Automation ROI: How to Calculate It Before You Buy Anything

Every vendor who wants to sell you an automation has an ROI slide. It usually says something like "save 10 hours a week" at the top and nothing at the bottom. The savings are best-case, the costs are missing, and the whole thing is designed to be nodded at in a meeting, not audited.

I build automations for a living. I am going to show you the calculation I actually use with clients, the one where both columns are honest. The arithmetic is simple enough to do in a spreadsheet in twenty minutes. What makes it useful is that the assumptions are written down, so you can swap in your own numbers and watch what changes.

The formula, said plainly

ROI = (net annual benefit − total annual cost) ÷ total annual cost

That is it. The entire game is in what you allow into each column. Vendor math stuffs the benefit column with theoretical hours and empties the cost column after the purchase price. Honest math does the opposite: conservative benefits, complete costs.

What earns a place in the benefit column

Labor hours, at loaded cost. Not salary. Take the fully loaded cost of the person doing the task (salary plus taxes, benefits, and overhead, roughly 1.3 to 1.4 times base pay), divide by their annual working hours, and that is the rate every reclaimed hour is worth. If the person is the owner, use a real rate anyway. Your hour is not free; it is the most expensive hour in the building.

Rework and error correction. Every process where a human re-keys data between systems produces errors, and someone gets paid to find and fix them. Estimate the errors per month and the minutes per fix. If you cannot estimate it, that is a signal the process is worth instrumenting before it is worth automating.

Cycle time that touches cash. A quote that takes three days instead of one loses deals you will never count. An invoice that waits a week for batch processing is a week of float you handed to your customer for free. This one is real but hard to price, so I put it in as a note, not a number. If a number is hard to defend, it does not get to inflate the ROI.

Decisions made earlier. Reports that arrive on Monday instead of the 15th change what management does in the month. I refuse to put a dollar figure on this in the calculation. It is the quiet reason most automations pay back faster than the arithmetic claims, and it belongs in the conversation, not the formula.

What belongs in the cost column (where vendor math goes quiet)

Build cost. The one-time work: design, integration, testing, the ugly last 10 percent that always takes as long as the first 90.

Run cost. Licenses, hosting, connector subscriptions. Annualize everything. A $50-a-month tool is $600 a year forever, and "forever" is what the formula is measuring.

Maintenance. This is the line almost everyone forgets. Integrations rot, APIs change, the platform updates. A rough but honest planning figure is a few hours a year per integration, billed at whoever actually keeps it alive. If the answer to "who maintains this when we're done?" is nobody, treat that as a yes/no gate before you trust any ROI number, not a line item. See the second question below.

Transition. Training, the awkward first month where the new process is slower than the old one, the parallel run. Small, but it is a real cost and it belongs in writing.

A worked example, assumptions on the table

A distributor around $10M in revenue. Two staff spend about six hours a week each assembling a monthly reporting pack in Excel: exports, copy-paste, formatting, chasing the one person whose data is always late.

State every assumption, because the calculation is only as good as these:

• 12 person-hours a week, about 52 hours a month
• Loaded rate of $40 an hour. Swap in your own
• The automation removes 70 percent of that time, not 100. Someone still reviews the output. Review is the job; assembly was not
• Annual benefit: 52 hours × 12 months × $40 × 0.7 ≈ $17,500

Costs:

• Build: $6,000 one-time
• Run: $150 a month in tools and connectors, $1,800 a year
• Maintenance: about 4 hours a year at $40, call it $200, plus a buffer, say $500

Year one: benefit $17,500, cost $8,300, net $9,200. ROI ≈ 111 percent.

Year two onward: benefit $17,500, cost $2,300. ROI ≈ 660 percent, and the payback period on the build was under five months.

Notice what this example does and does not claim. It does not cite a study, because there is no study. These are your numbers or your guesses, and a guess written down is at least correctable. It does not promise 100 percent time savings. And it treats the build cost as real money, because it is.

Run it on your own process and the ROI sometimes comes out negative. That is the point. A negative ROI on paper is the cheapest bad decision you will ever make about automation.

Three questions that kill bad automations early

One: how many hours a month does this process actually consume, and who is paid for them? If nobody can count them, you cannot reclaim them. Instrument first. Two weeks of honest time-tracking beats any vendor's benchmark.

Two: who owns this in two years? An automation without a named owner will break quietly and nobody will know who to call. APIs change, staff leave, the one spreadsheet nobody understood disappears with someone's laptop. If there is no answer to this question, the honest ROI is "unknown," and unknown is not "positive."

Three: is the underlying data good enough? An automation built on a CRM nobody updates automates garbage faster. If the data is a mess, [fix the data first](/blog/your-crm-is-lying-to-you). The automation will work when you get there, and it will not work before.

Where to start

If you have never counted what your manual processes actually cost, [the hidden payroll of manual data entry](/blog/hidden-payroll-of-manual-data-entry) walks through that specific calculation. If you have counted and now have a shortlist, [what to automate first at a $10M company](/blog/what-to-automate-first) is the ordering I use. And if the process in question is quoting, invoicing, or collections, the days between "yes" and "paid" are usually the highest-yield place to point this arithmetic. [Quote-to-cash](/blog/quote-to-cash) covers where those days go.

If you would rather have someone calculate this against your actual processes than your estimates of them, that is exactly what an automation audit is. [Get in touch](/contact) and we will do the twenty-minute arithmetic together, with your numbers in the assumptions.