Free tool
AI ROI Calculator
Work out what a piece of work costs you today, how many of those hours you could reclaim, and what that makes it worth spending. Every assumption is on the page and every one of them is yours to argue with.
What the AI ROI Calculator Is
The CoFabrix AI ROI Calculator is a free tool that works out what one process costs a business each year, how many of those hours could be reclaimed, and what that makes it worth spending to change. The visitor splits those hours by hand into three buckets: work a person drives, work that could run on its own, and work that needs judgement. Every constant it uses is published on the page, so a visitor can check every number that produced the result.
How this works. You enter three numbers, split the hours into three buckets by hand, and the page does the arithmetic. Two minutes, no sign-up, and no email required to see the answer.
Start with one process, not a whole department. The right size is work that already has a name people use: claims intake, quoting, scheduling, or code review. If you would have to explain what you mean, you have scoped too wide.
Your numbers
The result updates as you go.
What you can measure
Numbers you already have.
How many people spend time on it today.
Time one person spends on it in a typical week.
Base salary per person, before benefits and overhead. The model adds those.
Payroll taxes, benefits and overhead on top of base pay. Set it to your own loaded rate if you know it.
What you expect
Judgements about your own operation, not measurements. These are what drive the range.
Drag the dividers. Count the hours, not the jobs.
- Assisted, a person still drives it
- Runs on its own once the rules are set
- Needs a person's judgement
Not sure how to split them?
Take one person's week and walk through it.
The forty minutes spent looking something up in another platform and pasting it into the file is assisted. They are still in charge of the outcome, but the lookup and the re-keying would compress hard.
The status update that goes out every Tuesday, built from the same three fields every time, runs on its own.
The call with the customer whose request was denied needs judgement, and it stays that way.
If you cannot decide which bucket an hour belongs in, put it in judgement. The model will understate rather than oversell, which is the direction you want to be wrong in.
How much of each bucket is actually reclaimable
Neither bucket goes to zero hours. Assisted work still needs someone reading and approving, and work that runs on its own still needs watching. These are estimates, not measurements. Change them.
Fixed. Not an estimate.
45% assisted x 45% = 20.3 pts
25% self-running x 85% = 21.3 pts
30% judgement x 0% = 0.0 pts
= 41.5% of these hours reclaimable
Real, but deferred, and it depends on the growth actually arriving.
Reclaimed hours become dollars only when they change what you actually spend. This is the discount for that gap, and nothing else.
Licensing or usage fees, hosting, monitoring and maintenance after go live.
Net annual benefit $18,547. Worth spending up to $55,642 for a 36 month payback.
How the number is built
Each line is one step of the arithmetic. Change anything above and watch which line moves.
- What this work costs you per year
- $91,000
- 42% of these hours are reclaimable
- $37,765
- 65% of that reaches the budget
- $24,547
- less running cost of $500 a month
- -$6,000
- Net annual benefit
- $18,547
Those two percentages multiply. You have said 27% of what this work costs today can reach the budget, before running cost.
The shape of this work
Your reclaimable time splits fairly evenly between work a person still drives and work that could run on its own. Those are two different builds, with different risks and different timelines. They are usually sequenced rather than attempted at once.
Which reading you get depends on how your reclaimable hours divide between the first two buckets. The thresholds are set at one third and two thirds.
Where the money goes
The same arithmetic as the table above, drawn.
| Step | Amount |
|---|---|
| Annual cost | $91,000 |
| Not reclaimable | -$53,235 |
| Does not convert | -$13,218 |
| Running cost | -$6,000 |
| Net benefit | $18,547 |
What this is worth spending
The most you could invest and still be whole inside each window, at your numbers.
- Pays back in 12 months$18,547
- Pays back in 24 months$37,095
- Pays back in 36 months$55,642
Spend more than the 36 month figure and this work does not justify the project on cost alone. That does not make it the wrong thing to do, but the case has to come from somewhere other than this calculation.
Already have a quote, or a budget in mind?
Enter it and the payback view appears below. Leave it blank and it stays hidden.
Design, integration, testing and rollout. Vendor fees plus your own team's time if you want to count it.
42% of these hours are reclaimable.
Every assumption this uses
Every assumption behind the arithmetic above, with the reason it holds that value. Rows marked (Estimated) are a choice we made rather than a measured fact, so those are the ones to argue with.
| Assumption | Value | Why |
|---|---|---|
| Working weeks per year | 46 | 52 weeks less typical paid time off and public holidays, so the work is not costed for weeks nobody works. |
| Hours worked per year | 1,840 | 46 weeks at 40 hours. The same definition of a year used above, so an hour of task time and an hour of cost are the same hour. |
| Fully loaded multiplier(Estimated) | 1.3x, editable | Payroll taxes and benefits alone typically run above this. We use 1.3x deliberately so the model understates what the work costs rather than inflating the case for changing it. |
| Assisted hours reclaimable(Estimated) | 45%, editable | Assisted work keeps a person in charge. They read, check and approve, so the hours compress rather than disappear. Held below half deliberately. |
| Self-running hours reclaimable(Estimated) | 85%, editable | Work following a fixed sequence still needs monitoring, exception handling and the occasional intervention. Nothing goes to 100%. |
| Judgement hours reclaimable | 0%, fixed | Not an estimate. It is the boundary of the model. Hours you classify as needing judgement are never counted as reclaimable. |
| Conversion to budget(Estimated) | Set by mechanism | 85% for displaced contractor or overtime spend, 65% for avoided hiring, not applicable for redeployment. Estimates with reasoning, not measured findings. |
| Range width(Estimated) | Plus or minus 15 points | Applied to conversion only, because that is the input least anchored in anything you can check today. |
| Ramp to full effect(Estimated) | Immediate | The model assumes full effect from month one. This is its least conservative assumption and it will overstate the first six to twelve months. |
| Discounting(Estimated) | None | No discounted cash flow. Over a two to four year window on figures this rough, it would add precision the inputs cannot support. |
Questions about this calculator
- How big a piece of work should I put in?
- One process, not a department and not a single keystroke. The test that works: if the thing you are modeling already has a name people in your company use in conversation, it is the right size. If you would have to explain what you mean before anyone understood you, it is too broad, and the hour split at the end will turn into guesswork.
- What counts as a reclaimed hour?
- An hour that stops being spent on this work. That covers two different situations. Some hours stop because the work runs on its own once the rules are set. Others stop because someone is still driving but the lookup, the first draft and the re-keying take a fraction as long. The calculator treats those separately, because they do not compress at the same rate, and it never counts hours you have marked as needing judgement.
- What exactly am I buying with the investment figure?
- Rarely just software. Work like this usually breaks into three parts: the model or algorithm doing the reasoning, the technology and data plumbing around it, and the change to how work actually flows through the team. BCG's framing puts roughly a tenth of the effort in the algorithm, a fifth in technology and data, and the remaining seventy percent in people and process. That last part is the one most often missing from a budget. If your one time cost only covers building something, it is probably light.
- Why does it not ask what I plan to spend?
- Because most people do not know yet, and that is usually why they are here. Everything else this asks for comes off your payroll or out of your own observation. Cost is the one number you would have to invent, and inventing it would drive the whole answer. So the default runs the arithmetic backwards and tells you what the work is worth spending on instead.
- Are these CoFabrix prices?
- No. There is no CoFabrix pricing anywhere in this model. The ceiling it produces is derived entirely from figures you entered about your own operation, which means it is a number you can take to any vendor, including one that is not us.
- Why does the answer sometimes say there is no cash saving?
- Because sometimes there is not. If reclaimed hours get redeployed to other work rather than displacing contractor spend, overtime or a hire, the payroll does not change and neither does the budget. What you gained was capacity. That is worth having and worth measuring, but calling it a saving would be wrong, so the calculator does not.
- What does it deliberately leave out?
- Revenue uplift, per use case time saved benchmarks, and discounted cash flow. Each would require a figure we would have to invent. It also assumes full effect from month one, which is stated in the assumptions table because it is the model's most optimistic move.
You just estimated how your team's hours divide
Three buckets, dragged by eye: work a person still drives, work that could run on its own, and work that needs judgement. That split moved the answer more than anything else on this page, and nobody can get it right from a chair. Two days on site, at no charge, replaces the estimate with an observation of how the hours actually go.