KENNESAW, Ga. | Jan 14, 2025

Computing the ROI of any analytics, machine learning, or artificial intelligence (AI) process can be more complicated than expected … and so can getting stakeholders to buy into the computations! There are cases when computing ROI is straight forward, but there are also many cases when it is not. Further complicating things is the fact that analytics processes are often implemented for reasons outside of directly generating revenue or saving cost. Here, I’ll dive into some of the angles you can explore to justify the value of an analytics process, along with some of the factors that can make your computations difficult to pin down.

Cleanest: Revenue / Cost Impacts

We’ll start with the scenario where both computing an ROI, as well as getting buy-in for the computation, is straight forward and easy.  These are cases where a new analytics process is being added onto an existing business process and is the only change being made. Examples might include adding dynamic pricing models behind a website, adding predictive maintenance models into a maintenance system, or similar. The goal is a straightforward increase in revenue or decrease in costs.

In these controlled situations, whatever revenue increases, or cost decreases, occur are easily tied to the analytics process implemented and easily computed. There will be little dispute or debate about the computation, or about the ROI figures the computation generates. If only we were lucky enough to have these situations be the rule rather than the exception! 

Muddying The Cleanest Case

Unfortunately, the real world isn’t always so clean. For example, there may be an entirely new business process or product being rolled out with analytics as one component of it. Perhaps a company is totally overhauling its outdated marketing systems and approaches. A new platform is implemented alongside new marketing strategies, content, and creative. Various analytics processes are also developed to support those investments.

In such a situation, it might be possible to compute the ROI of the new system in totality, but very difficult to parse out which of the new components – platforms, strategies, content, or analytics – drove what portion of that ROI. You can expect much arguing and a heavy dose of subjectivity to enter the debate as each stakeholder (naturally!) argues that their portion of the work should be credited with an outsize percentage of the total.

There is no easy way to fix this dilemma. The statistically cleanest is to turn on and off various parts of the new system via A/B testing and measure what happens. However, with so many pieces and parts, this can be hard or even impossible from a practical perspective.

Fuzzier: Risk Mitigation

Risk mitigation is a different angle to use to justify an investment. In this case, we aren’t lowering existing costs but are instead avoiding potential future costs. Situations involving regulatory, transparency, or legal requirements can necessitate analytics processes to be built. For example, a process to validate that a machine learning model is fair and unbiased can help defend against a lawsuit or even avoid the lawsuit being filed in the first place. Of course, it is impossible to assign an exact monetary, reputational, and opportunity cost to the lawsuits that would have occurred without the process and, thus, very difficult to get hard figures on the return.

However, while a fuzzier approach, risk mitigation can be a terrific path to justify an investment. The reason is that the potential risks are often quite large and cause executives to become fearful of being the one in charge if they hit. The cost of the new analytics process can be dwarfed by the worst-case (or even moderate) scenarios that can occur without it. Thus, it can be easy to get buy-in as long as you can make the case that risks are large … even though you can’t pinpoint them exactly.

Fuzziest: Strategic Imperative

This is a less common scenario, but one I have seen used multiple times. I recall the CFO of a tens-of-billions of dollars per year health system talk about his approach to justifying a major data platform upgrade along with a new set of analytical tools and people to sit on top of it. He knew that the company’s systems were grossly outdated and that they were unable to build many of the analytical processes that had become standard in their industry. He pushed through the upgrades based on strategic imperative.

He was challenged by his team, who pointed out that they were always held to much harder ROI commitments and he wasn’t following his own rules. He acknowledged their concerns and the fact that while he had some high-level estimates, they certainly weren’t as firm as typically required. His view was that the organization could have confidence that between added capabilities and risk mitigation, the investment would pay off as it had for other companies. He made the case that NOT modernizing was too big of a risk to squash laying out the cash to modernize. A strategic imperative, perhaps, but still fuzzy ROI!

Use A Combined Approach

In practice, a given proposal can often be tied to multiple of the above ROI computation approaches. A process might generate some revenue (even if indirectly), mitigate a current risk, and have a strategic value. It might be that the process can’t be justified on only one of those ROI components alone, but that when considering them together the argument for investment can be won.

The best way to avoid massive arguments over exactly how to compute an ROI is to have enough different returns documented that people can believe a strong ROI is certain, though the exact magnitude might be unclear. Then, get buy-in up front on how to compute the post implementation ROI and document it well.

In the end, if the computations come back positive as expected, people will accept it. It is critical, however, to acknowledge the fuzziness of some of the computations up front and include as many components in the ROI projections as possible. Between revenue generation, cost take out, risk mitigation, and strategic imperative, you must find a way to make your case.