How to know when to call in The Machines: Making a Use Case for AI/ML
By Sara Beck
Advances in computational power, along with an unprecedented ability to store vast amounts of data, have resulted in staggering breakthroughs in the data sciences. Algorithms and models can be applied in ways no one thought possible – and with incredible accuracy. All this impressive progress has allowed organizations of nearly any size and any industry to leverage the power of Machine Learning and Artificial Intelligence in both their products and decision making. (Pssst- better decision making and smart product capabilities ultimately increase the bottom line.)
So, we've got the tech, but that's only half the battle. Is your team ready to leverage all these advances? With the all the buzz-cronyms around Machine Learning and Artificial Intelligence, does your team truly, genuinely understand them? If not, your organization is not fully prepared to benefit from the latest techniques.
If you think your team might not be ready – worry not. Slalom Build's ML practice is here to help! And as with most things we do, we can break this down together.
What’s your use case?
Sounds simple, but what are you hoping to accomplish? How do you determine if you need to leverage machine learning or data engineering techniques to realize your goals? See, Machine Learning and other data science models all have two things in common: uncertainty and variability. It helps to understand where your problem sits in relation to both. We need a bit of uncertainty in order to have something meaningful for ML to predict. And variability provides allows models and algorithms to leverage different information from different attributes to make better, more specific predictions.
For example, predicting next year’s widget sales numbers is a very different problem then auto-correcting text messages, even though they’re both Machine Learning problems in that they both feature uncertainty and variability. There’s definite uncertainty in predicting sales numbers because no one really knows the future. (Zoltar excluded, of course) What’s more, sales tend to be affected by many factors such as availability of item, public perception, price, etc. This is a great demonstration of variability.
Consider everyone’s favorite messaging scapegoat: Autocorrect. Until a word is fully typed there is a great deal of uncertainty in what a user was intending to type. (For some of us, a great GREAT deal of uncertainty.) There Is also variability in the sense that many words contain similar character strings, and that sentences often begin with the same words.
So, can you identify the uncertainty and variability in your ask? Without both, your use case is more likely a business intelligence or data engineering problem.
Now…how's your data?
Many software products require – or at least greatly benefit from – operational Machine Learning models from day one. Often before it's possible to collect any data at all!
Fortunately, lack of data does not need to be a deal-breaker for leveraging ML/AI techniques – depending, of course, on the problem being solved. Many Machine Learning implementations allow for continuous learning over time as new data is collected, which can even help make better predictions going forward. Spam filters, recommendation systems, and fraud models are just a few everyday examples of ML integrations with this effect. There are also entire classes of models devoted to estimating what cannot be directly measured! Some use cases may benefit from an expert to identify, select or modify open-source models that can be used from the get-go.
To obtain good results, other use cases might require a longer data history and a sufficient number of related attributes. When data for an important attribute is missing, it may be possible to use other variables as a proxy or perhaps separately model that attribute to be included in other models or analyses. Our Machine Learning Practice has experience training and applying models to many types of data from text, click-stream, and even video. (We are premier partners with AWS, GCP, and Azure and are confident in our ability to handle big and small data problems.)
No use case? No problem.
But hey, maybe the data sciences are a little out of your wheelhouse. Maybe you’re an expert in another line of business, so it might be difficult to identify how ML or AI could help your business.
One way to dive into this field is to start thinking about some of your general business problems or customer pain-points. How would you solve those problems? What would it look like if you had a solution? Often this line of thinking can reveal areas of uncertainty or variability – ingredients for a potentially viable Machine Learning solution!
Next, inquire if any related internal data sources are available within your company. Often, identifying internal sources and an area of uncertainty is enough to formulate a Machine Learning plan. All you need to start is to begin thinking of what a solution or design feature would look like!
If you’re still stuck, that's okay! Luckily, you have Slalom Build to turn to. We’re experts in this field. We focus on cutting-edge Machine Learning and Artificial Intelligence techniques all day every day to better enable you to focus on your own expertise. We will collaborate to understand what success looks like for your organization, explore what data you may have available, make suggestions for how to integrate AI/ML in an impactful way and then follow through with the build and integration of these models. It’s what we do.
This is an incredibly exciting and innovative time for Machine Learning and Artificial Intelligence advancement. Almost every organization is ready to start leveraging ML or AI in some form – and Slalom Build's Machine Learning Practice would love the opportunity to share our passion for solving the most challenging problems, while helping you unlock and build your possibilities.