Run the model
Run the model
Six inputs, all of which you already know. The results update as you type, including when the answer is that the automation is not worth doing.
Your six inputs
A worked example is loaded so the arithmetic is visible. The figures are placeholders for the calculation, not VSI prices, averages, or anybody’s results. Replace every one with your own.
How many times the task happens per week. Count one week rather than estimating.
Including the cost of switching to it, not the time it takes when someone concentrates.
Hourly cost including employment costs, not salary divided by 2,080.
Never 100%. Use the share of cases that are genuinely routine, measured rather than assumed.
What mistakes in this task currently cost per year. Usually the largest term and the least measured. Enter 0 if unmeasured.
The model works with anybody’s quote. Ours is fixed before you commit.
Monthly operation: hosting, model usage, monitoring, maintenance.
Where the benefit comes from
- Labour recovered
- $61,152
- Avoided rework
- $12,000
Cumulative position, 36 months
Starts at the build cost and climbs by the monthly net. Crosses zero at month 13.1.
Sensitivity: the automation rate
The largest distortion in this model is an optimistic rate, use the share of cases that are genuinely routine, measured rather than assumed. The same arithmetic at three rates:
- Conservative55%
- $60,048
- Your input70%
- $73,152
- Optimistic80%
- $81,888
The calculation
annual benefit = (volume × 52) × (handling minutes ÷ 60) × loaded hourly cost × automation rate + avoided rework cost
payback months = build cost ÷ ((annual benefit ÷ 12) − monthly run cost)
- Source
- Published model, not a VSI measurement
- Period
- Annualised from weekly volume
- Method
- Labour recovered plus avoided rework, less run cost. Deliberately excludes revenue upside, which is real and unprovable before the fact.
The model
How the estimate is built
No industry averages, no research-firm statistics. Each input is a fact about your business, and the quality of the answer is the quality of these six.
- 01
Volume
How many times the task happens per week. Count it for one week rather than estimating; the estimate is almost always low.
- 02
Handling time
Minutes per instance, including the interruption cost of switching to it. Not the time it takes when someone concentrates.
- 03
Fully loaded cost
The hourly cost of whoever does it now, including employment costs, not their salary divided by 2,080.
- 04
Automation rate
The share that completes without a human. Never 100%, and the residual is where the design effort goes.
- 05
Error and rework cost
What a mistake currently costs, and how often. This is usually the largest term and the one nobody has measured.
- 06
Build and run cost
One-time build plus monthly operation. Ours is quoted before you commit; the model works with anybody’s number.
What the model excludes
What the model includes
Naming the exclusions is what makes the rest of it usable.
| Criterion | Included | Excluded |
|---|---|---|
| Labour | Time recovered at fully loaded cost | Headcount reduction, rarely the actual outcome |
| Quality | Avoided rework where it is measured | Reputational cost of errors, real, unquantifiable |
| Revenue | Nothing | Faster response improving conversion, plausible, unprovable in advance |
| Capacity | Nothing | Work the team can now take on instead, usually the biggest term |
| Cost | Build and monthly run cost | Internal change-management effort, yours, not ours, but real |
The excluded column is where most of the upside actually lives. It is excluded because it cannot be estimated honestly before the deployment, and a model that includes unprovable terms is a model built to produce a number rather than to inform a decision.
Next
Run it with us
Twenty minutes. Bring one workflow and rough numbers for the six inputs above, and you will leave with a payback estimate, including when the answer is that it is not worth doing.
Book a 20-minute assessment