When you hear the phrase “I heard it through the grapevine” you may think of the 1960s hit song. But actually the phrase dates all the way back to the 1700s. Gathering useful information through informal, person-to-person social interaction has been with us for millennia. Only recently has it gone digital.
And now there is another development: leveraging the grapevine (social listening) for demand forecasting. Before we explain how, first let’s discuss why. Lora Cecere did a nice job of this in her blog post “Using the Bits: Building Digital Outside-In Processes”. She says it’s all about outside-in versus inside-out supply chain planning.
Cecere explains that most companies still focus solely on inside-out transactional data, “The traditional supply chain depends on orders and shipments. Both signals have data latency (out of sync with the market) and accelerate the Bullwhip Effect. MRP and DRP logic uses order and shipment data permeating the Bullwhip Effect."
"The movement to outside-in processes has the potential to reduce waste and prevent the distortion of signals between nodes in a value network,” she says. An outside-in process is not just an extension of a traditional inside-out optimization. It requires embracing new forms of data such as unstructured and semi-structured data. Cecere explains, “Historically we have lived in a world of structured data. We are just starting to envision a supply chain world that embraces the variety of data types. The redefinition reshapes our paradigms.”
She says that outside-in social data is often acquired and examined in the marketing department, but is seldom shared with the supply chain team. “Without sensing, and market-driven outside-in processes, the organization is out of step with the markets they serve,” she says. “Today’s supply chain processes respond. They do not sense. Because they cannot sense, they do not adapt.”
At ToolsGroup, we have begun to incorporate social media data or “social sensing” into demand forecasting by adding new data sources that enhance forecasting models, especially new product introduction forecasting and promotions forecasting. We are leveraging a new system called “Groover” that listens to social channels and gauges consumer sentiment to enhance supply chain planning. It monitors and archives live tweets on specific brands. This is big data—e.g., 45,000 tweets archived in just minutes.
Natural language processing, a subset of artificial intelligence, interprets the social communication by focusing on whether the sentiment is positive or negative. Groover can then look at the network impact by measuring the activity level—how many tweets around a sentiment or topic took place—and its reach—how many times it was retweeted, and how many people saw it. It also measures active reach—people commenting on and interacting with the sentiment; such as a “like” on Facebook.
Groover determines how active these people are by the kinds of comments they typically make; from this, you can cluster the sentiments, to understand what types of users are interacting with a post and what they are expressing. With Geolocation-enabled smartphones, you can also pinpoint the location where the activity is happening. If it’s taking place near a particular store or city, you can capitalize on the sentiment and activity level with promotions, and tweak the forecast to account for uplift. So you can answer questions such as:
- Sentiment - Do people like the new sunglasses?
- Location - What are people saying near a particular store?
- Activity levels – What are the reach and networks effects?
For example, in the image below shows a live feed of tweets that reference Costa Coffee. If you click on the image it takes you to a “zoomable” map (map with zooming capabilities) where you can see a geographic dispersion of tweets that hashtag the Costa brand.
It’s crucial to understand not only what people are saying but if the social messaging is a true signal and predictor of sales. For example, Instagram is typically a clearer signal, because people are sharing emotions and moments when they interact with a brand. You also have to factor in timing. Sales don’t happen at the same time that people share expressions. When people talk about a brand negatively, it is often a consequence of a bad experience with the product or service. Sales may have already suffered, and the complaint became evident afterward. These are excellent applications for machine learning.
Social sensing can be especially helpful for new product introductions. Imagine Nike launches a sneaker with a big campaign, and people start talking about it. It’s valuable to be able to measure pre-launch, and during the first days of launch, sentiment, activity, reach, and location, to get an early picture of how popular the product might be to adapt demand forecasts.
Another good application is helping to forecast the impact of promotions. Our research shows that social sentiment is highly correlated with the demand shape for a promoted item, so if a company is launching any kind of promotion—a discount, TV spot, or marketing campaign— social sentiment can offer early indicators of how the market perceives the offer. This can signal where to stock more goods, or to ramp up marketing efforts in a location with little social commentary.
Social listening is now used by companies to assess how their brand is perceived and how their marketing campaigns are being received. Now we can begin to correlate social sentiment with demand signals. A stochastic (probabilistic) demand planning system with embedded machine learning can use this data to improve demand forecasting.
Click below for a short whitepaper that describes many ways to leverage data to improve demand forecasting.