This Simple Arithmetic Can Optimize Your Main Business Operations

Deep down, we all know that we should embrace difficulties rather than dodge them. As John Adams put it, “Every problem is an opportunity in disguise.”

It follows that the greatest opportunities come from the greatest challenges, so let’s salivate over the most consequential business predicament of them all: optimizing our biggest endeavors, the largest-scale operations of each organization.

In their daily activities, companies face millions of “damned if you do, damned if you don’t” operational decisions, such as:

You can solve this. Here’s a big reveal that I believe precious few know. It takes only simple arithmetic to optimize any such decision: Calculate the downside of both options and go with the better choice.

As an example, take the first row in the table above: deciding whether to authorize a transaction or block it as potentially fraudulent.

First, calculate the downside of taking the action in the left column – the average amount you’d lose from being “damned if you do”:

The downside of the left column’s action = the chances you’re wrong × the potential cost

Example: If you are 95% confident that the payment is not fraudulent (i.e., valid) and we know that inconveniencing a legitimate credit cardholder costs you $100 on average – since they may stop using the card you’ve issued to them – then the downside of declining the purchase is 95% × $100, which comes to $95.

Second, calculate the downside of taking the action (or inaction, in some cases) in the right column – the average you’d lose from being “damned if you don’t” – which uses the same exact formula:

The downside of the right column’s action = the chances you’re wrong × the potential cost

Example: Since there’s a 95% chance the payment is valid, there’s a 5% chance that it’s fraudulent. If it’s a $600 purchase then, as a bank, you’d be liable for the full amount if it were fraudulent and you didn’t block it. So the downside is 5% × $600, which comes to $30.

Third, take the action with the smaller downside. Both options have a downside, but there’s a clear winner. We’re damned if we decline the transaction, expecting a $95 loss on average. But we’re less damned if we don’t, expecting a $30 loss. The math is telling us that, when we have 95% confidence in a $600 transaction’s validity, we should approve it.

The Tricky Part: Calculating Odds

Who’d have thunk? A simple bit of math based on the potential monetary loss from each misstep – aka the false positive cost (of an action in the left column of the table above) and the false negative cost (of an action in the right column) – and, bingo, millions of optimal decisions, each one based on its own straightforward cost-benefit analysis.

So, if the math is so easy, why isn’t it widespread, driving every operation, all the time? After all, many companies have deployed this approach successfully – but I wouldn’t say it’s nearly as widespread as it could be.

Here’s why. Coming up with the other piece of the formula – the odds that you’re right or wrong for each case – is the bottleneck. That’s the “magic ingredient.” You’re in great shape if you have it, but it takes some doing.

To be specific, what we need is a “probability calculator” that inputs the particulars about a case and outputs the chances things will go left or right – fraud or legit, cured or sicker, sale or no sale, etc.

Don’t fret – there’s already an established solution, a technology that learns from the outcomes of prior cases to calculate the odds for each new case. The technology is called machine learning. When used to drive these kinds of operational decisions, it’s called predictive AI or predictive analytics.

Surprise! It turns out that you’ve been reading an article about AI all along – but one that starts with AI’s value proposition instead of its brand. In an ideal world, predictive AI’s purpose and value for improving operations would be common knowledge. But in this world, where AI is branded in broad, abstract terms that declare first and foremost a silver-bullet quality – “intelligence” – rather than a concrete, value-driven purpose like that summarized above, it’s possible that you may not have realized this article was about AI until now.

The Final Challenge: Optimizing Total Value

Before you proceed with predictive AI, a word of warning: Without injecting a macroscopic view, your optimization project is liable to fail. Machine learning’s number crunching operates on the micro level, not the macro. It delivers the “probability calculator” needed to drive each individual decision. But you can’t launch a decision-making system without also gauging and tuning its overall performance. That step is usually missing – as a result, most predictive AI projects fail.

Enter ML valuation, an emerging practice that augments the typical predictive AI lifecycle with a reality-check for aggregate monetary value. ML valuation provides visibility into the total value a predictive AI system is expected to deliver across many decisions – in terms of business metrics like profit and savings.

By estimating the business value a system will deliver, it is possible to 1) navigate its development toward maximal value, 2) strike an informed balance where there must be tradeoffs (such as between the monetary bottom line and the number of inconvenienced
“damned if you do” cardholders) and 3) deliver an unshrouded monetary motivation for your decision makers to authorize the system’s deployment.

By focusing on concrete, operational value – rather than adopting traditional AI sales tactics that promise “intelligence” as a vague and often hyperbolic notion – you can take on your organization’s greatest operational challenges. As Unilever’s Global VP of Data & Analytics Morgan Vawter put it in the foreword to my book The AI Playbook, “Machine learning’s practical deployment represents the forefront of human progress: improving operations with science.”