coefficient that changes by customer is not a coefficient, it is a separate process that you have not admitted to yet.

On the count: one equation per process, not per variation. A focused cost-to-serve model runs on a handful of them. A full mid-market profitability model typically lands between 10 and 40. The full mechanics of writing them are in the time equations reference.

Run the equation over a year of orders

Four customers, one year, the driver volumes the equation runs on. Every figure in the two right-hand columns comes from the five columns to their left and the single rate of EUR 0.60, so any line can be reproduced with a calculator.

Read the per-order costs across: EUR 23.10 for Retail chain A, EUR 12.54 for Foodservice B, EUR 38.76 for Export wholesaler C, EUR 4.38 for Online pure-play D. Four customers of one business, in one warehouse, differing by a factor of nearly nine on what a single order costs to get out of the door. None of that is visible in a gross margin, and none of it is visible in any rate applied to revenue.

Note also what the model has not done. It has assigned EUR 395,400. The pool was EUR 448,800.

The missing EUR 53,400, and where it goes

This is the part of the method everything else exists to serve, so it is worth being slow about.

The team supplied 748,000 usable minutes and the four customers consumed 659,000 of them. The other 89,000 minutes were paid for and not sold. Valued at the same rate that priced the work, they cost EUR 53,400.

The critical word in the last row is separate. In traditional absorption costing that EUR 53,400 has nowhere else to go, so it is spread over the customers, and every one of them is charged for time it did not consume. In classic ABC the same thing happens more subtly: rates built on actual activity volumes silently carry the idle time inside them. TDABC is the method that refuses to do it, and refusing is the entire contribution.

Make it concrete. One picker supplies 93,500 practical minutes a year. So 89,000 unused minutes is 0.95 of a picker. The finance report does not say "unfavourable overhead absorption variance", it says: we are carrying almost exactly one picker we have not sold. That sentence starts a conversation. The variance never did.

What the number does not mean

It is not automatically a saving. Capacity is bought in lumps. You cannot dismiss 0.95 of a picker, and if December runs at 130% of the average you need the headroom in November. An annual rate hides seasonality by design, so read the unused-capacity line monthly before you act on it annually.

It is not a failure. Deliberate headroom for growth, for a tender you are bidding, or for resilience is a decision, and this line is what that decision costs. Naming the price is not the same as objecting to it.

It is not a customer's problem. The moment somebody asks for "fully absorbed" customer numbers for a pricing meeting, the separate line starts dissolving. This is the most common way a working TDABC model reverts, and it usually takes about two quarters.

It is not evidence on its own. Unused capacity can mean overstaffing, or a demand collapse, or a practical-capacity assumption that is too generous, or a time equation whose coefficients are too low. Check the last two before acting on the first.

The idea has an older name and an older source. Cooper and Kaplan set out the identity in Accounting Horizons in 1992: resources supplied equals resources used plus unused capacity. TDABC's contribution was to make it fall out of the arithmetic every month instead of being an annual exercise.

What the same year looks like under absorption costing

Run the conventional method over the same facts. There is one pool of EUR 448,800 and one base. Revenue is the base most businesses actually use, and the four customers billed EUR 8,000,000 between them, so the rate is 448,800 / 8,000,000 = 5.61% of revenue.

Look at the total of the difference column. It is EUR 53,400, and that is not a coincidence: it is the unused capacity, to the euro. Absorption costing has to put the idle time somewhere, and the somewhere is always the customers. So the two methods do not differ by a rounding tolerance. They differ by exactly the amount of capacity nobody bought, redistributed in proportion to revenue, which is to say in proportion to the one thing that has no causal relationship with picking time at all.

Follow it through to a decision. Under absorption, Retail chain A carries EUR 179,520 of warehouse cost against EUR 115,500 of work it actually caused, so it looks 55% worse than it is, and the account manager defending it in a margin review will lose. Foodservice B, with ten thousand small refrigerated orders, looks 15% better than it is, and nobody will touch it. The two conclusions a business would draw from that table are both wrong, and both are wrong in the direction that costs money.

This is the mechanism, at portfolio scale, behind what we found at a New Zealand distributor: 830 of 1,951 customers contributing negatively once cost to serve was properly attributed, EUR 1.335M of negative contribution, later brought down to around EUR 665K with 295 accounts still below the line. The full account is in the case study, and the way the ranking is read is in the whale curve.

When TDABC fits, and when it does not

The method has a shape, and forcing it onto work that has a different shape is how a good idea acquires a bad reputation. This is the test we run before agreeing to build one.

One row deserves expanding, because it is the fair criticism the literature actually makes. TDABC assumes time is a good proxy for resource consumption. For transactional and operational work it is an excellent one. For knowledge work it can be poor, and Cardinaels and Labro showed in The Accounting Review in 2008 that people asked to estimate task durations are systematically biased, particularly when the estimate is made in aggregate rather than task by task. That is not an argument against the method. It is an argument for estimating from timestamps and observation wherever they exist, and for treating interview-derived coefficients as the weakest ones in the model and the first to validate.

The one condition we treat as disqualifying is the last one. A model with no owner is a report, and it should be priced, scoped and described as a report.

What breaks, after the model works

Very few TDABC models fail at go-live. They fail quietly, several months later, and the failures are specific enough to list. These are the ones that belong to this method rather than to cost models in general.

1. The rate goes stale. Pay rises, a lease renews, two pickers leave and are not replaced. The numerator and the denominator both moved and nobody re-divided. Fix: refresh both on the same calendar as the budget, once a year, plus immediately on any structural change to the group.
2. Practical capacity gets re-argued. The month the unused-capacity line is uncomfortable, someone proposes that 85% was always conservative and 92% is more realistic. The rate falls, the idle line shrinks, and the model has been quietly edited to produce a preferred answer. Fix: the derivation from section 04, written down per group, changed only when the calendar changes.
3. The unused-capacity line is reallocated. Usually with good intentions, for a pricing meeting that wants fully absorbed customer numbers. Once it happens it does not get undone, and within two quarters the model is absorption costing with extra steps. Fix: if a fully absorbed view is genuinely needed, produce it as a clearly labelled second view. Never by changing the model.
4. Minute-hunting. Terms multiply, a fourteenth coefficient appears, precision becomes decorative and maintenance becomes a job nobody wants. Fix: the three tests, applied on the way in. Every term is something a controller has to defend for years.
5. The reconciliation is abandoned. Minutes times rate stops tying back to the ledger for the resources in scope, and because nothing visibly breaks, it goes unnoticed for two quarters. Fix: one line in the monthly pack: modelled cost against ledger cost for the groups in scope, with the difference shown even when it is small.
6. Nobody owns it. The most common cause of death, and the least method-specific. Fix: a named owner in finance before go-live, with the refresh in their objectives rather than in their goodwill.

The broader anatomy of that decay, with the symptom you see first for each failure mode, is in why costing models fail after go-live. It is the page to read before the build starts, not after.

The consolation is that TDABC recovers more cheaply than what it replaced. Updating a classic ABC model meant re-running a survey programme. Updating this one means editing two numbers in a divide and, occasionally, one coefficient. That is a morning, not a project, and it is the main reason these models survive their third year.

The rest of the method, in order

This page is the reference. The working pages sit under it, and this is the sequence we would read them in.

1. The data extract. Twelve files, the fields in each, the grain, and the fallback when a field does not exist. Read it before deciding whether to start.
2. The cost driver library. Which characteristic of a transaction earns a term, activity by activity, across six areas of a business.
3. Time equations in full. The mechanics of writing them, how many a model needs, and where the coefficients come from.
4. Capacity costing. Practical capacity, unused capacity and the denominator, in more depth than section 04 goes.
5. Build a TDABC model. Seven steps across three resource groups, ten customers, and a whale curve that closes to the cost base.
6. TDABC vs ABC. The comparison, feature by feature, including what classic ABC still does better.
7. Cost to serve and the whale curve. What the model is for, once it runs.

What we read, and what it actually says

Every claim about the method on this page traces to one of these. Where a source is paywalled we say so, and where the free version differs we say that too.

• Cooper, R. and Kaplan, R. S. (1992). "Activity-Based Systems: Measuring the Costs of Resource Usage." Accounting Horizons, 6(3), 1-13.The origin of the identity in section 08: resources supplied equals resources used plus unused capacity. Twelve years before TDABC was named.
• Kaplan, R. S. and Cooper, R. (1998). Cost & Effect: Using Integrated Cost Systems to Drive Profitability and Performance. Harvard Business School Press.Capacity costing at book length, and the ground TDABC grew from.
• Kaplan, R. S. and Anderson, S. R. (2004). "Time-Driven Activity-Based Costing." Harvard Business Review, 82(11), 131-138.The founding article. Paywalled. Substantially the same argument circulated as Harvard Business School working paper 04-045, which is free, and that is the version to read if you do not have HBR access.
• Kaplan, R. S. and Anderson, S. R. (2007). Time-Driven Activity-Based Costing: A Simpler and More Powerful Path to Higher Profits. Harvard Business School Press.The implementation manual. This is where the detail on time equations and multi-group models lives.
• Everaert, P., Bruggeman, W., Sarens, G., Anderson, S. R. and Levant, Y. (2008). "Cost modeling in logistics using time-driven ABC: experiences from a wholesaler." International Journal of Physical Distribution & Logistics Management, 38(3).Real time equations from a real wholesaler. The most useful published evidence that the structure survives contact with an operating business.
• Cardinaels, E. and Labro, E. (2008). "On the Determinants of Measurement Error in Time-Driven Costing." The Accounting Review, 83(3), 735-756.The serious critique, and the reason section 10 says what it says about estimated times. People asked to estimate durations are systematically biased.
• Kaplan, R. S. and Porter, M. E. (2011). "How to Solve the Cost Crisis in Health Care." Harvard Business Review, 89(9).Where the method went after distribution, and why most recent TDABC literature is clinical.
• Kaplan, R. S. and Narayanan, V. G. (2001). "Measuring and Managing Customer Profitability." Journal of Cost Management, 15(5), 5-15.The customer-profitability distribution behind the whale curve.
• Félix, Â., Cabrita, A., Garrido, D., Guimarães, M. and Pedroso, R. (2017). Dialysis unit costing study. Revista O Hospital no. 13, Associação Portuguesa para o Desenvolvimento Hospitalar, page 40.Our own published TDABC work, with our names on it. O Hospital is the professional magazine of the APDH rather than an academic journal, so we call it what it is. The figures are in the case study.

Two things we deliberately do not cite. Adoption statistics for activity-based costing that circulate without a traceable survey behind them, and any percentage attributed to "studies show". If we cannot name the study, the number does not go on the page.

Fair questions.

What is time-driven activity-based costing?

Time-driven activity-based costing, usually shortened to TDABC, is a costing method that assigns the cost of resources to products, services and customers using two parameters per resource group. The first is the capacity cost rate: the cost of supplying the group divided by its practical capacity, giving a cost per minute. The second is a time equation: a base time plus a term for each characteristic of a transaction that makes it take longer. Minutes multiplied by the rate give the cost of any transaction. Because the rate is built on practical capacity rather than on the minutes actually worked, the difference between the two is reported as the cost of unused capacity rather than spread across customers.

How do you calculate the capacity cost rate?

Divide the cost of capacity supplied by the practical capacity of the group. The numerator is everything you must pay to have that group ready to work: fully loaded employment cost, supervision, equipment, the space it occupies and the systems it uses. It excludes anything already traced directly to a product or customer, the cost of other resource groups, corporate cost with no causal link, and one-off items. The denominator is the time the group can genuinely deliver, which is contracted attendance time less absence, breaks, start-up, handover and briefings. In the worked example on this page, 448,800 euro of cost over 748,000 practical minutes gives a rate of 0.60 euro per minute.

Why practical capacity instead of theoretical capacity?

Because nobody supplies theoretical capacity, so nothing should be priced against it. Dividing by theoretical capacity understates the rate and leaves an unexplained gap between the model and the ledger. The opposite error, dividing by the minutes actually worked, is worse: the model then balances perfectly every month and unused capacity disappears by construction, so a quiet quarter automatically makes every customer look expensive. Practical capacity is the only choice that both prices the work correctly and keeps idle time visible. Kaplan and Anderson suggest roughly 80 per cent of theoretical for people and around 85 per cent for machines as a starting point, but the figure should be derived from the actual calendar and shift pattern and then compared with the rule of thumb.

What is a time equation, with an example?

A time equation estimates the minutes one execution of a process consumes: a base time plus a term for each driver. Quantities enter as multipliers and conditions enter as flags worth zero or one. The pick-and-pack equation on this page is T = 4.0 + 1.2L + 0.9C + 7R + 12E, where L is order lines, C is cartons, R is one if the order contains a refrigerated line and E is one if it needs export documentation. A four-line, two-carton domestic order takes 10.6 minutes. A twenty-two-line, nine-carton refrigerated export order takes 57.5 minutes. Same activity, same shift, five and a half times the work.

What happens to unused capacity in TDABC?

It is reported as its own line on the profit and loss account, valued at the same capacity cost rate that priced the work, and it is not charged to any customer or product. In the worked example the team supplied 748,000 minutes, customers consumed 659,000, and the remaining 89,000 minutes cost 53,400 euro, which is just under one picker of the eight. This is the single most important difference from traditional absorption costing, where that amount has nowhere else to go and is spread over customers who did not cause it. The idea comes from Cooper and Kaplan in 1992: resources supplied equals resources used plus unused capacity.

How is TDABC different from activity-based costing?

The ambition is identical and the measurement instrument is different. Classic ABC breaks the business into activities and asks people what share of their time each activity took, which produces a cost pool and a survey for every variant of a job. TDABC estimates how long a task takes, expresses the variants as terms inside one equation, and lets the transaction data do the counting. That makes it far faster to build, much cheaper to update, and able to scale without the activity list exploding. It also surfaces unused capacity, which surveyed percentages hide, because percentages always sum to 100 whatever the workload was.

When should you not use TDABC?

When the work is non-routine and judgement-heavy, so that two tasks of equal length consume very different resources. When the cost base is overwhelmingly bought-in material or per-transaction third-party charges, so a capacity model would explain a thin slice at full effort. When capacity is genuinely bought per transaction, as in fully outsourced fulfilment, because there is then no unused capacity to find. When the systems record nothing that distinguishes one transaction from another and nobody intends to start. And, most decisively, when no one in finance will own the model after go-live, in which case build a one-off analysis and describe it honestly as a snapshot.

Who created TDABC and when?

Robert S. Kaplan and Steven R. Anderson. Anderson had built time-based activity models at Acorn Systems, which he founded in 1996, and the capacity-costing groundwork was laid in Cooper and Kaplan's 1992 Accounting Horizons paper and in Kaplan and Cooper's 1998 book Cost & Effect. The method was named and introduced in their Harvard Business Review article of November 2004, volume 82, number 11, pages 131 to 138, and set out in full in their 2007 book with Harvard Business School Press.