It’s been five months since we departed the Microsoft Accelerator program in Seattle.
Time flies. We’ve made great progress and the list continues to grow based on customer feedback. That’s the exciting part. I’d like to reflect for a moment as to where we have needed to pivot or persevere in relation to what our aspirational goals were while at Accelerator.
First, the most important thing we have learned – this problem is difficult. There is a reason lots of organizations rely on Microsoft Excel to manually align anomalies and calculate growth. That said, we believe we’re moving the in the right direction and are almost there.
Pivot or Persevere
-
- The data is very unstructured. We anticipated some of this going in as we would be ingesting data dating back to 1999. What we didn’t anticipate is that it didn’t get much better as we traversed the decades. Hence, our first pivot – build a data mapping and validation mechanism so that we can ingest millions of inline inspection records. Otherwise, our algorithms wouldn’t be very effective.
- “Truth” data is hard to come by. We have ingested roughly 20,000 miles of pipe containing roughly 2mm+ of weld and 500K+ of anomaly “truth” feature types. We feel more confident with weld truth data and are currently producing a 99.65% confidence among matches, it’s the remaining .35% that we need to still validate. Anomalies on the other hand create less confidence in truth due to the increased volume and additional inline variables. However, we’re now sitting at 78% confidence and doing some pretty cool things here which I’ll cover in a future posting.
- Easy button. The daily work load and volume of data make it near impossible to break out of existing legacy work flows. It’s pretty universal that everyone wants an easy button. Any solution had better have the ability to easily ingest inline inspection data into the solution and rank the pipe segments with the greatest amount of corrosion at the top. We’ll continue to persevere – make business intelligence the front-end of the application and provide summary to detail anomaly growth analysis in a single step.
From time to time, we’ll continue to post about specific or unique lessons where we either pivoted or persevered. For now, we’re fortunate to have our current customers on-board and very actively working with us on a daily basis. We view them as pioneers and they are playing an important role in validating that machine learning can, in fact, evolve our product toward our ultimate objective of saving lives and protecting the environment. In fact, they are now picking excavation sites based on data analyzed in our solution which will expand our understanding of truth data and further build algorithm confidence.
We are very close to a general release of our product which was first conceived in March of this year. As with all cloud solutions, it will never be really done. In fact, we initially anticipate bi-weekly release cycles focused on our vision for the final product. We anticipate the next development cycle to be massaged with customer feedback, the constant evolution of our algorithms, and other normal product roadmap items.
We anticipate quite exciting times ahead. Stay tuned for further updates.
