Brian Parsonnet Kept Finding Better Uses for Yesterday’s Data

A factory can remember everything and understand nothing. Its historian keeps the temperatures, pressures, valve positions and vibration readings in obedient chronological order. Then an engineer asks why Tuesday’s batch behaved differently from Monday’s, and the orderly archive becomes a scavenger hunt. Data must be found, aligned, cleaned and exported. The question is urgent. The answer has apparently gone out for lunch.

Brian Parsonnet has spent more than four decades working in that gap between a system’s memory and a person’s decision. Today he is a founder and Corporate Fellow at Seeq, responsible for product strategy, operational-technology vision, intellectual property and research. The title sounds broad because the problem is broad. Industrial intelligence is never only an algorithm. It is also timing, interface, context, trust and the obstinate physical world on the other side of the screen.

His route to software began with machinery and control. After studying electrical engineering and computer science at Princeton, then control theory at Columbia, Parsonnet worked across diagnostics, maintenance, equipment health and distributed systems. He co-founded Loveland Controls Company, held senior technical roles at Honeywell and later helped start Ice Energy. Long before “industrial AI” became a conference badge, his work concerned the less decorative question beneath it: how does a complicated system give a useful answer?

01 / Cold storageThe battery was a tank of water

At Ice Energy, the abstract problem became a very large, very literal block of ice. Commercial air conditioning bears down on the electrical grid during hot daytime hours. The Ice Bear shifted part of that burden. It froze water at night, when power could be generated more efficiently, and used the stored cooling during the day. Instead of asking the grid for everything at the busiest moment, the machine arrived with yesterday evening’s cold in reserve.

Parsonnet led the system’s development. When Fast Company profiled him in 2011, he described the central move with a sentence that deserves a place in the small museum of lucid engineering explanations: “We replace the compressor with ice.” The publication reported that roughly 7,000 units had been sold by then.

The Ice Bear makes a useful prologue to Seeq because it reveals Parsonnet’s fondness for time as an engineering material. A kilowatt-hour available at night is not identical in value to one demanded on a scorching afternoon. A sensor reading buried in a historian is not yet an insight. Storage matters, but dispatch matters more. The trick is to deliver the resource when a human or a grid can use it.

“We replace the compressor with ice.”Brian Parsonnet, explaining the Ice Bear in 2011

02 / The archive problemA historian is not a conversation

Steve Sliwa met Parsonnet while advising Ice Energy during a reorganization. Sliwa had been considering semi-retirement after building the drone company Insitu. Instead, the two found a new problem. Industrial businesses were spending heavily to collect time-series data in process historians, but the architecture and available tools made serious analysis cumbersome. Plants were data-rich in the way an attic is furniture-rich: technically abundant, practically awkward.

In September 2013, Sliwa announced Seeq with Parsonnet as his partner, Mark Derbecker and Jon Peterson as co-founders, and a group of founding employees. The early description was plain: decision support for time-series big data. The product would sit with the infrastructure customers already had, helping process experts investigate periods, compare conditions and turn operational records into answers.

That distinction matters. Seeq was not founded on the claim that industrial companies lacked information. They had too much of it, arriving from equipment with names only an engineer could love. Nor did the company presume that the process expert should hand every question to a data specialist. The bet was that good tools could let the person who understood the plant remain close to the analysis.

The chart is conceptual, not a measure of effort. Its point is the accumulation. Embedded systems taught the discipline of physical constraints. Controls added feedback. Ice Energy made timing visible at grid scale. Seeq brought those habits to software: respect the process, preserve context and do not mistake a pile of readings for a decision.

03 / The patent trailBetween the average and the alarm

Patents are often presented as a number, the intellectual-property equivalent of counting medals. Parsonnet’s are more revealing when treated as a sequence of irritations. One concerns a managed virtual power plant using aggregated storage. Another distinguishes the settled portion of a streaming calculation from values still liable to change. A third reduces the dimensions of multivariate data while trying to retain fidelity to the calculation that matters.

Two Seeq patents granted in 2025 make the sequence particularly clear. One covers methods for learning from multivariate time-series patterns, including precursors that may predict a recurring condition. The other addresses a quieter interface problem: how to display dense time-series data at different levels of detail. Its slider moves between a single average and a representation that preserves minimum and maximum values, with useful gradations between them.

The display patent sounds modest until one considers how much industrial judgment occurs in a graph. An average can be tidy and disastrously polite. It may conceal the brief excursion that damages a batch or precedes a failure. Minima and maxima preserve drama but can turn a long history into a black thicket. A useful display must negotiate between legibility and evidence. The interface is not wrapping paper around the analysis. It participates in the analysis.

The recurring concern is not data collection. It is the last mile between stored evidence and an accountable choice.

04 / The fellow’s briefKeeping one foot in the plant

At Seeq, Parsonnet’s current remit spans product strategy, OT vision, patents and research. “OT” is the useful half of the familiar IT-OT pairing: the operational technology that monitors and controls physical processes. Its clocks are different from those of ordinary business software. A delayed report can be annoying. A delayed or misleading process signal can spoil material, waste energy or send an operator toward the wrong valve.

That helps explain why Parsonnet’s public arguments go beyond better algorithms. In comments filed with the California Energy Commission in 2023, he advocated an application layer that could connect raw data to the end result while supporting authorization, traceability, collaboration, reporting and protection of users’ intellectual property. He also argued that historical and real-time data should be usable without rebuilding every solution for each side of the divide.

This is not the glamorous part of artificial intelligence. It is the plumbing, governance and continuity that determine whether an impressive demonstration survives contact with a regulated factory. Models come and go. Plants keep their maintenance schedules.

Parsonnet also advises Ortomation, a company working on real-time optimization for process plants. Its biography of him makes an unusually helpful observation: his experience reaches from embedded systems to modern SaaS architectures. That span is less a boast than a form of productive suspicion. Anyone who has watched software meet a compressor knows that the word “seamless” is usually where the seams begin.

05 / The long viewWhat the machines remember

Colleagues writing public recommendations describe Parsonnet as a cross-disciplinary thinker who sees connections others miss, and as someone able to motivate a team around those ideas. His record gives the description some ballast. The early work crossed electrical engineering and computer science. The Ice Bear crossed refrigeration and grid management. Seeq crosses data engineering, process expertise and interface design.

There is also an old paper that makes the long arc unexpectedly vivid. In 1979, Parsonnet was among the authors of research on the beat-to-beat stability of implanted pacemakers, devising a way to measure tiny timing variations. Decades later, his patents would again concern streams, intervals, stability and departures from normal behavior. The applications could hardly be more different. The habit of attention is recognizable.

Seeq now sells into industries including chemicals, pharmaceuticals, energy, mining, food and beverage, and semiconductors. The scale is global; the founding complaint remains intimate. Somewhere, an engineer is staring at a trend and wondering what changed. Somewhere else, a plant has recorded the answer without making it easy to find.

Parsonnet’s career offers no grand finale to that predicament, which may be its most honest quality. Better tools create better questions, and better questions uncover more work. The tank of ice melts. The newest values in a streaming calculation settle. Yesterday’s data becomes today’s explanation, if the system is designed well enough and the person looking at it knows where to press.