Identifying opportunities for AI is really the intersection of three things: data, decisions, and opportunity. In most cases you start with the data. Many organizations think of data assets as whatever they have in their database or their spreadsheets, but data isn’t just your raw numbers.
The question you need to ask: what are you sitting on that you haven’t historically taken advantage of? It could be call center recordings, customer reviews on Google or Yelp, or mining social media. This is all unstructured data that you can turn to your advantage.
Next is decisions. What are some opportunities to enrich the data you have today with additional insights? Where are humans in your organization making impactful, frequent, and important decisions?
Last is opportunity. When I say opportunity, I mean the opportunity to create value. What is it that is consuming the effort of your team today? More importantly, what does that keep them from getting to?
One good rule of thumb is to start looking for cases where the humans in your organization are acting more like robots. When we are looking at the same patterns repeatedly, it becomes a monotonous task for humans, and it is an instance where AI can create value. Another good rule of thumb is if subject matter experts that are equally trained can agree and arrive at the same conclusion, chances are, you’re looking at a good opportunity to apply some AI.
And here’s some more concrete, tangible questions for you to start qualifying these opportunities:
These questions are your value proposition. They will help you get buy-in across your organization and help you focus on your higher valued work.
The real win with AI is that you have an incredible amount of untapped potential. As far as data goes, what are the unstructured data sets that you haven’t tapped into? Consider your process exhaust—logs, website analytics—things that produced by your existing processes and tools. What is created over the course of project management that we don’t necessarily think of as data sources, but are just sitting there waiting to be used to create insight?
Once you identify what have you not tapped into yet, what are some additional ways you can enrich your data? Whether you’re bringing in sentiment analysis using APIs or using some pre-trained models that can deal with audio and images, what are some opportunities to unlock your data for use in machine learning and AI?
Finally, think about the nature of the decisions you are making. Andrew Ng is a thought leader in AI, and he has a rule to help you think of tasks you can automate: “any task that generally takes you one second of thought or less to do is probably a good candidate for automation.”
Using this rule, we can look at an image and we identify the subject immediately. We can look at a comment and very quickly understand the tone. These are the types of human tasks that are well-suited to developing AI solutions. But remember, human agreement as a must.
To conclude, consider the two levers of value creation. How often do these decisions happen and what is the value created each time? AI is reusable and its value increases each time you identify an opportunity to use it.
Merilys Huhn in an Associate Data Scientist at Pandata.