One cannot open a business publication, or LinkedIn without a multitude of mentions of ‘the AI enabled future’. But what is Artificial Intelligence (AI) and it’s close cousin, Machine Learning (ML), and how do they apply to your business?
In this article I will go over a basic introduction to ML and AI and outline some key considerations in incorporating them into your company.
Igor Holas is a data scientist and consultant at Inviso by Devoteam.
Machine Learning and AI are data analysis approaches that enable you to estimate answers not directly contained in your data. Igor Holas, Inviso by Devoteam
Machine Learning and AI: What is it?
To explain it is as short as I possibly can:
Machine Learning (ML) and Artificial Intelligence (AI) are data analysis approaches that enable you to estimate answers not directly contained in your data.
This places ML and AI in the same family of data analysis approaches as statistics and forecasting – let’s call these approaches, as a group, ‘advanced analytics.’
Chances are you have already encountered some form of advanced analytics in your work. Common examples include:
- Forecast of sales / revenues / deliveries for next quarter
- Predictions of which customers will leave / leads are likely to convert in the next month.
- Recommendations to a customer of the next item to add to basket or recommendations to a team member of the next action to take in their workflow.
- Evaluations of unusual behavior that may be fraud or security breach.
- Determinations of market- or customer-segments .
Each of the above examples uses some form of advanced analytics to estimate answers not directly contained in the data. We do not know the sales for next quarter. Customers do not come with a ‘segment’ label attached to them. Fraudsters do not announce themselves. Thus, we use advanced analytics to estimate what are likely the correct answers.
By being able to estimate values not in your data, advanced analytics allow you and your team to know more, and ultimately ‘see into the future’. All the use case examples mentioned above are very common in many businesses, and they are all built around ‘advanced analytics’ such as statistics, ML and AI. Chances are you are using tools like these already in your daily work life.
In your personal life, you are almost certainly encountering AI-driven services on a daily basis. Services like Google Translate, searching images in your photo library, recommendations for things to watch on services like Netflix, or things to buy on shops like Amazon all use ML in the background. Your bank is also using ML and forecasting to monitor your account for possible fraud, and is also evaluating the likelihood that you are about to close your account and move to another bank,or the likelihood you are about to ask for a new loan. Of course, the examples are endless – ML, AI, statistics, and forecasting are all widely used by many of the services we use every day.
One reason to embrace advanced analytics and one not to
While you may already be using advanced analytics, I want to clarify a simple set of ground rules for when their use is appropriate and then it is not:
Reason to embrace: These analytical methods are unique in being able to answer questions about data you do not have (yet).
Reason to not: These analyses are estimating, or guessing. As a result, there is always a chance of error – this means they are ill-fit for situations where one needs 100% precision.
Adding your first Machine Learning or AI-project to your analytical portfolio
The best way to get started on a new project is to have a compelling use case that demands these analyses. The list above can perhaps serve as an inspiration as to what kind of projects to pursue. Whatever business problem you decide to solve, here are some key things to line up as you are getting started:
#1 To look into the future, we need data on the past
First, there is a prerequisite for any advanced analytics: having historical data. In order to ‘look into the future’ we need data on the past. This data does not need to be advanced or complicated – one can easily use historical data saved in an excel spreadsheet; use what you have. However, if you are predicting sales for the next quarter, we need a few years of historic data on past sales.. If you are predicting customers’ future behavior, you’ll need data on their past behavior.
If you do not have this historical data, your journey will likely need a ‘step zero’, meaning that you will have to begin by collecting the relevant data and organizing it for future analyses. We, at Inviso, are happy to sit down with you and review whether you have the right data for your use case.
#2 Start small
Having ensured that you have historical data available, the next step is to always start small. Like with any new tool, the best way to deploy and test it is in an isolated low-risk setup. Luckily, it is quite easy to set up your first predictive, or statistical models in a low risk way.
Here’s a few examples::
- Set up your prototype ‘on the side’ observing the results, without having the results impact your workflows and your clients. Measure whether the results are what you’re expecting and better than status quo
- Incorporate the results into your staff workflows, as added insights, or ‘next step recommendations’ – but still allowing your staff to make the key decisions themselves – here, measure whether the recommendations are relevant, helpful and precise.
- Run an A/B test if that is relevant for your business.
In starting you small, and ‘off to the side’ you are ensuring that you are free to experiment, and make mistakes, without attracting negative attention from decision makers at your organization.
#3 Get the management on board
Like any other project, showing value early, and avoiding causing disruptions will help with leadership buy-in. If you structure your goals in a way that shows very clear measurable value to your business, and keep delivering this value with your analytical product(s), you will continue to get more support to keep moving forward on the journey towards becoming data driven.
Case study: Implementing advanced analytics with Dribe
We did just this at Dribe, a danish car-subscription company, that required data automation and eventually predictive analytics to manage its fleet and make its cars available truly on demand.
As part of Dribe’s journey to derive better value from their data, our team began adding advanced analytics in the form of data predictions shown along measured data (data not calculated using statistics or ML). These values, such as predicted near-future fleet utilization, allow the Dribe team to evaluate the measures’ usefulness, refine them, and ask for more predictions – at the same time, the predicted values are not replacing values calculated from existing data, ensuring that if the prediction is wrong, the team at Dribe can fall back on a different value.
Our work on Dribe also shows how you can begin exploring the value of advanced analytics with tools you have. Dribe has eventually presented their data in Tableau dashboards and prepared the data in Alteryx, because those were the tools they were already using for their data. However, you could easily accomplish these tasks in tools your team likely uses like PowerBI, Power Automate, or even coding languages like SQL, R, or Python – whichever is relevant for your organization.
“…they unlock your ability to understand what will happen next in a way that other approaches cannot.”Igor Holas, Inviso by Devoteam
Advanced Analytics opens doors to answers to new questions
Advanced analytics, like ML and AI, thus represent a way to begin answering new questions in your work. At the same time, ML and AI are no longer some scary, high risk, cutting edge tools only for nerds and programmers. The tools you use currently with data likely can offer key ML models for the analytical questions relevant to your business. Getting started is thus truly easy.
How to take the next step
I hope this article has helped you get an idea of what ML is, AI and advanced analytics as a whole, and what these methods can do for you. I also hope you have come away understanding that these tools should not be feared, and indeed are likely close to tools you are already using – taking the next step into ML or AI is just another step in your data journey. Start small, and not be afraid of not getting it right the first time. If you take some of the precautions mentioned above at least.
Also, don’t be shy to reach out to me or one of my colleagues in Inviso by Devoteam if you want to to discuss your first ML project to get better at predicting your customers’ needs and growing your business.