Ahmad Fattahi

Observations at PI World 2018 from data scientists angle

Blog Post created by Ahmad Fattahi Employee on May 4, 2018

PI World San Francisco 2018 was a huge event hosting more than 2000 guests from multiple industries. From the data science angle we learned a lot about what our the community needs to move from raw data to value. We also learned how our partners are enabling this journey through their products and services. Here are some key observations:


1) Maturity: big industries are starting to get out of the hype phase generally. We saw multiple great presentations where dollar or man-hours were being saved leveraging data science.

2) Opportunity: there is still quite a bit of low hanging fruit in most industries. If you are a data scientist trying to squeeze a fraction of a cent from a dollar by optimizing your deep learning algorithm elsewhere you may be able to make much larger impact for many of these big industries. Examples abound such as oil and gas, power and utilities, metals and mining, pulp and paper, data centers, or life sciences.

3) Challenge: while there is low-hanging fruit the journey is not challenge-free. Unlike mainstream "data-centered" business models most of these companies have been around for decades. They are designed for purposes that may not have included "data" as a centerpiece. As a data scientist you may need to go extra distance to fetch, cleanse, shape, and transfer data. Luckily there are major advances in software and communication infrastructure, such as PI ecosystem, that makes the task increasingly feasible. There are also human and cultural challenges; many traditional industrial experts may still be suspicious of the impact of digital transformation on their career.

4) Data quality: data quality seems to be one overarching challenge across the board. Given the physical structure of industrial systems the data is prone to latency, sensor malfunction, sensor drift, cyber attack, time stamp misalignment, communication fault, configuration error, and several other issues. A wide range of efforts from software vendors as well as industries themselves is required to make the data more reliable. Fortunately there is already quite a bit of advances in the field in infrastructure and at the application layer; more work is underway.

5) Presentations: last but not least the recordings of my two presentations at PI World are now available. The first one was co-presented with Dallas Swift and the second one with Kleanthis Mazarakis.