ROI of AI (Sold With AI - Edition 11)

Doubts about the ROI of AI are beginning to pop up. However, it is critical to measure it, especially in the current environment. This post suggests how.

AI is everywhere today. Operations or HR; big company or small; a maker of welding tables in the midwest or vector databases in Palo Alto - everyone is exploring AI.

But how do you decide which AI is worth leaning harder into and which one to cut out? How do you ensure that you are not just “buying emotionally and justifying rationally”, like most people do?

How do you measure the ROI of AI?

This is a critical but hard question that doesn't always have great answers. At least not yet.

When it comes to Sales, only two metrics truly matter - Revenue, and Revenue Efficiency. Yes, there are a few more metrics that are useful for effective revenue management, but nothing really comes close to Revenue and Revenue Efficiency as beacons of truth.

We all understand what Revenue means. However, what is Revenue Efficiency? Simply put, it is the indicator that tells us if we are acquiring the revenue in a cost-effective way. At any point, a company should want to have as much more revenue as possible, at as much low a cost as possible.

Are these metrics peculiar to AI? Not really. These metrics could very well apply to acquisition of non-AI tools, and my advice would still be to measure if those tools move these two metrics up north.

However, these metrics become especially important when it comes to AI. It is important that one understand the reasoning for the same.

Before AI tools (aka ‘Intelligence’ tools), there were mainly two other classes of Sales tools - workflow tools (like Salesforce) and data tools (like Zoominfo). I have written about these classes of tools earlier). 

Over time, both these classes of tools have become necessary. And when a tool becomes necessary, the need to justify its ROI doesn’t remain very critical. The only question that gets asked is whether one needs such a tool; and if yes, then which tool should one go for (Salesforce or Hubspot; Salesloft or Outreach; Zoominfo or Apollo). The need to prove its ROI is not really critical.

However, AI tools, at least for the next 2-3 years, are not going to become necessary tools. They are going to be ‘nice to have’ - in the sense that one could do without them. (Whether one should is a different question.)

In that sense, AI tools face a harder test. They need to prove their ROI, more than their ‘workflow’ and ‘data’ counterparts have ever needed to prove. 

Which brings us back to the original question - how should one measure the ROI of AI? Should we always try to measure the impact on revenue or revenue efficiency? What if a tool’s promise is only in terms of impacting the pipeline, or in terms of saving reps time? How do we justify purchase of such tools if we cannot measure their ROI?

The answer lies in establishing ‘deductive’ metrics. These are the metrics that eventually lead to revenue or revenue efficiency, even if they are something different themselves. As long as one can establish the relationship between these deductive metrics and the core metrics (Revenue and Revenue Efficiency), it should be possible to use these metrics for measurement of ROI.

Let us take an example of a tool ‘A’ that holds the promise to increase outbound pipeline significantly. Just because we cannot measure its impact on revenue, should it mean that we should not consider this tool? That would be quite foolish, assuming the tool can live up to its promise.

The key is in ensuring you know what your outbound pipeline’s contribution to revenue, and conversion, is. If you know this, you can now easily compute the impact of tool A on revenue. Even when all that you could directly measure was the change in pipeline because of tool A. 

For example, if tool A increases outbound pipeline by 40% and outbound pipeline is 30% of your overall pipeline, it means that tool A will increase your revenue by 12%. If tool B increased your email response rate by 100%, then as long as you can compute how much the increase in pipeline for each unit of change in email response rate is, you will be able to compute the ultimate impact on revenue. 

Otherwise, you run the risk of getting stuck with a vanity metric, or worse, a misleading metric.

Apart from the need to prove its value, there are two more reasons why measurement of AI’s ROI impact is crucial.

Not many people fully comprehend the fact that AI is a non-linear change. Which means that the impact of ‘good’ AI can be exponential. And unless you measure such an impact, odds are that you will underestimate it. 

I have seen that firsthand at Humantic AI. One of our customers, a Nasdaq listed SaaS company, saw positive but somewhat mixed feedback from its sellers when they polled them about the Humantic AI product. However, to get a quantifiable measure of the impact, they decided to analyze the change in win rates on deals where Humantic AI had been used (vs. those where it had not been used).

They were surprised to find out that the difference between deals where Humantic AI had been used vs. not used was 37%. Which means that deals where Humantic AI had been used had a 37% higher win rate!

For a company that is doing 00s of millions of USD in revenue and adding $100 million every year, that totals up to potentially $37M in additional revenue from using just one impactful AI product. It is hard to imagine how much the impact can be from using 4-5 AI tools effectively and one might simply dismiss it as ‘too good to be true’. But as I have written earlier, this is often a mistake given how capable ‘good AI’ can be.

And finally, there is a not so good reason to measure the ROI of AI. The reality is that while good AI tools can have exponential impact, there are 00s of AI tools in the market today that do not necessarily offer significant impact. However, it is very hard for the buyers to differentiate which is which from the look of it. Hence the ROI case becomes vital.

I also think that the onus is on the vendors to not just offer proof of ROI, but to start offering measurement of ROI in the tool itself. It is one of our major initiatives at Humantic AI this year and we hope that it gives our customers the confidence as well as the peace of mind when they are making a commitment. 

In summary, it can be said that measuring the ROI of AI tools is important for multiple reasons. And more so than earlier, non-AI tools. While the best measure is the impact on revenue or revenue efficiency, it is not always possible to do that. In such cases, one should aim to measure a metric that can be tied to revenue directly or indirectly. More direct the association is to revenue, the better it is.