Automation ROI models that survive a CFO's scrutiny
A free reference library of editable spreadsheets for sizing agentic automation — payback period, fully-loaded cost-per-task, error-cost avoidance, and fleet-scaling math, with every assumption shown.
- Payback & breakeven calculator
- Fully-loaded cost-per-task model
- Error & rework cost avoidance
- Fleet-scaling sensitivity tables
Most automation ROI decks are vibes with a chart on top
We got tired of pitch numbers that fall apart the moment finance opens the cells.
The hard part of an automation business case isn't the headline savings — it's the denominator. A model that only counts loaded salary divided by task volume ignores the cost of the agent itself: tokens, orchestration, human review on exceptions, and the engineering to keep integrations from drifting.
These templates are the ones we use internally before we quote a build. They force you to enter the unglamorous inputs — exception rate, review minutes per exception, model and infra cost per run — so the payback number you walk into a budget meeting with is one you can defend line by line.
Five models, one shared methodology
Each is a standalone spreadsheet with labeled inputs, a sensitivity table, and a short note explaining what it does and does not capture.
Payback & Breakeven
Enter build cost, run cost, and monthly savings to get payback in months and a 24-month cumulative cash curve.
Cost-per-Task
Fully-loaded per-task economics: human baseline vs. agent run cost including tokens, review, and exception handling.
Error & Rework Avoidance
Quantify the savings from fewer downstream errors — the line item most ROI decks leave out entirely.
Fleet-Scaling Sensitivity
See how unit economics shift as you go from one workflow to a governed fleet, with volume and cost tiers.
Build-vs-Buy
Compare an in-house agent build against a managed engagement across a three-year horizon.
Methodology Briefs
The technical briefs behind every assumption, so finance and engineering read from the same page.
From download to defensible number
Roughly an afternoon of work to turn a template into your own business case.
Pick a model
Start with Payback if you have one workflow in mind, or Fleet-Scaling if you're sizing a program.
Swap our defaults
Replace benchmark inputs with your task volumes, loaded wage rates, exception rate, and run costs.
Read the sensitivity
Use the low/base/high columns to see which assumptions actually move the payback — usually exception rate.
Pressure-test it
Bring the populated model to a working session and we'll stress the inputs against real deployment data.
Exception rate is the whole ballgame
Every automation looks profitable when you assume 100% straight-through processing. Real agents hit edge cases, and each exception pulls a human back into the loop — that review time is the single largest swing factor in agent ROI.
Our models make exception rate and review-minutes-per-exception first-class inputs, then chart how payback degrades as they rise. It's a less flattering picture than a vendor slide, and that's exactly the point: budgets approved on honest numbers don't get clawed back in month four.
- Exception rate as a primary, visible input
- Review minutes priced at loaded labor cost
- Breakeven shown across a range, not a single point
A typical ROI deck vs. these models
Same goal, very different rigor.
| A vendor ROI deck | An Automatic.co model | |
|---|---|---|
| Cost basis | Loaded salary only | Salary + tokens + infra + review |
| Exceptions | Assumed away | A primary input with sensitivity |
| Formulas | Hidden or hard-coded | Unlocked and auditable |
| Scenarios | One optimistic case | Low / base / high |
| Ownership | Their spreadsheet | Yours to edit and keep |
Frequently asked questions
What format are the models in?
Each model ships as a documented spreadsheet (Google Sheets and Excel) plus a one-page methodology note. Every input cell is labeled, formulas are unlocked, and assumptions are called out so finance can audit the math.
Do I have to talk to sales to get them?
The library is gated by a short form so we can send the right templates and a sane set of starting benchmarks. There's no sales gauntlet — book a call only if you want help plugging in your own numbers.
How accurate are the default benchmarks?
Defaults are drawn from deployments we've run across regulated industries, but they're starting points, not promises. The models are built so you replace our assumptions with your own task volumes, wage data, and error rates.
Can you build a custom model for our workflow?
Yes. On a working session we'll take one real workflow, instrument it for cost-per-task and cycle time, and hand back a model wired to your numbers — the same artifact we use internally before any build.
Keep reading
The rest of the reference library.
Bring your numbers. Leave with a defensible payback.
One working session: we instrument a real workflow and hand back an ROI model wired to your task volumes, costs, and exception rate.