AI and machine learning are going to start making a lot more decisions. They probably still won’t be used in the near future to make “big” decisions like whether to put a 25 percent tariff on a commodity and start a trade war with a partner.
However, nearly anything you’ve stuck in Excel and massaged, coded, or sorted is a good clustering, classification, or learning-to-rank problem. Anything that is a set of values that can be predicted is a good machine learning problem. Anything that is a pattern or shape or object that you just go through and ”look for” is a good deep learning problem.
And business is full of these. Just like the world processor replaced the typewriter pool, AI will soon replace hordes of office workers staring at Excel—and replace some analysts too.
Companies need to be preparing for this change. Just as companies that didn’t prepare for the web and e-commerce got left in the dust, so will companies that don’t adapt to AI and machine learning. If you’re not looking at the vast amounts of data you process and decisions you make and asking, “Can’t I go the last mile in automating this?” or looking for things you don’t do because you can’t decide “in real time” enough to gain an advantage—I’ll be seeing your company’s closure in the papers in a few years.
To prepare for this change, you have five prerequisites before you can even begin a business transformation. You need a strategy to spread AI throughout your organization that starts with these five prerequisites.
AI prerequisite No. 1: Education
You can’t make everyone in your company a data scientist. Moreover, some of the math is running too fast for us mere mortals to grasp—the specific algorithm people think is most efficient this week isn’t likely to be the right one next week.
However, some basic things aren’t going to change. Everyone in your organization should understand some basic capabilities of machine learning especially developers:
- Clustering: Grouping things together.
- Classification: Sorting things into labeled groups.
- Prediction on a line: If you can make a line graph, you can probably predict what that value will be.
- Prediction of variance: Whether it is liquidity risk or vibrations or power spikes, if you have a set of values that fall in a range, you can predict what your variance is on a given day.
- Sorting/ordering/prioritizing: I’m not talking about the simple stuff. Whether it is for search or prioritizing which call your sales or support person takes next, this is something that can be handled.
- Pattern recognition: Whether it is a shape, a sound, or a set of values ranges or events, computers can learn to find it.
One key thing is to have a set of people around who can dumb it down for people based on their skill level. Your developers might be interested in specific algorithms or techniques, but your analysts and executives should understand the basic business problems and computer techniques. Your executives many not need to know how clustering works, but they do need to recognize that a problem “looks like” a clustering problem.
Finally, you need a regular education refresh, at least yearly, because the capabilities are expanding.
AI prerequisite No. 2: Componentization
Some of the recent tools around componentization are “notebooks” for data scientists; a lot of the other tools grow out of these. These are great tools for data scientists and their collaborators.
The problem is that they encourage bad practices when it comes to production. The interface to a classification algorithm looks roughly the same as all the other algorithms. A particular classification algorithm implementation doesn’t change with the business problem.
Just like many companies had to figure out how to make one representation of a customer (rather than totally different ones in each system for each business problem), you need to do the same for algorithms. This isn’t to say you come up with the one true clustering algorithm, but that you componentize what is different.
AI prerequisite No. 3: Systemization
Despite all the hoopla, most systems still look the same. There is some process for getting the data into an algorithm, some process for executing the algorithm, and a place to spit the result out. If you’re custom designing all these things over and over for each algorithm, you’re wasting time and money—and creating a bigger problem for yourself. Just like SOA changed how many companies deploy application software, similar techniques are needed in how AI is deployed.
You don’t need a bunch of custom Spark clusters running around with custom “notebooks” everywhere and custom-built ETL processes. You do need AI systems that can do the heavy lifting regardless of the business problem.
AI prerequisite No. 4: AI/UI componentization
AI prerequisite No. 5: Instrumentation
None of this works without data. Let’s not go back to creating big, fat data dumps where we just collect a bunch of garbage on HDFS and hope it has value someday, as some vendors have urged you to do. Instead, let’s look at what things should be instrumented.
If you’re in manufacturing, there are simple starting points: Anyone pulling out a manual gauge is wasting your time. However even in sales and marketing, you have email and mobile phones—data can be automatically gathered from these that is clearly useful. Rather than nagging salespeople to get their data entry done, why not let the systems do that themselves?
Get moving on your AI strategy
To recap, the five key prerequisites are:
- Spread AI knowledge throughout your organization.
- Everyone should understand the basic everyday things that machines can do on their own.
- Build systems and components for your AI.
- Build AI/UI mixins for easily adding AI to your business applications.
- Instrument your systems to gather the data you need to feed the algorithms to make decisions for you.
If you put these prerequisites together, the rest should follow as you transition from the Information Age to the Insight Age.
Join the CIO Australia group on LinkedIn. The group is open to CIOs, IT Directors, COOs, CTOs and senior IT managers.