Fixing Savings Estimates That Get Rejected

01What gets challenged is the source, not the formula

Say you expect to save around forty hours a month. The response is rarely about the calculation. It is about how the forty was produced, and the moment it becomes clear the number came from intuition, every figure after it is read with suspicion.

Approvers are not trying to deny that savings exist. They are avoiding a position where they cannot explain the gap if the estimate turns out wrong. A number with no traceable origin puts them there.

Behind the scrutiny of sources lies a gap between efficiency and revenue. In IPA's DX Trends 2026, covering 1,799 Japanese companies surveyed between 17 April and 12 June 2026, 91.6% named process efficiency among the areas where they were seeing results, while only 3.9% named revenue growth. Management is cautious about savings estimates because it has watched that gap open repeatedly.

Before you submit an estimate, check that you can answer the three questions below for every figure. Any figure that fails one of them is likely to stall the meeting.

• When was it recorded, for which process, and by whom?
• If it was scaled to a month or a year, what was multiplied by what?
• If it is a post-change figure, which item falls, and because of which change?

02How to measure one cycle before you start

Before making any change, measure a single cycle of the target process. For a monthly process that means one month; for a daily one, a representative day. No tooling is required; a log kept by the operator while working is enough.

If you have not yet decided which process to measure, run a process inventory first. Limit measurement to the processes where you intend to claim savings.

Measurement changes the shape of the argument. Instead of "about forty hours a month" you can say "42.5 hours last month, of which 11 were spent waiting." The second is harder to dispute, and when it proves wrong you can return to the record and find out why. Traceability is what approvers are actually asking for.

03What to capture so the number holds

Four items are enough: time spent processing, time spent waiting on another step, rework loops, and cases where the operator hesitated on a judgment. Waiting and rework matter most—they show the largest improvement and nobody counts them today.

Volume is not the point. One measured cycle supports a monthly or annual projection. What matters is having any grounded basis at all, rather than a large sample.

The log sheet below is all you need. Paste it into a spreadsheet as CSV. Leave the waiting reason and hesitation detail as free text and categorize them when you total up.

Measurement also produces a by-product: you learn where the time actually goes. That makes it the first step of choosing what to improve, not a separate exercise.

04The time goes into collecting and reconciling data

When you measure, you often find that most of the time goes not into processing itself but into collecting and reconciling data. Sustainability disclosure data collection is a typical case. Each site aggregates data differently, at a different level of detail, with different fields, so exceptions pile up in head-office consolidation.

Someone also has to merge fields that have different names but mean the same thing. On top of that, tracking which department was asked for which data, and whether it has arrived, takes real effort. Because this work lives in people's heads, it rarely appears on a process list and is easy to leave out of an estimate.

When you build a savings estimate, measure collecting, reconciling, and chasing as separate lines. These are the areas with the most room for improvement that nobody normally counts.

05Separate the figures into three layers

Present three layers distinctly. First, measured values: when, which process, how many hours, stated as recorded. Second, projected values: the monthly or annual extension, with the arithmetic shown. Third, expected values: how far the number should fall after the change, with the lower confidence stated plainly.

Mixed together, the whole set reads as speculation. Kept apart, the first layer stands as fact. When you write expected values, tie each one, line by line, to a measured item, a specific change, and an amount. Any reduction you cannot tie that way is safer left out of the expected layer.

Do not force effects that resist monetization into a monetary figure. Days of stoppage if one person leaves, or the probability of a delayed close, can be shown in their own units. Converting everything to money weakens exactly the part with the thinnest basis, and drags the credibility of the rest down with it.

If you do convert to money, state where the hourly rate came from. Use the rate HR or finance already uses internally if one exists. If you know the approver works from a different rate, use theirs; the discussion moves faster.

• Keep measured, projected, and expected values separate
• Show the arithmetic behind projections
• Tie each expected value to a measured item and a change
• Leave hard-to-monetize effects in their own units

06Worked example: a monthly accounts payable run

The following is a fictional example. It shows one cycle of an accounts payable run logged on the sheet and then laid out in the three-layer table. Every number here is illustrative; none of it is survey data or a client result.

Look past the size of the numbers to the fact that every row carries its source and arithmetic. The 72 hours in the expected layer map line by line to measured items: waiting time and rework. If an approver asks where the 72 came from, the answer is already on the table.

Note also that the 14 hesitations are not converted to money. The effect of writing down account-coding rules is easier to defend as a case count compared month to month.

07Stress-test the estimate with AI before the meeting

Whoever built the estimate tends to assume the sources are obvious. Having something look for gaps from the approver's side before the meeting lets you find the questions you cannot answer while there is still time.

Paste your estimate text or table into the prompt below. The key is that the AI only classifies and flags; it must not fill in numbers. If you use an external AI service, follow your internal data-handling rules and mask company names and amounts first.

Act as the approver and review the savings estimate below.

# What to check
1. Classify each figure as "measured", "projected", or "expected".
2. List any figure whose source (when, which process, who recorded it) is not stated.
3. Recalculate every projected figure and point out anything that does not add up.
4. For expected figures, list any reduction that is not tied to a specific measured item and a specific change.
5. Give the three questions an approver is most likely to ask first.

# Rules
- Do not fill in or guess any numbers.
- Where information is missing, write "not stated".

# Estimate
(Paste the estimate text or table here.)

08What to keep measuring after approval

If measurement stops at approval, the next proposal starts from the same position. Capture the same items after the change, using the same method. Kept comparable, the following case begins from actuals rather than estimates.

There is a second benefit. When results fall short of the expectation, you notice early. Correcting course after one month costs far less than discovering it in a review six months later.

Most savings estimates that fail do so because the supporting material was never gathered before the work began, not because the presentation lacked polish. Before your next estimate, start by handing out a one-cycle log sheet.