← All perspectives

Perspective / Strategy & transformation

Measuring transformation at the unit of value: cost-to-serve, throughput and revenue leakage

Transformation investment needs an economic model that survives contact with operational data. Connect technology events, service quality and cost allocation to a meaningful unit of value before presenting a return to the board.

10 min readReviewed

Finished precision components arranged beside a production line.
Conceptual editorial image.

Digital transformation should be measured through the work a business completes and the value that work creates. A lower technology bill can accompany a worse customer outcome; a higher bill can support profitable growth. Executives therefore need a model connecting investment, operating behaviour and commercial results. This article develops that model using an illustrative service-operation example. It separates unit efficiency, capacity, cash savings and attributable benefit so that a delivery programme can be challenged with evidence rather than supported by an attractive but ambiguous headline.

The executive view

  • Define a completed business unit and quality controls before calculating improvement.
  • Separate unit efficiency, capacity, cash savings and attributable commercial benefit.
  • Trace measures to source data and use explicit continuation or stopping decisions.
01

Choose a unit that represents completed business value

Start with the outcome the organisation is trying to improve. Suitable units might include a correctly fulfilled order, a resolved service case, an accepted booking or an approved application. A page view, API request or generated document can be useful operational evidence, but it is rarely sufficient as the primary commercial denominator.

Define completion carefully. An order that is cancelled, refunded or manually reconstructed may impose cost without delivering the intended value. Decide whether such work belongs in the denominator, in a quality measure or in a separate failure-cost calculation. Make that treatment consistent between periods.

The FinOps unit economics framework distinguishes resource-efficiency measures from business-unit measures. That distinction helps prevent a programme from proving that infrastructure became cheaper while leaving the economic performance of the service unexplained. Record the unit definition, exclusions and accountable owner alongside the formula.

Define the denominator through a business state rather than a convenient system event. A service case may be marked closed automatically after a timeout, while the customer's problem remains unresolved. Consider a completion definition that includes the agreed outcome and a defined observation period for repeat work. That definition will depend on the service; document why it represents value instead of assuming a status field already does so.

Keep comparable cohorts. A specialist complaint and a routine address change may both count as one case while consuming very different effort. Segment by meaningful complexity, then report both the segment results and a mix-adjusted total where appropriate. Otherwise a shift towards easier work can look like a technology improvement. The measurement owner should be able to explain every material change in the denominator before interpreting the unit cost.

02

Build a benefit tree with traceable assumptions

Connect each proposed intervention to an observable operational change and then to a commercial consequence. For example, better case routing may reduce avoidable transfers; fewer transfers may release handling capacity; released capacity may improve response time or accommodate additional demand. Each arrow is an assumption requiring evidence.

Separate benefits that can coexist from those that would double count the same change. A reduction in handling time cannot automatically be counted as both a full staff-cost saving and the value of all additional work completed with those same hours. The business must explain how the released capacity is actually used.

Assign an owner to each assumption. Engineering can establish that routing changed; operational management can verify the process effect; finance can challenge the cost treatment. A benefit tree becomes a governance tool when it identifies which evidence could invalidate a claim, rather than merely arranging optimistic numbers into a diagram.

Use a causal chain with observable intermediate evidence. If automated routing is expected to reduce transfers, establish the transfer count before delivery and identify which transfers are avoidable. If fewer transfers release time, measure handling and waiting separately. If the business plans to redeploy capacity, name the work it will absorb and the operational decision required to make that happen.

For each benefit, classify the commercial treatment: cash reduction, avoided future cost, additional capacity, revenue protection or a qualitative risk improvement. These categories cannot simply be added as though all were current cash. A hiring plan that can be postponed may support an avoided-cost case, while unchanged headcount producing better service supports a capacity case. Finance should approve the category as well as the arithmetic.

03

Worked example: lower unit cost without a lower total budget

Consider an illustrative service operation. Its baseline monthly operating cost is £96,000 and it resolves 12,000 comparable cases. After a change, monthly operating cost is £104,000 and it resolves 16,000 comparable cases.

MeasureBaselineLater period
Monthly operating cost£96,000£104,000
Completed comparable cases12,00016,000
Cost per completed case£8.00£6.50

The unit-cost reduction is 18.75%, while total spend has increased by £8,000. This could support a credible capacity argument. It does not establish a cash saving or prove that the technology change caused the improvement.

The example still requires checks on case mix, completion quality, seasonality and unrecorded rework. If the later period contains easier cases or transfers cost elsewhere, the apparent efficiency gain may be overstated. These are hypothetical figures, not KTE client outcomes.

At the later volume, applying the original £8 unit cost would imply £128,000 of monthly cost. Comparing that hypothetical amount with the actual £104,000 suggests £24,000 of capacity-related cost avoidance under a constant-unit-cost assumption. It remains a counterfactual, not money removed from the original £96,000 budget. Whether it is credible depends on the cost behaviour that would otherwise have supported the extra work.

Separate fixed, variable and step costs before extrapolating. A licence may rise only when another user band is reached; a support team may absorb some demand before needing another hire. The apparent saving at 16,000 cases might therefore differ from the saving at 13,000 or 25,000. Show the relevant demand range rather than multiplying the unit-cost change across an unlimited forecast.

04

Instrument the measurement before delivery changes the process

A reliable model needs traceable data. Define the business event that starts work, the event that establishes valid completion, and the identifier linking them. Record event time separately from ingestion time so that delayed reporting does not distort lead-time calculations.

Cost data also needs lineage. Decide how to allocate shared software licences, support, integration operation and relevant staff time. An allocation method does not become objective simply because it is expressed as a spreadsheet formula. Document its rationale and test whether a different reasonable allocation changes the investment conclusion.

Version the metric definitions. If the meaning of a resolved case changes midway through the programme, retain enough information to recalculate or clearly separate the periods. The measurement design should survive staff changes and supplier handover. A dashboard without documented definitions and source ownership is a presentation layer, not an auditable benefit model.

Preserve the difference between throughput and work in progress. A process can complete more cases temporarily by clearing an old backlog; that does not necessarily establish a sustainable increase in steady-state capacity. Record opening backlog, arrivals, completions and closing backlog with consistent definitions. Reconcile these quantities before drawing conclusions from a throughput increase.

Design identifiers to follow transfers and reopened cases. If each handover creates a new case identifier, the system may count extra work as extra value unless a parent journey reference connects it. Similarly, record correction costs against the relevant period and cohort. This data model matters commercially: an incomplete event trail can make the programme look efficient precisely where the organisation has lost sight of repeat work and unresolved demand.

05

Distinguish causal evidence from a favourable comparison

Before-and-after comparisons are useful, but many conditions can move at once. Demand, staffing, product mix, pricing and customer behaviour may change alongside a new system. Establish the strongest comparison the situation permits: a controlled cohort, phased rollout, matched segment or an explicitly limited time-series analysis.

Record assumptions before results are available. Select the observation window, exclusions and quality checks in advance where practical. This reduces the temptation to choose whichever period produces the most persuasive story.

The Service Manual’s measurement guidance distinguishes service performance from evidence about wider organisational objectives. For a commercial programme, use that distinction to keep process measures and investment claims separate. Where attribution remains uncertain, show a range and explain the uncertainty. Decision-makers can act on imperfect evidence more responsibly when its limits are visible.

Use staged rollout where operationally feasible. If comparable teams adopt the change at different times, the design may provide a better comparison than a single organisation-wide switch. Check whether teams differ in case mix, staffing or customer segments, and whether the control teams are indirectly affected. A plausible comparison needs an explanation, not merely a label.

Retain the original analysis plan and record changes. An exclusion can be valid when a source system failed, but removing an unusually difficult period because it weakens the result is a different decision. Report the effect of material exclusions. If the evidence supports improved operational performance but cannot isolate the technology's contribution, say that directly and use subsequent releases or targeted experiments to reduce the uncertainty.

06

Pair economic measures with quality and delivery controls

A programme can lower cost per case by closing cases prematurely, shifting work to customers or deferring difficult exceptions. Pair the economic measure with controls such as repeat contact, correction rate, unresolved backlog age and the proportion of work needing manual repair. Choose controls that would expose the specific ways the metric could improve while the service deteriorates.

Software delivery measures answer another question: whether the organisation can change the service effectively. DORA’s April 2026 Quick Check update includes deployment rework alongside throughput and stability. Those signals can explain delivery constraints, but they do not substitute for customer or commercial outcomes.

Use a compact scorecard with distinct layers: delivery capability, operational performance and economic effect. Do not combine them into an unexplained composite score. A board should be able to see which layer improved, which deteriorated and what decision follows.

Choose a guardrail that would detect each likely shortcut. For a case-resolution programme, track repeat contact within the agreed window, customer-reported resolution, unresolved high-severity cases and downstream manual repair. For an ordering programme, check cancellations, incorrect commitments and fulfilment exceptions. A generic satisfaction score may miss a concentrated failure affecting a small but commercially important group.

Inspect distributions and tails. A lower mean handling time can coexist with a growing queue of difficult cases. Segment by complexity and examine backlog age, including cases excluded from the completion denominator. Delivery measures should likewise stay attached to the service they describe. Frequent releases can support rapid improvement, but a release count says little about whether the organisation is solving the right commercial problem.

07

Make the investment decision explicit

A decision paper should identify the next commitment, the expected benefit mechanism and the evidence needed before further funding. Include implementation cost, ongoing operation, transition effort and the cost of retaining the present arrangement. Show sensitivity to the assumptions most capable of changing the decision.

Define continuation, correction and stopping conditions. If adoption remains low, the next action may concern workflow design or operational ownership rather than another release. If capacity improves but demand does not materialise, the commercial case may require revision even though the software works.

The strongest transformation model is explainable from source event to executive decision. It does not require certainty or a universal ROI benchmark. It requires a meaningful unit, disciplined cost treatment, an honest account of causation and named people who can act when the evidence changes. That is what turns measurement into practical ownership of an investment.

Stress-test the decision with a small number of plausible scenarios. For example, model demand below expectation, slower adoption and a longer coexistence period. Identify the assumption that moves the investment from attractive to unattractive, then decide what evidence is worth purchasing before committing further capital. This is more useful than a single precise ROI percentage built from uncertain inputs.

Assign benefits to operating owners who can change staffing, policy or workflow, not solely to the delivery team. Software can create an opportunity to reduce handling effort; management determines whether the opportunity becomes service capacity or cash. Schedule the benefit review after enough real work has occurred, and retain a route to correct or stop the intervention if quality deteriorates. Measurement earns its cost when it changes these decisions.

08

What a decision-ready transformation scorecard contains

A useful scorecard connects four things: the completed unit of value, the full cost boundary, the service-quality guardrails and the next investment decision. Show baseline and current performance with their observation windows, cohort definitions and material changes in demand. Keep cash effects separate from capacity and avoided-cost claims.

Add the confidence and limitations that affect interpretation. An incomplete staff-time allocation or immature adoption cohort can be acceptable for an early steering decision if it is visible. It becomes misleading when a provisional estimate is presented as an audited saving. The scorecard should identify which uncertainty the next measurement step will resolve and who will act on the result.

Commercial transformation governance is strongest when finance, operations and delivery can explain the same causal chain. The numbers then become a shared operating model rather than competing departmental accounts of whether the project succeeded.

Further questions

Questions this raises

Is lower cost per transaction sufficient evidence of ROI?

No. It must be considered alongside implementation and ongoing costs, quality, demand and attribution. A unit-cost improvement may release capacity without reducing cash expenditure, and a favourable comparison may reflect changes outside the programme.

Which transformation metrics should the board review?

A small set linking delivery capability, operational performance and economic effect. Each measure needs a definition, source, owner and decision it informs. Avoid a composite score that obscures deterioration in one area.

How should uncertain benefits be presented?

Show the assumptions, plausible ranges and evidence still required. Separate measured outcomes from forecasts and identify what would cause the organisation to continue, correct or stop the next investment commitment.

References

Sources & further reading

  1. FinOps Framework: Unit Economics
  2. Measuring the success of your service 6 August 2018
  3. DORA Quick Check updates 22 April 2026

Connected capabilities

Strategy & digital transformationSoftware & CRM

From thinking to doing

What could this mean
for your business?

Your next chapter

What could your
business do better?

Tell us where you want to go, what’s getting in the way, or what you need to deliver. We’ll start there.

Start a conversation

Start a conversation

Tell us what
you have in mind.

Interactive preview. Your information stays in this page and is not sent.