On Automated Testing ROI

  September 28, 2005

There are so many variables that make up an automated testing project that it can seem insane to even talk about estimating what kind of savings/ROI to expect.  So, I'll be insane and predict that you could save $7,200 by automating just one significant test. I'll explain how, but first, a story about ROI:

One day, early in my programming career, I came to my manager with a great idea to automate one of the company's business processes. It was just the beginning of an idea, but I was so sure it would save the company money I wanted to tell him about it right away. He was just as excited about the idea, maybe more so. We spent lunch on the whiteboard brainstorming and I was ready to get started right away. He stopped me from running off, smiled and dropped the bomb on me, "This is going to be a huge improvement. How about you work up some numbers for me so I can get the VP to approve it?"

"Numbers? What kind of numbers?" I asked, afraid of what was coming next.

"Just whip up some estimates on the savings that automating will give the company, you know, ROI numbers, return on investment."

I went on the defensive about how it wasn't possible to estimate the savings, there were no numbers on how much the current process was costing, there were too many intangibles, too many variables, it would take longer to calculate the ROI than it would to automate the process…

"Relax," he said, "just guess."

Just guess? Guessing was heresy to me in those days. Weren't we using computers so we wouldn't have to guess? 

He politely ignored my objections.

"Ok, he said, how much will this automation save the company?"

"I don't know, I can't know, we don't have enough information!"

"Will we save a million dollars a year?"

"No, that's silly."

"Higher? 10 million? 20 million?"

"No, no…" Now he was just mocking me. "I don't know… uh, $200,000?"

He wrote down $200,000.

"And how long will it take us to automate the process? One year?" he asked.

I started to protest again, then I saw where this was going.

"Two weeks.", I answered. I thought I could do it in a week, so I gave myself a nice margin for error. He wrote down four weeks. We put down a few more numbers, then he turned the paper around to show me what we had created. He had taken my guesses and put together a simple ROI calculation. Seeing it on paper I noticed a few problems immediately and made some corrections. It still wasn't accurate, but this guesstimate ROI actually made some sense. In just a few minutes we had something we could stand behind to help justify the project to our VP.

My manager explained that my gut instinct was correct, it would take longer to try to accurately calculate the ROI than to complete the project, and the estimated ROI would never be accurate anyway. There were too many unknowns and too many variables. But when I just threw out some numbers, some wild guesses, the shape of the data would emerge and give me tools for thinking rationally about an irrational process. I was still guessing, but the guesses made sense. And they gave my manager and VP a good starting point to talk about approving the project. 

Ok, now, what about automated testing ROI? Let's throw out some guesses about that. What would be the ROI for automating one half-day long regression test by one tester?

Estimated tester salary = $20/hr

One 4 hour manual test executed 3 times a week = 12 hours @ $20/hr = $240 week.

This manual test uses about 600 hours a year of tester time @ $20/hr = $12,000 per year just to execute this single test.

If this test were automated we could guess that the tester would spend 1/3 of the original test time to run and maintain the new version. So, the tester would spend 4 hours a week with the new version instead of 12 hours. That's 8 hours saved per week, or 400 hours per year, for a rough savings of $8,000 per year.

The highest price of TestComplete Enterprise is $800. (It's less with concurrent licenses and quantity discounts.)  So, the first year savings for automating one significant manual test would be the Estimated Cost Savings minus the Cost of TestComplete:

$8,000 est. savings - $800 TestComplete

1 Tester / 1 Test ROI = $7,200/yr

Of course, these are just insane guesses, I've left out lots of important details, and your mileage definitely will vary. Remember that old project of mine I told you about? It ended up taking about 3 weeks to complete and it was a big success. The ROI estimate was a big help in getting the project approved, but after it was approved, we never mentioned the ROI again. Does that mean the time spent creating the estimate was wasted? Not at all, it never would have been approved without the estimate. Management needed to see that the costs and benefits of the project had been seriously considered. They needed an illustration of the possible benefits, not an accurate prediction of the future. The point is not to worry too much about your ROI estimate being perfect, it just needs to be useful and sensible to help you show management how much automated testing could help your team and your company.