Identifying AI Opportunities: A Hypothetical Case Study

Previously, we discussed how there is a large class of business problems that AI can address.  Here, we’re going to follow up that thought with a hypothetical case study. In particular, we’ll look at how AI could be part of a product for addressing elder care.

When we hear elder care and artificial intelligence some remarkable examples come to mind. We imagine robotic companions for the elderly. Smart home automation, which can help those with cognitive decline remain independent longer also come to mind.  Self driving cars are also poised to make a huge impact on the quality of life of the aging. These are all real examples of technologies that are ready now or will be soon.

They’re also all examples of unbelievably complex systems. Systems so complex that they need entire teams of researchers to produce. Some even represent ongoing avenues of research. These aren’t business opportunities that can be enhanced with or enabled by AI. These are research programs with commercial applications.  It’s important to recognize that not all AI opportunities are like this.  Some AI opportunities arise when we recognize how well studied AI problems relate to our own.

Looking for the AI Problems in In-Home Care

Consider the problem of in-home elder care.  People want to spend as many of their golden years in the homes they’re already comfortable in.  Unfortunately, certain realities of aging make that challenging.  In-home elder care seeks to fulfil that desire to stay in their own homes by providing a variety of services to people:

• Household maintenance
• Personal Care (grooming, etc.)
• Health care
• Transportation services
• Meal preparation

Noble intents aside, in-home elder care is straight forward from a business perspective.  Clients would like to receive services as either a one off occurrence or on a regular cadence.  The business needs to employ staff to provide those services to clients. There are lots of (incredibly important) things around the periphery of that core.  How does the business differentiate itself from competitors? How does it deal with the regulatory burden?

Service Call Scheduling Might Benefit from Automation

While this is all critical, it isn’t as exciting from an AI perspective as the core of the business: scheduling service calls.  Scheduling service calls fits the “Can AI Help?” pattern we established in the previous article:

• It takes a significant amount of human effort to schedule shifts generally, and here is no exception.
• That effort is repeated as we bring on new clients, as employees request time off or have to take sick days, and so on.
• The cost of scheduling is high.  It is someone’s (or several individuals’) job. Payroll is one of the largest expenses for most companies.
• Bad outcomes (suboptimal schedules mean less happy employees, not taking on a new client) are non-catastrophic
• Schedules have some objective measurements of quality
  • Time driving vs. time serving customers
  • Under-utilization of workers’ abilities
  • split shift count

Scheduling Service Calls Fits the AI Mold

If we could get AI to help offload some of the scheduling burden, the business could spend a larger portion of its payroll on service providers.  This would mean a reduction in overhead and the ability to handle more clients overall.  One of the key requirements for an AI-able task is that the problem can be rigidly defined. While we are working with a hypothetical, here is a believable rigid definition:

• We offer a fixed set of services (for example those listed above)
• We have a number of customers that
  • live places
  • want services
  • have times that are preferred and those that are not
• We have a number of employees / providers that
  • work out of locations
  • are qualified to provide various types of care
  • have constraints on when and how they can work

Our Problem Looks Similar to Formally Studied Domains

This business problem, distilled in this way, looks very familiar through the AI lens. Someone might compare it to a number of well known optimization problems in AI, including:

• The multiple traveling salesman problem
• Vehicle routing with time windows
• Job shop scheduling with material travel times
• Oversubscription planning

It might seem overwhelming that there are a large number of classic problems similar to this business case.  That’s actually a very, very good thing. Our use case is close to several text-book examples, each with decades of research. There’s likely a technique known to work well for something like our situation. This would greatly reduce development effort.

In our next article, we’ll look at where we go from here.  Once we’ve identified our business case as an AI problem, how do we go about building software to solve it. We’ll discuss how to tease apart small, useful increments so we can build our AI in stages, rather than all at once.