Two Projects Claim the Same Savings. What Belongs in the Decision Deck?
Two Projects Claim the Same Savings. What Belongs in the Decision Deck?
Two slides arrive on the same afternoon. Project A will cut six minutes from a support task. Project B will cut five. Each estimate was built against the same unchanged baseline — the work as it runs today — and each was built without knowing the other existed. The obvious next move is a summary slide reading eleven minutes.
That move assumes something nobody checked: that A and B remove different minutes.
Here is the arithmetic that should sit underneath the summary slide. Take a 20-minute task. A removes six of those minutes; B removes five; and three of the minutes A claims are the same three B claims — one re-entry step both owners are reaching for against the same baseline. Add the stand-alone figures and you get 11. Remove the overlap once and you get 8. The three-minute gap is not a rounding difference. It is the part of the claim both projects are pointing at, and the deck has to decide what to do with it.
Put both claims against the same unchanged work
Before six minutes can be compared with five, both numbers have to describe the same minutes. That requires the same task, the same people performing it, the same count of how often it happens, and the same comparison horizon.
Construct a deliberately tidy example. An operations analyst handles a customer reimbursement correction. Today the correction takes 20 minutes: look up the account, key the change, verify it, and re-enter one field into the finance system after it posts. The team handles roughly 40 corrections a week, and the proposal looks at the next 12 months. Intervention A is a pre-filled template; its owner estimates six minutes removed — three minutes of lookup and keying, plus three minutes of the shared re-entry. Intervention B is an auto-approval rule; its owner estimates five minutes removed — two minutes of review, plus the same three minutes of re-entry. Both figures are measured against today's 20-minute process. Neither is measured against the other.
The unit check matters as much as the baseline check. Six minutes per correction and five minutes per correction are comparable. Six minutes per correction and an hour per week are not: the second figure hides the volume the first one exposes. If you want an annual number, both estimates have to multiply by the same correction count over the same months — and the count is itself an assumption, since a change to the intake process can move it. Agreement on the baseline is necessary. It is not sufficient. Two teams can define the task identically, count the same corrections, share a horizon, and still be describing the same minutes twice.
The UK government's project-delivery guidance, the Teal Book, treats benefits against a counterfactual — the state that would have existed without the intervention — and asks how several interventions contribute to one outcome, when they contribute, and how their dependencies fit together (Chapter 19, benefits management). That is guidance for public project delivery, not a rule for a private deck, and it does not measure your task. But the comparison principle travels: with two interventions there are no longer two states to compare, there are at least four — neither change, A alone, B alone, and both together. A stand-alone figure describes one of those comparisons. It says nothing about the others.
Separate shared work from dependencies and alternatives
Two benefits can relate in four ways, and only one of them lets you add.
- Independent. A and B change different steps, with no shared minute and no interaction. The minutes add.
- Overlapping. A and B each remove part of the same step. The union is smaller than the sum. The example sits here.
- Mutually exclusive. A and B are two ways to remove the same step; you would choose one. The minutes do not add at all, because you only get one.
- Dependent. B only works once A exists. B's stand-alone estimate describes a world without A, which may be a world B cannot actually be built into.
Sorting the pair is not bookkeeping. It changes the question. In the example, A's six minutes break into three of its own plus three of the shared re-entry; B's five break into two of its own plus the same three. An overlap diagram puts both sets side by side, marks the three shared minutes once, and the sum of the stand-alone figures stops looking like a total. That is what the diagram is for. It exposes the question; it does not settle who gets credit for the shared minutes. An overlap graphic is a claim about the process — which step, which field, which person does that work today — and a claim about a process can be wrong. The example stipulates that the three minutes are genuinely shared. In a real proposal, that stipulation is a fact to verify, not a fact to assume.
Two related cases deserve the same scrutiny. If A and B were two different fixes for the same re-entry step, they would be alternatives, and the deck's real question would be which one to back, not how to present their sum. If B's auto-approval rule only functions once A's template is in place, then B has no honest stand-alone number for a universe without A — the five minutes describes a configuration that will never be procured on its own.
Implementation work belongs in the comparison, because it sits on the opposite side of the same ledger. A template and an approval rule each cost something to build, train, and maintain. If A changes the finance system's field, B's rule may have to integrate with a different field than its stand-alone estimate assumed, and B's build cost moves with it. A comparison that counts every minute of released time while ignoring the build does not describe a saving; it describes a gross benefit. Adverse interactions count too. If the two changes together trigger a new check — the rule flags corrections the template could not fill, say — that new work is part of the combined process and pulls the joint benefit down.
APMG International's explanation of benefit management from a portfolio perspective names the failure directly: the same benefit claimed to justify more than one initiative. It is a training provider's account of a known failure mode, so it identifies the mistake rather than measuring how often it happens or what it costs when it does. Naming it is still worth doing, because the overlap diagram in your deck is precisely the situation that failure describes.
Show either the joint result or a conditional increment
There are two honest ways to present the pair, and they do different jobs.
The joint account compares the combined process with the unchanged baseline. One 20-minute correction becomes a 12-minute correction. The pair reduces the work by eight minutes, and both owners are credited with the same combined outcome. The joint total does not depend on who describes it first.
The incremental account compares the second intervention with the state that would otherwise remain after the first. Put A first. After A, the correction takes 14 minutes; adding B brings it to 12, so B's additional contribution is two minutes — not its stand-alone five. Put B first. After B, the correction takes 15 minutes; adding A brings it to 12, so A's additional contribution is three minutes — not its stand-alone six.
Notice what changes and what does not. The joint total, eight minutes, is the same in either order. The increments are not. B adds two after A; A adds three after B. The sequence controls who is credited with the shared three minutes — whoever goes first carries them, because the second intervention only adds what the first left behind. Sequence does not change how much work the pair removes.
That distinction keeps an order-dependent increment from masquerading as a finding. You can sequence A first to give A the larger increment, or B first to give B the larger increment, and the total is unchanged. Sequencing to flatter an attribution is a slide decision. Sequencing because B cannot run before A is a delivery decision, and it should be argued on delivery grounds — dependencies, integration, who is available to build first — not on which order makes a project's incremental number look better. If the deck's ordering is doing work the evidence is not, someone in the room will eventually ask which sequence you are actually going to run, and on what grounds.
Hold the combined claim when its basis is missing
The eight minutes is arithmetic. It holds only while the sets are what the example says they are: A removes six, B removes five, and exactly three of those minutes are the same minutes. Subtraction gave you the union of two sets. It did not measure what happens when you run both changes together.
That is where subtraction stops being enough. Both changes act on the same re-entry step, and the example assumes that removing it once removes it fully and that nothing new appears in its place. A real process might disagree. The step could be replaced by something else — a correction the template leaves blank and the rule cannot approve, which somebody then resolves by hand. Or the template and the rule could disagree about the same field, and every correction could carry a manual reconciliation neither owner estimated. Nobody has run the two changes together on the same corrections, so nobody knows which of those happens. The joint eight minutes is a statement about two sets. It is not yet a statement about the combined process.
When the interaction is unresolved, the deck carries the joint account as conditional, or it carries a request for the missing evidence. Presenting eight minutes as a measured combined benefit is a claim the evidence does not support. Presenting it as an arithmetic result under a stated assumption — that the shared step disappears and no new work replaces it — is defensible, and it tells the reader exactly what to test. A range belongs here only when its ends have a basis; a lower bound pulled from imagination is not a hedge, it is a second invented number dressed up as caution.
The monetary step needs the same discipline. Released time and its dollar value are one benefit expressed two ways, not two entries. Eight minutes per correction, against the team's volume and its loaded cost, is a single gain described arithmetically and then financially. Listing "time released" and "value of time released" as separate lines adds the same minutes twice — the same error as adding A's and B's stand-alone estimates, just one layer further down. If the released time converts to cash because fewer hours are worked, or because hours move to work that would otherwise have been hired for, that conversion belongs to finance, with its assumptions showing. The deck can state the released time and the conversion rate. It should not present the converted figure as a new gain, and it should not derive attribution from an overlap graphic. Credit for the shared minutes is a process question, and the overlap diagram does not answer process questions; it only asks them.
The comparison the deck can support
What the summary slide wanted was one number. What the evidence supports is narrower, and it is more useful. Show both estimates against the one unchanged baseline. Show the shared minutes once, so the diagram states the problem instead of hiding it. Show what each intervention adds after the other — three minutes for A after B, two minutes for B after A — and keep both increments on the page so that nobody quietly returns to adding the stand-alone figures. Then name what remains open: whether the combined process actually removes the union, or whether a new step appears in its place.
If that interaction is unresolved, the supported decision is to go measure it — run both changes on the same corrections for a defined period, against the same baseline, and compare the combined process with the unchanged one. Until that evidence exists, present the joint account as conditional, hold the total, and let the deck do the job it can honestly do: show the choice that is still open, rather than close it with arithmetic the process has not confirmed.
Frequently asked questions
Why can't six minutes and five minutes simply be added?
They may overlap. In the example, A removes six minutes and B removes five, but three of those minutes are the same shared re-entry step. Adding the stand-alone figures gives 11; removing the overlap once gives 8. The gap is the part both projects are pointing at, so the deck must decide what to do with it.
What must be true before two minute estimates can be compared?
They must describe the same task, the same people performing it, the same count of how often it happens, and the same comparison horizon, against the same unchanged baseline. The unit check also matters: minutes per correction and an hour per week are not comparable. Annual figures must multiply by the same correction count over the same months, and that count is itself an assumption.
What is the difference between a joint account and an incremental account?
A joint account compares the combined process with the unchanged baseline: 20 minutes becomes 12, so the pair reduces work by eight minutes, and the total does not depend on who describes it first. An incremental account compares the second intervention with the state after the first: B adds two minutes after A, while A adds three minutes after B. The total is still eight, but the increments differ, and sequence controls who is credited with the shared minutes.
Why is the eight-minute result conditional, and what decision is supported if the interaction is unresolved?
Subtraction gives the union of two sets; it does not measure running both changes together. The example assumes the shared step disappears fully and nothing new appears. A real process might replace it with a correction the template leaves blank or the rule cannot approve, or create manual reconciliation. Until the interaction is resolved, present the joint account as conditional under a stated assumption, or request the missing evidence. The supported decision is to go measure it by running both changes on the same corrections for a defined period against the same baseline and comparing the combined process with the unchanged one.
How should released time and its dollar value appear?
They are one benefit expressed two ways, not two entries. Listing time released and value of time released would add the same minutes twice. If released time converts to cash because fewer hours are worked or because hours move to work that would otherwise have been hired for, that conversion belongs to finance with its assumptions showing. Do not present the converted figure as a new gain or derive attribution from an overlap graphic.