Practical Applications of AI in Business
Last week I had the honor of hosting a number of executives through a discussion of Practical Applications of AI in Business in Evansville, Indiana—right in the industrial heartland of America. What a wonderful forum it was to discuss the business challenges of the day and the places where AI might be helpful to their mission of creating value. Most of the participants came from middle market companies, with no time or excess budget to waste on science projects. These folks were serious and deliberate about making their companies better, faster, and cheaper. They wanted no-nonsense advice about how AI could help them…now.
One principle I stressed over and over to that group was the need to apply AI in small, surgical doses. I favor that approach over the big, sweeping enterprise-level, grandiose AI projects that I see taking place in some companies.
AI-Enabled Agents
Most often these small doses of AI take the form of agents. Agents are essentially software objects that carry out a defined task, many times in the background so that they are never seen. Like an agent that senses that a longstanding customer is trending downward in their order volume, or an agent that predicts a sales increase in a certain region, or another that analyzes inventory and suggests that manufacturing is not on pace to replenish it. To me, these are the silent heroes of the AI revolution, and where AI will have the greatest impact on company value.
Agents live among large enterprise software applications like CRM, SCM, ERP, MES, HRM and so on like ants among the elephants, sprinkled throughout the enterprise to accomplish small, defined, usually fully autonomous tasks. A diagram of this is shown below.
AI as it stands today is not a product that you can buy off-the-shelf in spite of some vendor claims to the contrary. You have to build them. That is not as scary as it sounds. Let’s talk about how to do that.
AI-enabled agents need three ingredients:
- A triggering mechanism
- Business logic
- Access to data
Database Triggers
Agents are typically triggered by a time interval or by a condition in the business. Time is straightforward—every day, week, month, or quarter you want some agent to wake up and do its business. Changes in the condition of the business can be a bit more complicated, but still just as feasible to implement. The best way to trigger an agent on a business condition is through its data, sitting in a corporate database. Database triggers are a feature of nearly every modern database system these days and can be much more sophisticated than triggering off of the change in one database field, allowing multiple database fields to comprise an “if this and this and that and another thing then activate the trigger” mechanism.
Once the trigger is activated, a predetermined script runs. This in turn can be linked to a computer program. Our favorite language for creating business logic these days is Python, as it is both an elegant language for constructing any arbitrarily complex logic, and it comes with a wealth of open source libraries for accomplishing an astonishing variety of common computing tasks. It is our practice to create very detailed diagrams of the business logic, vetted by company experts, before committing that logic to the programming process.
As the business logic runs, it needs data to accomplish its computing. Agents are attached to databases either directly or through Application Programming Interfaces (APIs) that are a communications channel to the application that holds the data the agent needs. Most good, modern enterprise applications publish the specifications for their APIs so that external programs (agents in our case) can easily access data from the application.
Now to be sure there are security issues associated with a bunch of agents scattered across an enterprise doing various intelligent tasks. It is an important consideration, but it should not stop you from building agents in the first place. It is a bit like teaching your teenager how to drive a car. A car is dangerous, and risky in the hands of a teenager. The teenager needs good training and sensible thinking skills to mitigate the risks. But the importance of having a teenager learn to drive—the capability itself—is too important to ever consider the option of never driving.
While we have talked a lot about technology here, the real trick of getting AI to work adding value in the company is in the design of the business logic. I cannot stress enough the importance of pulling out a sheet of paper and making a good, solid diagram of the business logic and sharing that diagram with others in the company until you get a consensus on the design. The technology is fairly straightforward once the design is finalized.
So there you have it—agents. Small fragments of intelligence that propel our companies forward in countless ways every day. I encourage you to consider agents in your mix of analytical capabilities.
As always, I wish you the best of luck on your AI journey.
About George Danner
- George uses the very latest scientific techniques and methods to improve the performance of mid-size and large organizations through problem-solving, the optimization of existing practices, and advanced forecasting
- 37 years of experience in corporate strategy, specifically operational and financial analysis, across a wide variation of industries
- Author of the best-selling books, “Profit from Science” and “The Executive’s How-To Guide to Automation”
- M.S. Massachusetts Institute of Technology
