4 Practical Tips to Create Value with AI
Setting realistic expectations and taking lessons from previous tech trends can help you plan for successful AI programs.
- By Cal Al-Dhubaib
- September 26, 2024
AI talent is scarce; builders and highly skilled users are in low supply. Data science was once the same. In the mid-2010s, my company Pandata (acquired by Further in 2024) helped solve the data science talent pipeline shortage in Northeastern Ohio by building a diverse, talented team of data scientists equipped to work on problems for major organizations.
Based on what I learned while data science talent was scarce, here are my recommendations that can help you create successful AI programs today.
Suggestion #1: Make AI Boring
Whenever we talk about AI in popular culture, we think of the Terminator or fancy flying cars. Even with widespread use of ChatGPT, it’s still somewhat a novelty that has a mystique. This perception gets in the way of using AI productively.
When we make AI boring, we eliminate the hype and stop treating it like a curiosity.
Think about driving a car, for example. You probably drove to work or the grocery store in the last few days. Modern-day engines are marvels of engineering, and yet, we just get in the car, turn on the ignition, and go.
As complex as these engines are, automotive giants have reduced quality assurance to a science. There are NIST standards and checklists in place, and a part either meets those standards or it doesn't. And more importantly, we know the exact number of tools and staff needed to execute these standards.
All of this makes safe scalability possible, and it is what we're striving towards with AI.
Suggestion #2: Accurately Measure the ROI of AI
There are two sides to every ROI equation:
- How you measure value (dollars)
- Costs (dollars)
Even though everybody in business understands this equation, when either side is poorly defined, it becomes really challenging to measure ROI—especially for AI projects.
The two most common ROI calculation mistakes are inflating your expectations of the value generated and undercounting potential costs. You don’t want to end projects prematurely because you didn’t correctly measure value.
Although time-savings is a popular choice, it’s a very limited view into the value AI brings to the table. To avoid these pitfalls, consider all the different ways that value can be generated from AI (many of these should mirror how you measure the value of your workforce today).
- Volume of outputs
- More satisfied customers
- Higher basket sizes or spend
- Improved output quality, even if it takes a little bit more time
- Shorter time to results
- Enhanced work-life balance
Plus, keep in mind the true cost of building and implementing a model:
- The cost of strategizing, planning, and building the model
- The time it takes to build, test, and run the model
- Ongoing maintenance costs
- Human training and hiring to accommodate new or improved workflows
Suggestion #3: Understand Your Problem and How Decisions Are Made Today
I recommend AI projects always start with a discovery and design phase. One of the most dangerous trends I’ve seen is the rush to use AI to solve problems without a pre-existing human process in place.
For instance, consider a scenario where an organization wants to use AI to qualify patients for a program. If there isn’t already a human process for this, why hurry to implement AI?
We must take a step back and ask: If a human were to handle this task, how would it be managed? What would success look like? What does solving the problem look like? The discovery and design phase allows for an in-depth exploration of decision-making processes.
Once we understand how decisions would be made, evaluated, or influenced by humans, data scientists can more effectively determine:
- What types of models are possible
- Potential limitations
- The accuracy we might achieve
This foundation helps us align AI solutions with decision-making needs. We consider factors like whether the task allows for any failure or if the AI must provide comprehensive information for a human to make an informed decision.
Suggestion #4: Recognize That Culture Is an AI Business Decision
I've always seen culture as a business decision, and organizations that have found success with AI do as well.
AI so often fails because we’re surprised by unintended consequences.
Most people aren’t trying to create malicious uses of AI. Most of the AI-related issues that we hear about in the news are the result of a system breaking in a way we didn't expect it to break because patterns didn't match reality or there was some unintentional negative bias that was reflected in the data set.
Having individuals on your team that feel comfortable voicing those types of concerns and building an open culture where this dialogue is welcomed is the secret sauce behind building robust, safe, responsible AI.
Final Thoughts
While there is no secret formula to perfecting AI (let’s face it: if there was, we’d see way fewer mistakes in headlines), these strategies and ways of thinking have consistently led to greater chances of success for organizations.
Treating AI projects with flexibility and setting realistic expectations for all involved will go a long way.
Editor’s Note: Members of the Further team will be presenting more insights at TDWI Transform Orlando, running October 20-25, 2024.