How to plan an effective AI Solution to a challenge

Artificial intelligence (AI) can be used to solve a wide range of problems in the healthcare industry, for patients and practitioners as well as for healthcare businesses.

At Eularis, for example, we’ve successfully leveraged AI and machine learning (ML) to improve business processes relating to customer engagement, precise physician targeting, customer segmentation, Next Best Action modelling and customer journey mapping, and pricing and market access, to name just a few.

There is no shortage of opportunities for the application of AI solutions. And yet, businesses consistently struggle to meet these opportunities when using such tools. An astonishing number of AI projects never make it off the ground, encounter disappointing and costly setbacks, or else fail to produce the kinds of results businesses hope for and expect.

Like so many things in business (and life), these difficulties often stem from a fundamental lack of planning and preparation. Artificial intelligence comprises a powerful set of tools, but understanding which of these to use and how best to implement them is key to its successful deployment. This article draws on insights from industry experts in AI and Eularis’ own experiences to provide a roadmap for planning an effective AI solution to any business challenge.

Where to begin: “a problem well-stated is a problem half-solved”

The titular quote above eloquently alludes to a problem many leaders encounter when bringing AI to bear on a problem. Namely, that the problem they hope to address is ill defined. And, perhaps more importantly, defined from the wrong perspective.

As a business term, “AI” is fantastically imprecise. It refers to a vast array of tools and solutions that, while sharing (at least in theory) some core principles, are diverse in nature and function. Having heard of the value of AI and hoping its implementation will reduce costs, increase efficiency, and boost productivity (which it indeed can), businesses try to “apply AI” to solve a problem or else go hunting for problems AI can solve. These are good ways to spend a lot of time and money going nowhere fast.

Instead, the first step in truly leveraging AI, is to identify a business problem—not an AI problem. And I would take that one step further and ask businesses to first define a customer problem. From here, it becomes much easier to clearly define the problem.

For example, rather than asking, “How can I use AI to boost sales?” (AI-centric and ill-defined), ask instead, “What would make it easier for the right practitioners to land on the right script at the right time?” This is now a customer-centric, business problem that AI is well-suited to handle: data can be used to define and find the right practitioners at the right times, provide them with the right drug for the right patient, just as or even before the patient needs it.

From problem statements to problem solutions

With a clear problem statement in hand, it’s time to consider solutions. These solutions needn’t be the sole realm of AI. Only some, perhaps even only one of the possible solutions, should involve the use of AI.

Both with AI and alternative solutions, try to engage in “lateral thinking”, using an indirect, creative approach—the kind that led Alfredo Moser to invent the water-bottle light—and “Design Thinking”, which involves “developing an understanding of the people for whom we’re designing the products or services” (and is thus decidedly customer-centric).

Buzzwords aside, in both cases, the goal is to come up with novel, creative solutions to your existing problem. This kind of thinking thrives in cross-departmental environments, where multiple different perspectives can be brought to bear on the same problem. Next, determine which of these solutions is best, according to, for example, how simple they are to implement; how likely they are to give good returns compared to the level of investment; and which are most likely to work well in 2, 5, or even 10 years.

Just as the problem statement should be customer-centric, so too should your solutions. Think carefully about each in light of the end customer and which serves them best. Take your best ideas and reimagine them from a customer perspective. Does your path from problem statement to problem solution provide the most benefit to the customer? If not, how can it be improved to do so?

Organizing your project in this fashion—with a customer-centric problem at the start and a customer-centric solution at the end—provides the framework for a successful roadmap.

In the land of the blind, data is king

Businesses operating in data-poor environments are blind. You cannot draw the roadmap from problem to solution without identifying, collecting, organizing, and analyzing the right data. This cannot be understated when implementing AI-centric solutions to business problems, because AI solutions—and especially ML solutions—are only as good as the data going in.

Data can take innumerable forms, although businesses will be most familiar with its more traditional formats, like spreadsheets and databases of text, audio, or images. There are a number of steps that need to be taken before this data can be used, but only a few key questions need to be answered in the planning stage:

• Data quality: Does the data seem to be of good quality? That is, is it both precise and accurate, a faithful and sufficiently detailed representation of the real world? Is there significant variation in its quality?
• Accessibility: Can the data be easily accessed and manipulated? Are there data protection (e.g. HIPAA, GDPR) issues that need to be addressed? What other challenges will make it difficult to use?
• Unique identifiers: Is there a simple and problem-specific manner of identifying data points? How easily can the data be organized, structured, and integrated?
• Relevance: How old is the data? How frequently is it being updated? Does it relate to an appropriate population for the project at hand?

Many of these questions can only be fully answered by consulting with a data scientist, but any preliminary efforts you can make will be invaluable for defining the shape of the path from problem to solution. Review your available data against the project and your objectives, identify any discrepancies or deviations, and come up with ways they might be remedied. Do this internally first to solve the most glaring problems, and then with the help of a data scientist for the more subtle and insidious ones.

How much tech do you really need?

We mentioned above that not all solutions to your business problem need necessarily employ AI. It may be that alternative solutions, like outsourcing, are more appropriate. At the next level up, your project may best be described as a “data science” project, one which leverages computing power but doesn’t require ongoing automation or intelligent decision-making.

But for projects that involve intelligent automation or decision-making (that is, do something that a well-trained person could do, but on a much, much larger and faster scale), with ongoing ingestion of fresh data, you will likely need a tech stack—that is, a combination of technologies used in the conception and function of your application or algorithm. A stack comprises all the technology, including all the programming languages, frameworks, databases, front-end interface, back-end tools, and external connections (e.g. via API) necessary for an AI project “to work.”

There are a huge variety of tools available to businesses depending on the volumes and types of data being processed (e.g., as a stream or in batch), and numerous software tools available for each part of the process. In addition, new tools are coming out of Silicon Valley and other tech centers, each with its own strengths and weaknesses.

Creating the right tech stack is arguably the most tricky part of bringing an AI project to fruition. There is no “one size fits all” solution when it comes to AI, which is why clear understanding of your problem, solution, and data is so essential. But it’s also important to understand how AI-based solutions fit together and which are best suited to your project, something that’s born out of years of experience and the reason why realizing a truly successful AI project usually means hiring or consulting with experts in the field.

Getting your project off the ground

Once your customer-centric, clearly-defined, problem-solution pathway has been established, including an understanding of your data and technological requirements, there’s just one step left: convincing people it will work.

Planning your ROI

“Show me the money,” say shareholders and executives. No company will invest in an AI project if the financial benefits are not made clear. Start by determining any initial (CapEx) and ongoing (OpEx) expenditures and a clear picture of cash flows from these investments. Establish a minimum return required by your company, and then use a statistical approach to validate your model. There are multiple ways to calculate ROI, including breakeven analysis, payback period, net present value (NPV), and internal rate of return (IRR).

Preparing your business case

A business case outlines the who, what, when, where, why, and especially how of your project. Clearly present the steps you’ve taken, as outlined above, including the business or customer problem to solve, the benefits of doing so, the risks involved in undertaking the project, your ROI calculations including initial and ongoing investments, the nature of your solution including the data (and any steps that will need to be taken to prepare it) and tech stack (with initial research demonstrating how these moving parts will fit together), the timescale, and any impact on ongoing operations.

Your AI business case should clearly demonstrate the problem, the solution, and your roadmap to getting there, including all the financial and operational steps along the way

.