What Is ‘Explainable AI’ And How Can It Help Your Business?

What Is ‘Explainable AI’ And How Can It Help Your Business?

By Birgit Starmanns, Global COE, Finance and Risk, SAP

When it comes to Enterprise AI (Artificial Intelligence), we often focus on automating repetitive business processes for a very simple reason and it doesn’t take much imagination to see the value. But what if you wanted to gauge the impact of an unexpected event, such as a hurricane, on your business’s bottom line? Maybe you’d like to compare the probable financial outcomes of a strategic decision before you make it? 

Explainable AI, which combines Human Intelligence with Artificial Intelligence, means employees now have the visibility to make these decisions. Let’s look at some challenges— and Explainable AI solutions— in the context of a real-world business scenario.

A Sample Scenario: Sales and Demand Planning

Divvy, Chicago’s bike share system, allows Chicagoans to go green and get a little exercise. But with over 500 stations, having bikes where they’re needed, when they’re needed, can be a significant challenge. Whether it’s a better customer experience, greater operational efficiency, or optimizing profit, more accurate predictions of volume can lead to a very real strategic advantage over the competition.

In Divvy’s case, having accurate projections on the number of rides originating from each station helps to ensure the bikes are available where and when they’re needed, and re-allocated when they are not, leading to a positive customer experience, increased repeat use and customer loyalty, greater growth and increased revenue.

Many factors can impact how many bikes are used on a given day, such as:

  1. Location
  2. Temperature
  3. Precipitation
  4. The day of the week and the week of the year

There are also factors that impact some stations more than others, for example a Cubs baseball game.

Artificial Intelligence has a proven track record for making reliable forecasts in such cases, but in today’s enterprises, accuracy is not enough – predictions have to have understandable. Otherwise one of two things will happen:

  1. Employees will reject the forecasts outright, because they don’t understand how the system arrived at them.
  2. Employees will blindly trust the forecasts without question, instead of using the results as a decision support tool for complex scenarios that require the human element.

To solve this problem, we need accuracy that can be understood.  We need Explainable AI.

Explainable AI

Explainable AI provides a whole new layer of insight by allowing analysts to clearly see why a prediction was made. Insights for Divvy Bikes may include:

  1. Average number of rides for a station
  2. Weather related factors (precipitation and temperature)
  3. Time-related factors
  4. Effect of a Cubs game

This additional information helps analysts make sense of how the underlying model works, which helps build confidence in its skill and precision.

Now let’s take this approach a step further. 

Adaptive Forecasting

Referring back to our Divvy Bike example,  to make future predictions, we need to provide assumptions about those same factors: the weather and the Cubs at-home schedule.

The Cubs schedule is easy to find, since games are scheduled well into the future. We can also obtain the weather data from any number of weather services, but as we all know, weather forecasts aren’t always accurate. The same is often the case with financial forecasts, and highlights the necessity of a what-if analysis.

In our scenario, after analyzing the Divvy demand schedule, we may notice that the weather assumptions are incorrect. With an interactive model, appropriate business drivers can be updated graphically, resulting in a visual depiction of the simulation results.

This same type of analysis can be performed whenever new information comes to light. This capability allows us to include highly predictive factors in our models even when they aren’t easy to predict themselves.

Adaptive Forecasting can be applied just about anywhere we need an accurate, explainable, and interactive forecast, such as:

  1. Predicting operating expenses based on high-level economic indicators
  2. Quantifying the impact of hard-to-predict events, like a natural disaster, on sales, revenue and cost
  3. Analyzing whether increased R&D or marketing spend would have a significant impact on net income in the next quarter, even the next year

As the technical complexity of creating these solutions decreases, power users and analysts will gain the ability to answer questions in a mathematically sound way that was never before possible. 

Not only will early adopters gain financial, operational, and strategic advantages over the competition in the short term, they will also be preparing their workforce to deliver greater value in the future.

JS Irick and Daniel Settanni of TruQua also contributed to this story.

Business News