It takes time for new technologies to be accepted, let alone become ubiquitous. At some point, technologies such as cell phones, desktops, laptops, email, apps were looked at with skepticism before they became essential parts of our everyday lives. In a manner similar to the aforementioned technologies, Machine Learning and Artificial Intelligence are slowly but surely permeating the CPG industry.
What Is Machine Learning?
Broadly speaking, Machine Learning encompasses a set of algorithms that are used to identify and predict patterns in data. These algorithms work well with large data sets and can ingest a wide range of data types, which are essential to effective demand planning.
When combined with today’s powerful computing capabilities, machine learning algorithms enable us to view and leverage data in ways that weren’t common years ago. The creation of machine learning models is an iterative process with the goal of precision.
What Is Artificial Intelligence?
The term Artificial Intelligence is commonly used to refer to Machine Learning algorithms that have achieved high levels of accuracy. While not an actual artificial brain, the predictive capabilities of an industry-specific AI can exceed the predictive capabilities of a person. Ironically, these predictive capabilities typically are the result of a team of data scientists who have tuned a set of advanced machine learning algorithms on a specific set of data.
How Data Science Can Help CPG Companies
In the CPG and retail industries, using machine learning algorithms to predict demand is a better approach than current industry practices, such as using manual spreadsheets. Here are three business benefits of implementing ML and AI.
1. Better data, better formulas, better demand forecasts
Depending on the firm (and its resources), some professionals analyze historical sales data to predict demand using Excel formulas. Others may try to take into account the impact of a particular time of year, such as a holiday, and “guesstimate” future sales.
But the ability to predict SKU-level sales using spreadsheets is not a scalable practice given the sheer amount of data that should be accounted for when forecasting sales, inventory, manufacturing and distribution. While these aforementioned, data-driven approaches may at times be sufficient, there’s no consistent benchmarking of predictive error, which can result in serious forecasting errors.
Ash Patel, the Chief Information Officer of IRI, a market intelligence company, suggests letting the machines "do the heavy lifting… [to] sort through trillions of combinations of relevant data points."
2. Spend marketing dollars wisely for e-Commerce
Another use for machine learning within CPG is in understanding how best to deploy marketing dollars when selling via an online marketplace, such as Amazon or Shopify. This means ingesting large amounts of data, including keyword search statistics, social media ad spend statistics, search engine keyword performance reports, and product reviews for all competitor products. Based on this information, data science teams will model out different scenarios in an attempt to quantify the impact of each data point and ultimately choose the optimal scenario.
Once again, the number of different parameters consumed and examined is quite large and lends itself to a machine learning process run on enterprise computing infrastructure.
3. Maximize sales and reduce working capital
The benefits of applying machine learning models to the retail industry are often quantifiable in dollars. For example, increasing sales prediction accuracy can translate to fewer expedited shipping costs, less working capital tied up in inventory, fewer unplanned production runs, and fewer lost sales from stock outages. Deploying marketing dollars more efficiently can result in a greater advertising conversion rate and increased sales.
How to Implement Machine Learning and Artificial Intelligence at Your Company
1. Access and organize all data
This process can be somewhat challenging if a CPG company’s data, from inventory to sales, is spread across different legacy systems. Other complicating factors include poorly organized data, or data that’s siloed in different departments across the company.
Meanwhile, the number of different data points that can affect sales is massive. This should include all sales history, consumption data, weather, search engine statistics, social media mentions, ingredient lists, brick-and-mortar foot traffic data, and much, much more. As you can imagine, consuming all of this data is better suited to a machine learning approach run on enterprise-grade servers rather than an Excel model run on an analyst’s laptop.
2. Customize the machine learning models
Model customization is typically done by a team of data scientists who combine a firm’s data with various external data points (many of which are previously mentioned). This process involves iterating over different machine learning algorithms until a predictive model is created that meets or exceeds predetermined benchmarks. From there, a team of software engineers can use the customized models to create a scalable solution that interfaces with a brand’s specific internal data and systems.
3. Leverage the results
Similar to the most popular software apps, user adoption is dramatically increased when machine learning models are coupled with intuitive software and a high degree of automation. It’s one thing to have the most sophisticated algorithms; it’s another thing to be able to empower your organization to interact with it in an efficient manner.