Time saved / month
Fill in the required inputs
Hours of human labor freed, before quality adjustment.
The AI ROI calculator
Time saved, hard-dollar savings, soft capacity, and payback — every output a range, not a point. The math is on the page. The assumptions are on the page. The hard-vs-soft split is answered by you, not hidden in a footnote.
Range, not point
Every output on this page is a range: low · expected · high. Every input is uncertain — handle times are medians, exception rates vary by team, AI cost per execution varies week-to-week. A single point estimate implies precision that doesn't exist.
The confidence band below controls how wide the range is. 20% is the default. Widen it if you're estimating. Narrow it if every input came from system data.
Setup
Workflow identity. Two fields — name and analysis date — so this calc is found again and attached to a pilot.
Use the same name as the Pilot Plan Worksheet. If it doesn't match, the ROI number can't be traced back to a pilot plan.
Volume
Executions per month. Steady state, not peak. If seasonality matters, use the average over the last six months of comparable months — not the busiest one.
exec/mo
One execution = one run of the workflow (one ticket processed, one claim reviewed).
Applied to net savings per execution to produce the low/high ends of the range.
Today's workflow
The pre-pilot state. Total minutes per execution including rework — if the workflow is 6 minutes of work with 20% rework, enter 7.2, not 6.
min
$/hr
Loaded = salary + benefits + overhead. Ask HR for the fully-loaded rate, not salary ÷ 2080.
Tomorrow's workflow
The post-pilot state. Includes the automated portion and the HITL review portion. Exception rate is the percentage of executions where a human actually intervenes.
min
Human time spent watching or navigating the automated step. Often near zero. Don't enter the model's processing time — that's not human labor.
min
The time a human spends on an execution that requires intervention.
%
Percentage of executions requiring HITL review. 100% = HITL on every action; 20% = 1 in 5; 0% = fully automated.
%
% of executions that come back for rework. Applied to the review minutes as a multiplier.
AI costs
The cost to run the workflow. Per-execution cost is token/API spend. Fixed is the platform fee. One-time is setup, integration, and initial training.
$
Token cost per execution. If your pricing model is per-volume rather than per-token, this is the per-execution figure.
$/mo
$
Integration build, initial training, review checklist design. Not recurring.
Most one-time costs recur — re-integration, retraining, model migrations. Defaulting to annual is honest.
Freed time — the question every ROI tool skips
When the workflow saves 4 minutes per execution, what happens to those 4 minutes? If nothing — the team just works a bit less each day — there is no money saved; there is capacity. If headcount drops or contract labor ends, there is. Most tools add these together and call it ROI. This one won't.
%
The rest is soft. The tool will report both. Do not report just the hard share.
Your result
Every figure is a range. The low and high ends come from your confidence band, applied to net savings per execution. The expected value uses your inputs directly. The math is in the expandable block below — check it yourself.
Hard-dollar savings / month
—
The portion of saved labor that leaves the budget. Zero if you chose "soft".
Soft capacity / month
—
Hours redeployed, not cash saved. Real value, different line item.
Payback period
—
Months to recover one-time cost from net savings. Uses hard-dollar savings only.
Show the math
Every line below uses only the inputs you provided. Recalculate by hand if you want — anything you can't reproduce line-for-line is a bug, not a feature.
minute_rate = loaded_cost_per_hour / 60
today_labor = today_minutes × minute_rate
tomorrow_hitl = review_minutes × (exception_rate/100) × (1 + rework_rate/100)
tomorrow_labor = tomorrow_automated_minutes × minute_rate + tomorrow_hitl × minute_rate
tomorrow_ai = ai_cost_per_execution
net_saved_per_exec = today_labor − (tomorrow_labor + tomorrow_ai)
net_saved_per_month = net_saved_per_exec × volume_per_month
net_after_fixed = net_saved_per_month − fixed_monthly_cost
hard_savings_per_month = net_after_fixed × hard_share
soft_capacity_hours_per_month = (net_saved_per_exec / minute_rate) × volume_per_month × (1 − hard_share)
payback_months = one_time_cost / hard_savings_per_month (if hard_savings_per_month > 0)
low_end = expected × (1 − confidence_band)
high_end = expected × (1 + confidence_band)
Today's labor cost per execution
—
Tomorrow's labor cost per execution
—
Tomorrow's AI cost per execution
—
Net savings per execution
—
Net savings per month (before fixed cost)
—
Fixed platform cost per month
—
Net savings per month (after fixed, expected)
—
Assumption list is included in the CSV/JSON export. Attach it to the ROI number when you share it — a range without its assumptions is a guess.
See a worked example (insurance · claim summaries)
The case study from the hub's regional insurer case, run through this calculator. It shows what a filled-in analysis looks like, and why the hard-vs-soft question matters.
Inputs
- Volume
- 10,500 claims/month
- Today
- 8 min per claim · $85/hr loaded adjuster cost
- Tomorrow
- 0.2 min automated + 4.5 min review × 30% HITL rate, with 10% rework
- Costs
- $1.50/claim API + $8,000/mo platform + $45,000 one-time
- Freed time
- Mixed — 60% hard (contract adjusters reduced), 40% soft (capacity redeployed to complex claims)
- Confidence band
- 20%
Result
- Time saved / month
- ~780 hours (range 624–936 at ±20%)
- Hard-dollar savings / month
- ~$17,000 (range $13,600–$20,400)
- Soft capacity / month
- ~310 hours redeployed (range 248–372)
- Payback period
- ~2.6 months on hard dollars alone (range 2.2–3.2 months)
What this example shows
Most ROI tools would report $28,000/month in "savings" — the full labor substitution before the hard/soft split. That number doesn't survive a budget review, because only 60% of it is actually hard dollars. This calculator reports $17,000 hard and $11,000 of capacity as two different things, and the payback uses only the hard number. Slower payback. Defensible payback.
What this tool refuses to do
Five specific refusals. They're visible so you can hold the tool to them — and because most ROI calculators will happily do all five.
Point estimates.
Every output is a range. If you screenshot the "expected" number and hide the range, you've defeated the tool. The range is the deliverable.
NPV or IRR.
Those are finance models. This tool produces workflow-level inputs; the finance partner produces cash-flow timing, discount rate, and horizon. Two different jobs.
Qualitative benefits.
Employee satisfaction, CSAT improvements, reduced cognitive load — real, valuable, and not money. Don't ask this tool to include them. Put them in the narrative, not the model.
Downstream effects.
"If the workflow is 3 min faster, four downstream decisions get faster, which saves…" — that's a different model. This one prices the workflow in front of you.
A board-deck number.
This is the analysis the board deck rests on. Present the range and the assumptions, not the "expected" figure alone. If your board deck promises a single number, the deck is lying to itself.
Where to go next
The calculator is the last of the four chain tools. With a range in hand, the workflow enters Expand — or, if the range says stop, the pilot plan's stop rule kicks in.
Frequently asked questions
Why is every output a range instead of a number?
Because every input is uncertain. Handle time is a median across many workflows; HITL exception rate varies by team; AI cost per execution depends on token usage that varies week-to-week. A single point estimate implies precision that doesn't exist.
The range is a statement about confidence, not hedging — a good calculator tells you how much it doesn't know.
What's the difference between hard-dollar and soft-capacity savings?
Hard-dollar savings means the money leaves the budget — headcount that isn't backfilled, contract labor that's cancelled, overtime that stops. Soft capacity means the time is freed but the budget doesn't change — the team redeploys the time to other work.
Most ROI calculators add them together. That's how they produce numbers that don't survive a budget review. This one keeps them separate and asks which one applies.
Does this calculator account for time to deploy?
No. It assumes the workflow is already running at its target state. Time to deploy is a project workstream, not an ROI assumption — the one-time cost field captures setup, integration, and initial training, but not the months when you're building.
For a payback projection that includes ramp-up, use seasonality on the volume field or run the calculation at expected steady-state and discount the result.
Why doesn't this calculate NPV or IRR?
Because those are finance models, not workflow models. NPV requires a discount rate, cash-flow timing that accounts for when headcount actually drops, and a horizon that matches your investment committee's conventions.
This tool produces the workflow-level inputs those models need. Hand the CSV to your finance partner — don't try to make this page produce a board-room number.
What's the confidence band, and how should I set it?
The band applies a percentage swing to your net-savings-per-execution figure, producing the low and high ends of the range. Default is 20% — appropriate when your inputs are measured. Use 10% if every input came from system data. Use 30% or more if any input is an estimate.
The band width is not scientific; it's an honesty signal. A 40% band on a defensible plan beats a 10% band on a guess.
Does this work for a workflow with near-zero human labor today?
Not usefully. If there's no human labor in the workflow today, there's no time to save — the ROI is a cost question (can AI do it cheaper than the software you already pay for?) rather than a labor substitution.
Fill in the costs and set the time inputs to near-identical values, and you'll get a cost-only view. That's an honest output but a limited one.
Where does my data go?
Nowhere. Every input is saved to your browser's localStorage on this device. Nothing is sent to a server, no account, no email, no analytics tied to your numbers.
Use CSV or JSON download to move results to a team drive. If you clear your browser data, everything resets — download before clearing.
What does this tool refuse to do?
Five things. Produce a point estimate (every output is a range). Produce an NPV or IRR (that's a finance model). Include qualitative benefits (employee satisfaction, CSAT improvements). Include downstream effects (faster evidence elsewhere). Give you a number to put in a board deck without the assumptions attached — the assumption list is part of the export.