AI Basics: How to Spot Real, Helpful AI

AI Basics: How to Spot Real, Helpful AI

Distributors are in a fight to compete and differentiate. With increasingly demanding customer expectations, distributors must adopt a technology-first business model that improves the customer experience and lowers cost-to-serve. Winners in this fight will be the ones that implemented artificial intelligence (AI) early and effectively.  

There are countless applications for AI in a distribution business, including pick and pack robotics, intelligent routing, inventory planning and predictive sales analytics. Plenty of technology vendors will tout their AI features and the impact they might have on your business. But it can be difficult for distributors to cut through the noise and evaluate the legitimacy of these claims. 

How can you measure the sophistication of a technology vendor's AI and whether or not it’s right for your business? 

How to Think About AI 

AI enables a machine to think and learn like a human. Because of exponential increases in computer processing power, data storage and breakthroughs in programming techniques over the last 15 years, AI is now a multi-trillion-dollar industry. 

Like the human brain, AI observes and classifies data at massive speeds and scale. It learns over time as it observes and processes more data. Because of its sophistication, AI can be used to make decisions, recommendations or complete complex tasks. The quality of AI outputs depends heavily on how much data is used and how robust the models are. 

For example, you could build a basic hotel recommender that uses one data parameter – income. If income is greater than $500,000, suggest a 5-star hotel. If income is lower than $500,000, suggest a 4-star hotel. Simple, non-earth-shattering stuff.  

Alternatively, you could build an AI-powered hotel recommender that uses tens, thousands or even millions of data parameters to compare travelers and find patterns that help make better recommendations – income, age, location, gender, travel history, dietary preferences, family size, hobbies, etc. 

Unsurprisingly, a well-built AI model that’s leveraging more data and parameters will deliver exponentially better results. 

Through the Lens of a B2B Distributor

Because distributors have large and complex data sets, they are uniquely poised to benefit from AI. Let’s look at an example.  

Imagine you’re trying to increase sales by identifying wallet-share gaps.  

A simple approach would be to build a model that looks at one or a few data parameters – like customer type. With this data, you could build an image of what the average wallet looks like for each customer type. 

But distributors have so much more data to leverage. A more effective approach would be to use an AI model that not only looks at customer segments, but also learns from your entire data set – using customer metadata, purchase history and frequency, customer location, the names of each customer, size, D&B data, brand-preference, NAICS codes, etc. to find patterns. 

In this scenario, instead of reductive customer segmentation, AI unlocks the ability to learn about each customer individually, find the most-like customers, and identify granular category – or product-level – wallet-share gaps.

3 Things to Look for When Evaluating AI Vendors

Not all AI vendors are equal. Knowing how to differentiate between good and bad AI will help you make more impactful business decisions. Here are three important factors to consider:

1. Data sources, processing frequency and storage 

The more data an AI model can chew on, the more impact it can have on a business. A good AI vendor can store a lot of historical, varied data and will capture new data over time to keep learning. 

It’s a bad sign if a vendor puts a cap on the amount of data they ingest from you. For example, if your vendor for eCommerce personalization limits you to two years of historical transaction data, you’re directly limiting the amount of complex, multi-dimensional patterns it can find.  

Comprehensive data about each shopper allows Amazon’s AI to find patterns for every customer, providing accurate product recommendations, search relevancy and audience monetization that improves constantly. This is central to Amazon’s success.  

Vendors need to be able to capture new sources of data and safely store it in a virtual private cloud. They should also be able to control who accesses the data.   

Ask your vendor: How much data can we send you? Are there any caps? How frequently will you ingest new data?  

2. AI sophistication 

Sophisticated AI uses deep learning, which is a way to automate predictive analytics (the application of current and historical data to forecast activity, behavior and trends). Deep learning uses algorithms stacked in a hierarchy of increasing complexity. The AI then applies these algorithms until the output reaches an acceptable level of accuracy. 

Knowing the gender of the viewer is not enough data for Netflix’s AI to make a good movie recommendation. With more details about what the viewer watched in the past (i.e., genre, actors, etc.), AI zeros in on what the viewer might like. As more data accumulates about each viewer’s preferences, such as what movies they gave a higher rating to, how “bingeable” a show was or what movies/shows they never finished watching, the recommendations for what to watch next become better. Additionally, AI takes what it learns from other viewers who look like you to find more suggestions. Based on your response, it learns and adjusts its recommendations based on the new data.  

Distributors have high-dimensional data with large product catalogs and many transactions and customers. Simple models can’t deal with this complexity and, as a result, won’t identify true patterns or connections. Imagine trying to automatically identify substitute items in a catalog of 1 million SKUs by text matching – a simple AI model. You might get some substitutes, but to do it effectively, AI needs to map to higher-dimensional data to understand what the text means. It needs to be able to figure out that the color navy is similar to blue; or that two different manufacturers are creating similar items even though the descriptions are different. Deep learning models can sort through these nuances to solve higher-dimension problems. 

Ask your vendor: Do you use deep learning and natural language processing? Deep learning is a highly effective form of machine learning. If your vendor uses more basic predictive tools like segmentation and clustering, they won’t be able to identify complex patterns. Deep learning models can handle messy, real-world factors like seasonality and deliver real results.

3. AI built for distributors

Distributors may look to AI vendors to supply them with an advantage they can’t build in-house, but selecting a generic solution won’t cut it. AI designed specifically for distributors will be more effective. 

Distributors have unique challenges and types of data. Generic AI models – like a model optimized for direct-to-consumer businesses – aren’t sophisticated enough to process all the data distributors have and see the complex patterns within. Generic solutions are also not built to answer questions distributors specifically care about. After all, recommending a pair of pants to match a sweater is a much easier problem to solve than matching a conduit with the appropriate fitting. 

Ask your vendor: What questions does your AI answer for my business? Is there an omnichannel element to the AI? Can it be used on digital (online) channels in addition to offline (via customer-facing people)? 

Asking the hard questions and selecting the right AI solutions today will pay off quickly and sustainably. 

Join our thought bubble

Get industry insights and Proton news.