The most common experience people have today with “artificial intelligence” is through spoken interactions with Apple’s Siri, Google’s Assistant, or Amazon’s Alexa. These are really “intelligent assistants” rather than true AI, but the way they make the interaction tangible is a good way for us to start this discussion about the technology that underlies LevaData’s full cognitive AI capabilities.
An underlying philosophy to LevaData’s approach to User Interface / User Experience is to minimize the cognitive load on users. This is particularly important in Sourcing and Procurement because professionals deal with complex decisions that are influenced by hundreds, if not thousands, of data points. If we can help make processing that complexity easier, then we’ve added real value to our customers. But, why is an Intelligent Assistant interface to our underlying AI important?
An Old Standby: Spreadsheets
Let’s start by understanding how most Sourcing teams interact with the complex data sets that they need to do their jobs. Our 2018 Cognitive Sourcing Survey (see the infographic summary here) revealed that 61% of teams still use spreadsheets to create reports to make decisions. Microsoft Excel is their primary “user interface.”
These spreadsheets often have years of history to them. Teams feel like they understand how to build the reports. They know where to look for data, and the similarity from year to year provides a feeling of continuity. Reports also give a sense of thoroughness to an analysis, with column after column, and row after row of information to scroll through.
Processing these reports generates a high cognitive load for users. It’s hard to see patterns at a glance when looking at information in this format. On top of that, reports are for the most part historical. The data is lagging information, often from the prior quarter or before. Up to the minute market data is hard, if not impossible to integrate. To use reports to project future changes, users must manually build projections with the limited and unwieldy statistical tools and formulas in excel.
In summary, reports are labor intensive, built on data that is often incomplete and out of date, and the outputs can be difficult to understand.
What You See is What You Get: Visualization
Visualizations are often the next level of user interface with sourcing data that teams employ to reduce the cognitive load in reviewing the reports. Charts and graphs can show trends over time. They can also be used to project the information forward, by showing a range of possible ways a graph line might continue.
However, while visualizations help you focus on important trends, they do so by simplifying complex data into a more limited representation. You may be able to compare a couple of factors against each other in one chart. Add more variables, for example, more sub-commodities that might affect your product’s cost, and you quickly run into hard to read graphs. Once again, the cognitive load on the user goes up.
Both reports and visualizations suffer from the quality and age of the data in the spreadsheets. They also introduce risks related to cognitive biases. We’ll be doing a deep dive into what cognitive biases mean in other articles, but in summary, these biases occur in part because humans are pattern-seeking by nature. As a simple example, if we look at a graph of prices that seem to be going up consistently quarter-after-quarter, we naturally project the line upward expecting prices to increase in the same way. Experienced sourcing professionals may know that’s a fundamental mistake, so when presenting a visualization, we add verbal caveats to call this risk out or draw a worst-case line that doesn’t continue the trend. That again adds to the cognitive load on the viewer and introduces debate about what to believe.
Even if the above could be addressed through better visualization tools, the lagging information in both reports and visualizations means that you may walk into a negotiation with a playbook that is out of date. From our survey, 39% of leading organizations have migrated from simple reports to advanced analytics and purpose-built sourcing platforms. A key feature of these platforms is that they integrate more data sources from both within and outside of the organizations, often in close to real time. This allows for adding a key User Interface element: Alerts.
Information Just-in-Time: Alerts
Alerts warn you of changes as they happen or soon after. These notifications allow teams to react in a shorter time frame to changes in the market. If you know that the price of a raw material that a supplier uses has gone down, you may have more leverage in an upcoming negotiation. This up to date information can become a competitive advantage and streamline the negotiation process.
However, while alerts may tell you that something is happening, it’s still on the user to try to understand what that might mean and what the best action to take might be. This still requires a significant cognitive load and often further manual analysis and modeling. The user has to factor in the relationship these changes have to other variables, be it other suppliers, sub-commodities, or parts. So, while alerts may help you identify risks and opportunities, they don’t tell you what to do. Furthermore, alerts can become overwhelming. Think about the constant notifications you get on your smartphone as new emails, news alerts, tweets, messages, and updates arrive. It becomes distracting at best and draws your attention away from what is important at worst.
This is where Intelligent Assistants and AI Recommendations become a real advantage to the sourcing professional. By combining analysis informed by deep AI capabilities with an Intelligent Advisor User Experience on the front end, sourcing teams can not only be alerted to a new risk or opportunity but also be given guidance on what to do as they arise
For example, say that the price of an input to a supplier has dropped by 10%. This is data that comes from an external source outside your enterprise data. No matter; a sourcing AI has access to many data streams. The AI would use a number of methods, in this case maybe a Monte Carlo simulation to determine your likelihood of sayings on an RFQ for a variety of price points. The Intelligent Advisor might tell you as you are about to head to the meeting that you have an 80% chance of saving 7% but only a 5% chance of saving the full 10%. You’ll know how hard to push the negotiation without alienating your supplier; after all, good relationships are a key to long-term success. Because the advisor is always on, if another market change happens while you’re in the meeting, you’ll be advised what to do about it then and there. True AI recommendations reduce the cognitive load on the team to the point where they can focus on the highest value add actions; in this case, effectively negotiating with a supplier while maintaining the best possible relationship.
On the Road: An Illustrative Example
To summarize, it might be useful to think of the evolution of the User Interface from reports, to visualizations, to alerts, and finally to AI recommendations using the history of navigation technology in your car.
In the pre-internet days, directions for a road trip came in verbal or written form. This might be along the lines of: drive 120 miles on I-35, then take the exit for Round Rock. Keep on the right for about twenty miles; if you want a bite to eat, there’s a great diner with a red sign about 5 miles from the exit. When you come to a fork in the road, head towards the three large towers. If you look to the left, you’ll have a great view of the river. Turn left on Main street, then find parking around the fountain, it’s the cheapest. This is similar to a report, with a lot of detail describing places to stop and landmarks that are easy to see.
Of course, most of us would take these written directions and find the road on a paper map, a basic visualization. We might highlight the route with a marker and maybe circle roughly where that diner might be. Perhaps we’d lose the landmark information and the note to look for the beautiful river, but that might not be useful information anyway. However, a map is hard to use while driving and may require you to pull over, so hopefully you have a navigator next to you to help.
Early GPS devices took the visualizations to the next level and added alerts. They’d notify you when you need to make the turns, so you don’t miss one. And if you did it would reroute you, saving you from fumbling with the map to find your way back. You could even add the diner as a stop so the system could alert you just before you pass it. If you updated your device maps recently, you could find more efficient routes and perhaps be alerted to planned road closures.
Connected GPS apps on your phone, such as Waze, have taken things to the next level, bringing real AI recommendations to your drive. Not only do they provide the turn-by-turn alerts of basic GPS devices, but live data, including the real time speed and position of other drivers, information on obstructions, and traffic. Waze doesn’t just tell you that traffic is building up or a road ahead is flooded. It automatically suggests alternative routes to save you time, giving you a range of options from which to make a choice. You’re still in the driver’s seat, but you don’t need to pull over and decide how to route around obstacles. You can stick to the route you’re on if you still want to stop at the diner or take the app’s advice and adjust to the shortest course to the destination.
Just like Waze, AI Recommendations provide a User Experience that minimizes the cognitive load on the user. In the case of a driver, this can be a safety improvement, reducing the distractions of navigation, or uncertainty about the correct right exit to take. In the case of Sourcing professionals, it can mean better negotiation outcomes, faster identification of new opportunities, and the ability to focus on strategic tasks, as the system handles tactical decisions.
As mentioned, while AI Recommendations are the most visible part of a Cognitive Sourcing platform, to be effective they require a full suite of underlying predictive analysis and deep AI components. The following articles in this series will discuss the various approaches LevaData employs to make Leva, our Intelligent Advisor, that much more intelligent and useful to you and your procurement team.