How to start with AI?
Don’t start with AI
Many companies want to start with “AI”, whilst this may sound like a good idea for marketing purposes, in order to generate business value from AI I recommend a different approach.
Let’s assume you have your data available, you want to start using this data for the benefit of your company. What do you do? Don’t start with a random machine learning project, but consider doing this first.
Photo by Jon Tyson on Unsplash
Even though the buzzword may have been around for a while, chances are this topic is a new topic for your company. People may not understand AI yet (AI Literacy), and probably have the impression it is magic.
A good way to start is to get them familiar with using data in their daily business with some ‘easy’ but oh so important data science. About what you ask? Here are a few steps to follow.
#1 Know the company’s strategy
What are your company's key strategic focus areas? And what are their KPIs related to the strategy? Check the annual report, the internal portal, or speak to your manager, what are the topics that people are talking about in the corridor? Perhaps your company wants to grow into a certain customer segment, or there is pressure to save costs.
#2 Choose a handful of topics
Now that you have an insight into your strategy and KPIs, choose 3–5 topics that you can further investigate.
You can choose them based on where you know you have the data available, where you have a network in the organization, or where people are actually sitting close by in the office. For example, choose the topics related to sales and marketing, where most data should be available from your CRM solution.
#3 Exploratory data analysis
Now with these topics in mind, start your exploratory data analysis. For each topic, write down 5–10 questions you can ask the data and find the answers to these questions. Examples of questions could be
- How many customers have more than one product?
- How many of these customers have been actively approached by sales/marketing?
No need for anything fancy just yet, a quick and dirty Jupyter Notebook will do.
#4 Create actionable insights
Now go through your EDA for each question, and write down the five actionable points that could be of interest to your stakeholders. An example could be that you have found that customers actively approached by sales in a certain customer segment A have bought multiple products.
And for another customer segment B, marketing has been the key to the success of upselling for a different product.
The actionable insight would be that sales should focus on customer segment A for cross-selling, and marketing on customer segment B for upselling.
These messages you can support with a nice visualization that your management can understand. Publish in a nice format that works within your company. Where? There were your audience is.
Publish where your audience is in order to gain the right traction.
Is there an internal newsletter? Do people use slack channels? Perhaps hang it up on a printout in the coffee area?
#6 Communicate, communicate, communicate
Now just publishing is not enough. Communicate with your stakeholders, set up a meeting, speak about it during your lunch break, etc. Make sure people see the insights multiple times and encourage them to take action. Once they see the value from your insights, they will come back for more.
You’ll see, actionable insights show value with data science in a way that people will understand. Once they do, they will come back for more.