I am working on a project where we are analyzing customer demand for a public service. The project is to predict future demand. Pretty simple Artificial Intelligence (AI) type of scenario.
There are a lot of factors that go into demand variance in this scenario. The variety of sources of impact do make it more in the realm of AI than just a straight regression in a spreadsheet.
We are considering weather since that makes an obvious difference for what we are providing as a service. I expected weather data to be fairly straightforward and it is. In the old days of AI, we would download a big text file of weather conditions by location, find the nearest station and use relevant fields as part of a model.
New weather data fields are more dynamic especially during the prediction phase. We can dynamically query the projected data for a weather station to makes predictions using the model. Cool, but not dramatically different since the data is the same just the feed is more interactive, up to date, and so on.
The really interesting new feed had me thinking about all the new external data sources available on line in the last 5 or so years. There are tons of demographic, census, health, traffic, and other data feeds. There are so many that it is hard to make sense of why many are useful.
I have found that is especially true if people only think of using data in dashboards and reports. Descriptive analytics only looks at the history and, while it is neat to see if there was a sporting event nearby on a particular day that might affect demand, that kind of information mostly just clutters up a fancy chart on the dashboard without adding any insight.
It is when you use the new data feeds (like events happening near particular zip codes) inside of AI modeling that the data feeds make more sense and are useful. The detail possible is exciting.
For example, AI models can consider events at museums or plays at theaters to project changes in demand not just enormous sporting events. This should help with staffing and logistical preparations prior to and during the event. Dashboard based analysis might look at big sporting events or large festivals, but the ability and time required to find correlations and effects limit how deep a human can go into the depth of the events data feed. AI models are limited in detail examined only by the time required to run them.
Being able to let the machine learning algorithm find the quantifiable impact of different types of events based on distance is not something a human is going to glean easily from the dashboard. Business analysts, understandably, scratch their heads on how to use such a data feed in a historical report or dashboard.
When a company moves to analyzing data using AI methodologies, then the mass of the new data feeds becomes useful. Useful both in the analysis and the ongoing predictions coming from the model.
So, if you are a business person or reporting analyst and have been wondering who finds the massive amount of online data services useful, think about how AI can make use of all that data, combine it with your internal sales and operations data, and then provide more insight into future activities.