What’s your use case? “Send Everything!”

Earlier this week, one of our YouTube subscribers posted a question on a video—"How to get OPC data into Microsoft Azure”—that we published back in October 2020 [time flies]. While it’s an old video, his point was relevant four years ago and it’s relevant today.?

The question:?

“This looks great and simple process for sending data to the cloud. While this approach is promising and straightforward, we need to consider real-world industrial scenarios. For instance, a single injection molding machine might generate over 300 data points per second. With 20 such machines, this translates to 6,000 messages per second, which could significantly increase cloud costs and configuration efforts. This is one of the use cases and issues that I am facing in my current journey and figuring out how can we optimize this process to minimize costs while ensuring data integrity and real-time cloud delivery? I would be very glad if you could recommend some solutions. Once again this is a great start and much more helpful. Thank you.”

With the last 4 years in perspective, let me walk you through what I commonly see.

1)?????? Customer starts by sending all raw tag data to the cloud with no context other than a tag name. They spend a lot of effort to do this and struggle with assembling the context in the cloud to enable use cases. The ROI to the business can't be measured, it's expensive, and the project dies.

2)????? Customer still wants to send everything, but realizes they need to contextualize at the edge. They model their assets and send everything to the cloud. Someone goes to use the data for application X, Y, Z and realizes the data they need isn't there, isn't in the right format, or isn't scanned fast enough, etc. Enabling use cases is still hard, there is a high upfront cost to modeling everything, and cloud spend is still high. ROI is hard to measure, cost is high, and the project dies.

3)????? Customer takes a use case-based approach. It could be energy monitoring, scrap reduction, etc. Customer creates the data model required by the cloud application and pushes it to the plant. The plant then maps the use case to the required data. There is immediate ROI, cloud spend is predictable, and the use case sticks and scales across factories. While doing this, plants start building out their asset model as needed to make connecting future use cases faster.

I think too often we as an industry treat cloud like factory historians, where we send everything "just in case" we need it. But cost structure wise that's not what cloud is and this approach leads to high effort/cost initiatives that don't deliver ROI and end up hurting manufactures and cloud vendors in the long run.

It's a long answer, but hopefully a useful perspective. The short answer is only send what you're going to use.