ROI and decisions ยท 5 min read
Where to Automate: Decide With Numbers, Then Prove the Return
By FlowSharp
November 11, 2026
Every automation project promises savings, but few are checked afterwards. How to choose where to intervene and how to prove the result against a frozen baseline.
Promised savings versus verified savings
An automation project usually starts from a quote and an estimate: hours freed up, fewer errors, cases closed faster. These figures are plausible, but they often stay what they were at the start: estimates. Once the project is over the team moves on, and the question 'how much did we actually save?' has no answer that would hold up in front of a CFO.
The problem has two halves. The first comes before the investment: if you do not know what your processes really cost, you may automate what is most visible or most noisy rather than what weighs most. The second comes afterwards: without a fixed starting point, you end up comparing a result with a memory. This article deals with exactly these two halves: the investment decision and the proof after the fact.
What McKinsey says about digital transformations, and what it does not
According to McKinsey, in its report 'Unlocking success in digital transformations' (October 2018), based on a survey of 1,793 participants carried out in January 2018, only 16% of respondents say their digital transformation improved performance and equipped the organization to sustain changes over time. Another 7% report improvements that were not sustained.
The limits should be stated plainly. The survey covers digital transformations in general, not process automation in particular. The answers are self-reported, so they reflect how respondents perceive their results rather than an independent measurement. The data dates from 2018. For these reasons the 16% should not be read as the odds of success for a single automation project in a small or mid-sized company.
What the figure does describe is a problem: improving is one thing, sustaining the improvement is another, and the difference only shows up if something was measured. It does not prove that any tool, FlowSharp included, solves that problem. It can, however, be a good reason to ask, before approving a project, how its result will be verified.
Before the decision: what each process really costs
The Analysis section of FlowSharp shows average Lead Time and the period's Total Cost, which includes human labor, automations and AI tokens. It also shows how cost and time are split, the Flow Ranking, and the points where cases get stuck.
For anyone deciding where to invest, the difference is practical. A flow with a high cost and cases that stall at a specific step is a different candidate from a flow that only looks heavy at first glance. The analysis does not decide for management, but it puts comparable evidence on the table instead of impressions.
The ROI Simulator helps turn that comparison into figures. You enter the human cost of a task, its frequency, the time that could be saved and the cost of the automation or AI; the simulator returns the estimated saving and the return over time. It remains an estimate, and it is only as good as the data entered. AI cost is also counted explicitly: the cost of an AI scale adds up each evaluation at the price of its model.
Fixing the starting point with a frozen baseline
The cost baseline captures a process before the improvement. It records the measured cost per case, the volume, the estimated manual work and the quality of the data. Recording data quality means stating up front how solid the foundation for the comparison is.
A baseline value can only be corrected with a traced justification, so every change leaves a readable explanation. Freezing then makes the baseline immutable, with a SHA-256 fingerprint.
For a CFO the point is practical. The starting point cannot be adjusted after the fact to make the result look better, and if a value was corrected before freezing, the reason is on record. The final comparison is then between two measurements, not between a measurement and a recollection.
Proving the result: measured savings and estimated capacity
The premium Saving Challenge module links the baseline to a duration and to economic terms. It distinguishes measured (hard) savings from estimated capacity, consolidates the months, and allows an Excel export after the first consolidated period.
The distinction matters for anyone who has to answer for the numbers. Hours freed up are not automatically euros saved: estimated capacity can have value, but it should be presented as what it is, separate from the savings that show up in measured costs.
The Automation Trajectory adds another checkpoint. It places the structural percentage, meaning how the process is designed, next to the automation observed in the period, meaning how it actually runs. You can set a target, for example from 25% to 45% by a given date, linked to the Implementation Plan, and check over time whether the gap between design and practice is narrowing.
In short, before approving a project it is worth asking three questions: what does this process cost today, against which frozen baseline will the change be measured, and when and by what criteria will the result be verified. They are simple questions, but they move the discussion from promise to proof.