Enhance your demand forecasting with recurring events
Recurring events are scheduled, attended events which happen multiple times over a period of time – most commonly, every year. For example, the San Francisco pride parade is a recurring event that takes place every year. Recurring events may or may not take place at the same time and place each year, which complicates tracking at scale.
PredictHQ uses the concept of recurring event groups to link events together using our entities system. Recurring event groups make it easier to identify, search and retrieve all instances of a specific event within our APIs and our web application, Control Center.
This quarter, we’ve launched new recurring events models to increase the pace and scale at which we create recurring event groups. More coverage for recurring event groups means it’ll be easier to identify and plan ahead for recurring events that impact your business locations – and instantly see all other instances of any given recurring event.
Let’s take a look at how to use this exciting improved feature to get even more out of the PredictHQ platform and our data.
Use recurring events to enhance demand forecasting and optimize your operations
Annually recurring events present an opportunity for companies to compare external data and the impact of the event against their own historical sales data. For example, let’s say you manage a Dunkin’ Donuts located in downtown Boston. Each spring when the Boston Marathon rolls around, your store location is completely slammed. By analyzing past years’ transactions, you can align demand spikes with this recurring event – and accurately predict demand increases for future years. This powers the ability to fine-tune inventory to make sure you have just enough product for that day, every time. Now we know the Boston Marathon is in the same location every year, so it’s a bit more predictable. But what about the events that move around each year?
Understanding why you were busy last year means you can better plan your resource requirements for the same period next year. What if the event changes location? Recurring events gives you the level of visibility needed so you can align staffing to match incoming demand with reliable accuracy. Keeping your stores properly staffed and stocked during high-demand periods makes for a positive customer experience, and prevents lost sales opportunities.
The Boston Marathon is a popular, well-known event – but what about the less-visible recurring events near your business locations? Maybe you want to be able to see events of a similar profile that may recur in the future so you can plan accordingly.
Every month, PredictHQ captures an average of 390,000 new events globally and adds them to our dataset. These include recurring events such as conferences, expos, festivals, and more, all of which could be influencing your demand. With recurring events and by leveraging our intelligent event data, you can in a highly-predictable way, track and plan accordingly.
How recurring events models work
In order to ensure the best quality, confidence and coverage, our recurring events model has a two-pronged approach:
Firstly, it scans all of our existing event records (millions of events in the past, present and future) in order to determine any patterns of recurrence - this means it looks at events across time, geographies and it even leverages Natural Language Processing (NLP) in order to scan for patterns in names, descriptions, performers, teams and more. If a strong pattern of recurrence is identified, a new event group is automatically created and all eligible events are linked to it.
Second, the model processes all incoming events as we become aware of them. If the incoming event is found to be recurring, it is automatically linked to its corresponding recurring group. Otherwise, the event may simply not be recurring, or there are not enough instances to create a recurring group around it - for example, it may be due to it being a newly introduced set of events. If an event is not linked in this way, it is periodically reassessed for its linkage eligibility as new groups and events become available.
Combine this with our stringent data quality checks and the result is an automated model that we are sure will meet and exceed the quality and coverage standards that we are known for.
Having seen and used the results of these models ourselves, we are quite happy with the outcome - to attempt to replicate this process manually would require over 20 minutes per event group, give or take a couple of minutes based on the size and complexity of the set. And this is a benchmark set by our engineers who live and breathe event data. Needless to say, we’ve succeeded in automating a very time intensive and complex process.
To date, we’ve invested a lot of manual time into recurring event groups to get 6,400 recurring event groups.
How recurring event models boost data quality + event coverage
The recurring events models will provide a huge boost to our events coverage, which means ever better data quality across event categories. Over the coming weeks, the coverage of our recurring event functionality is expected to expand from 6,400 recurring event groups to over 192,000 – all thanks to the models.
This influx of 186,000 new event groups would take 50,000+ hours of manual collection and grouping. Automating this process means our data assurance team can spend those hours focusing on other ways to improve the PredictHQ user experience.
This is only the first phase of the recurring events model. As the model gathers more recurring event groups, we’ll be investigating new ways to further improve how we identify and link events to existing event groups.
Want to learn more about how recurring events work? Dive deeper with the free whitepaper, The technology behind recurring events.
Find recurring events near your business locations
Want to learn more about how recurring events work? Dive deeper with the free whitepaper, The technology behind recurring events. your business locations? Log in or get started with your 14-day free trial today.