The internet of things generates a lot of data that needs to be processed, and some innovative startups recognize that artificial intelligence can lighten the load.
Network World | AUG 21, 2018 2:22 PM PT
Plants, factories, and manufacturers in general are embracing IoT, which in turn is driving the use of artificial intelligence at the edge of corporate networks as a way to streamline industrial processes, improve efficiency and detect maintenance issues before they become problems – perhaps even big problems that could force plant shutdowns.
The AI-powered IIoT space already has a couple of its own unicorn startups, such as Uptake with $258 million in funding and a $2.3 billion valuation and C3 IoT with $243 million in VC funding and $1.4-plus billion valuation (the C3 IoT valuation was not publicly updated with its latest $100M round of funding, so its valuation is outdated and conservative).
The competition in this sector is brutal, but the opportunity is big enough that the 10 startups highlighted here still have room to maneuver and time to scale up. Keep an eye on them because one or more could well be the next unicorn in this hot market. (Click here for a description of how we chose the final 10.)
Link: Theatro
What they do: Provide an AI-powered personal assistant and voice-controlled productivity platform for employees in manufacturing, brick-and-mortar retail and hospitality
Year founded: 2012
Funding: $8.8M from Khosla Ventures and angel investors
Headquarters: Dallas, Texas
CEO: Chris Todd, who was previously the CEO of AppTrigger
Problem they solve: According to the U.S. Department of Labor, the 32 million hourly service employees in the U.S. represent the fourth-largest employment sector in the economy. However, enterprises with large staffs of hourly employees have yet to realize the massive productivity gains from smart mobile devices that enterprises with large staffs of knowledge workers are achieving, nor have they benefited from IoT trends that are boosting productivity in other sectors, such as manufacturing.
The problem is that hourly service and manufacturing workers tend to be slowed down if required to carry a smartphone, tablet or other device while doing their jobs because they need their hands free and their eyes focused on their work. Moreover, most hourly employees don’t have a company email or voicemail, and in many states, hourly labor laws make it difficult for them to access necessary company information when off the clock.
How they solve it: Theatro’s “heads up, hands free” mobile solution provides a software suite of productivity and communication applications, delivered as SaaS and designed to optimize employee, sales, and operational performance.
Theatro allows workers to access an AI-powered virtual assistant via voice commands to gain information about their store, plant, or warehouse almost instantaneously. Theatro’s suite of workforce productivity applications operate in conjunction with an enterprise’s backend systems.
For example, when working the retail floor, an associate can connect a Theatro-enabled device to an AI-powered virtual assistant that talks to the employee through that person’s earpiece. The assistant can help employees understand things like inventory and back-of-the-house operations, so more employees can be on the floor working with customers instead of physically searching for answers in outdated systems in the back of the house.
The platform also connects employees to one another, to headquarters and to other enterprise information systems. Using Theatro, employees can find each other by name, expertise or even location. Employees can collaborate directly by tapping their Theatro-connected earbuds and saying “Hello [coworker’s name].”
With Theatro, employers are also given insights into what their employees do, how they work as a team, and what makes them successful. The AI-powered Employee Analytics application measures social interaction data to understand what impacts productivity and who the top performers are. It then performs predictive analytics, enabling businesses to create a streamlined repeatable model that maximizes productivity.
Theatro says that customers are using its platform to decrease customer wait times, connect regional leaders to their teams, and boost employee productivity by as much as 10-30 percent.
Competitors include: Zebra, Inkling and Tulip
Customers include: The Container Store, Cabela’s, Neiman Marcus and Total Wine & More
Link: Alluvium
What they do: Provide an IoT analytics platform that turns complex streams of data into simple, real-time insights
Year founded: 2015
Funding: $4 million in seed funding from Work-Bench, IA Ventures, Bloomberg Beta and Lux Capital
Headquarters: New York, N.Y.
CEO: Drew Conway, who also co-founded DataKind, a global non-profit, and is the co-author of Machine Learning for Hackers.
Problem they solve: Industrial operations collect more raw data than even expert operators can manually observe and analyze on their own. It can be difficult to determine what data matters and what is just noise. Important but small blips in the data are easy to miss, which can lead to underperformance, downtime and even safety incidents. Missing important signals within the noise ends up being costly to an industrial operation, both in terms of monetary and productivity losses.
How they solve it: Leveraging machine learning and AI, Alluvium helps industrial companies achieve operational stability and improve their production. Alluvium’s flagship product, Primer, uses machine learning to help companies distill complex, massive streams of raw sensor and production data into usable insights.
Industrial teams can rapidly navigate data to identify where and when deviations occur, determine the source of the problem and decide what action to take. Primer creates a Stability Score analysis based on collected data drawn from any timeframe – from last night to, say, the past year. Primer helps operators identify anomalies from sensors, single machines or an entire facility to make the changes necessary to keep operations running smoothly.
Competitors include: OSIsoft, C3 IoT, Uptake, Foghorn, Presenso, Falkonry, and Manna
Customers include: At the time of publication, Alluvium did not have any customers willing to go on the record.
Why they’re a hot startup to watch: Alluvium has locked down $4 million in seed funding, and while their team is small and a bit green (not unusual for an early stage startup), Alluvium has entered a high-growth but confusing space. In such a noisy market, Alluvium’s focus is on simplicity. Alluvium boils down the data generated from complex production systems into a set of metrics that non-experts can understand. The resulting “Stability Score” provides at-a-glance data points operators can use to easily track key variables that could prevent them from meeting business goals.
Link: Arundo Analytics
What they do: Helps heavy industrial companies with complex physical assets improve operations through machine learning and other advanced analytical applications
Year founded: 2015
Funding: $35 million. Their most recent round, a $28 million Series A, closed in the first half of 2018. Investors included Sundt AS, Stokke Industri, Horizon, Canica, Strømstangen, Arctic Fund Management, Stanford-StartX Fund, and Northgate Partners.
Headquarters: Houston, Texas
CEO: Tor Jakob Ramsøy, who was previously a Senior Partner at McKinsey & Company, where he led the technology service lines for the Global Energy Practice and EMEA Big Data/Advanced Analytics. Ramsøy was also country manager for McKinsey Norway and led the Business Technology Office in Scandinavia.
Problem they solve: Industrial companies face unique challenges when attempting to integrate machine learning and other advanced analytical applications into their daily operations. These challenges include managing complex, highly engineered physical assets; dealing with layers upon layers of equipment and instrumentation that has accrued over many decades; and coping with varying levels of control systems, ERP systems and data stores that are often scattered across multiple operating companies, subsidiaries, acquired entities and even third-party vendors.
With that much complexity, ingesting live data and then streaming it to a highly available, cloud- or edge-based machine-learning system is no small task. Even more difficult is pushing out models based on that data in a timely fashion, so the data can inform business decisions.
Once you accomplish that, you must still figure out how to scale machine-learning applications from a handful of models to dozens or hundreds of models across the numerous assets and use cases in a typical industrial operation.
How they solve it: Arundo Analytics automates the end-to-end challenges that emerge when large industrial companies from oil and gas, power and shipping attempt to use edge analytics to drive daily business decisions.
Arundo applies machine learning and advanced analytics to edge data inputs, integrating those inputs into daily business operations and scaling these applications throughout the enterprise. Arundo helps companies deploy machine-learning models to the cloud in order to create enterprise-grade software applications, which the platform then manages.
Arundo also offers configurable, out-of-the-box applications for common industrial challenges, including equipment-condition monitoring, system-anomaly detection and a virtual multiphase flow meter, which the startup developed and jointly markets with the global industrial technology company ABB.
Competitors include: GE Predix, Siemens Mindsphere, ABB Ability, C3IoT and Uptake
Customers include: Equinor, AkerBP, Carnival Maritime, DNV GL and INEOS
Why they’re a hot startup to watch: Arundo Analytics has the second-largest VC haul – $35 million – of all the startups in this roundup. Their leadership team gained experience at McKinsey & Company, Aker Solutions, Siemens and other high-tech and industrial-focused companies. They have a respectable list of early adopting, on-the-record customers, and Arundo’s out-of-the-box industrial applications help manufacturers quickly overcome common headaches such as anomaly detection.
Link: Canvass Analytics
What they do: Provide an AI-powered predictive-analytics platform for industrial applications.
Year founded: 2016
Funding: The first of two seed-funding rounds led by Real Ventures with participation by Barney Pell closed in May 2017, and financial terms were not disclosed. The second closed in July 2018 and will be announced this month.
Headquarters: Toronto, Ontario, Canada
CEO: Humera Malik. She previously served as Executive Director for Quexor Group and Director of Product Management for Redknee.
Problem they solve: Typical industrial environments are data rich but information poor. A plant’s connected assets and dynamic processes can generate hundreds and thousands of data points every minute, but less than 10 percent of this data is used to derive insights or aid decision making.
Instead, decisions are made based on the personal experience of operators and/or by using outdated tools that are unable to handle voluminous, frequently changing data that comes from a variety of sources.
How they solve it: Canvass Analytics’ real-time AI data models identify trends to help operators understand the variables impacting their industrial processes. The Canvass AI Platform responds to changes in the data in real time, providing operations teams with the most up-to-date information so they can continuously adjust their operations to improve quality, reduce energy consumption and optimize asset health.
The Canvass AI Platform simplifies the challenge of rapidly processing large, complex volumes of data by using AI to automate the entire data-science process. The platform distils the millions of data points generated by industrial machines, sensors and operations systems, and identifies patterns and correlations hidden deep within the data to create new insights. These self-learning models adapt to new conditions in real-time ensuring that decisions by operations teams are made with the most accurate data possible.
Competitors include: GE Predix, IBM Watson, Uptake Noodle.ai
Customers include: At the time of publication, Canvass Analytics did not have any customers willing to go on the record.
Why they’re a hot startup to watch: As an early stage startup, Canvass Analytics has locked down two rounds of seed funding to pursue early adopters. The leadership team gained relevant experience at Quexor Group, Redknee, Bell Canada, NorthWest Energy and CHR Solutions. The company has built its platform to continuously ingest massive volumes of data in real-time, which its AI platform uses to improve and then automate operational processes.
Link: Falkonry
What they do: Provide machine-learning software for industrial operations
Year founded: 2012
Funding: $10.9 million. Falkonry closed a $4.6 million Series A in June 2018. Investors included Polaris Partners, Zetta Venture Partners, Presidio Ventures (Sumitomo) and Fortive.
Headquarters: Sunnyvale, Calif.
CEO: Nikunj Mehta. Prior to Falkonry, Mehta was the VP of Customer Success at C3 IoT. Before that, he was at Oracle, where he led the team that created the IndexedDB standard for databases that is embedded inside all modern browsers.
Problem they solve: To compete globally, industrial companies must improve the productivity of their operations and/or embrace new business models, and many industrial companies are turning to data analytics to drive that change.
The data generated in manufacturing or process operations, especially time-series data, is very rich in information that can provide actionable insights on the health of the production systems and the products created.
Machine learning is ideally suited for analyzing such massive amounts of data. However, hiring data-science consultants has proven inefficient, as they lack the subject matter expertise of the operations teams, and, as a result, such projects can take more than a year to see results.
How they solve it: Falkonry applies feature learning and machine learning to multivariate time-series data that is generated by the equipment and production systems in most discrete manufacturing and industrial process operations.
Given the high volume and number of signals, most industrial data goes unutilized in operations today. The Falkonry operational machine-learning system is able to discover hidden patterns in the data that cannot be observed by humans or traditional analytics. These patterns, in turn, provide insights to the operating state, and they identify conditions that precede undesired events to issue early warnings. Depending on the process being monitored, such early warnings may occur hours, days or even weeks in advance.
Falkonry says that its system acts like a “data scientist-in-a-box,” meaning no data scientists are required, and it can be quickly deployed by manufacturing engineers or process engineers. Customers start gaining actionable insights within three weeks of deploying the system, which could potentially save companies millions of dollars annually.
Competitors include: Cylance, Alluvium, Presenso, Seeq, Sight Machin and SparkCognition
Customers include: Toyota Industrial Equipment Manufacturing, Kawasaki Heavy Industries, Ternium, Ciner Resources and Energias de Portugal (EDP)
Why they’re a hot startup to watch: Falkonry has locked down nearly $11 million in funding. Founder and CEO Nikunj Mehta previously served as the VP of Customer Success at C3 IoT, a unicorn IoT startups. Add a customer list that includes the industrial divisions of Toyota and Kawasaki, and Falkonry was a no-brainer for this roundup.
Link: Interactor
What they do: Provide an edge-computing platforming that connects to any AI system
Year founded: 2014
Funding: The startup is backed by an undisclosed amount of funding from private equity (Telos Ventures) and governmental (National Science Foundation, Department of Homeland Security) sources.
Headquarters: Sunnyvale, Calif.
CEO: David Jung. Previously, he was an engineering leader at both Brocade and Cisco.
Problem they solve: As the IoT trend gains steam, the amount of data generated is becoming so large that AI processing at the edge is transitioning from a nice-to-have to a must-have IoT feature.
Processing data at the edge is difficult, however, because computing resources are constrained. This is why so many AI IoT vendors attempt to push data to the cloud first, which isn’t practical in many industrial applications.
How they solve it: Interactor’s IoT edge software enables companies to apply the latest AI technologies at the edge without any lengthy deployment effort. Interactor acts as small-footprint IoT gateway that sits between edge devices and the cloud. Interactor abstracts all of the components and microservices needed for technologies to interact and/or integrate with one another into a small (about 50MB) executable.
Using Interactor, developers and operators can easily integrate the AI of their choice and apply the intelligence closer to the devices for faster response times.
Interactor IoT edge software can be run on any IoT gateway or server, and it includes pre-packaged device configurations, security and authentication, messaging, device visibility, logging and error handling.
Competitors include: AWS Greengrass, Microsoft Azure Edge, EdgeX Foundry and PTC Kepware
Customers include: Cisco, Panasonic, Malaysian Government and MacroBlock
Why they’re a hot startup to watch: Interactor hasn’t released details of its funding, but its backers include a VC firm and the Department of Homeland Security. CEO David Jung previously led engineering teams at Brocade and Cisco, and customers include Cisco, Panasonic, and the Malaysian Government. Its deployment model of adding a small executable to an IoT gateway sidesteps WAN connectivity and cloud issues. Moreover, the AI can apply instant change/control/morph to any part of an industrial application on the fly.
What they do: Develop IIoT sensors and machine-learning softwareYear founded: 2014
Funding: The startup is backed by an undisclosed amount of seed funding from True Ventures, Felicis Ventures and unnamed investors.
Headquarters: San Jose, Calif.
CEO: Abhinav Khushraj. He previously led strategy and product for Nokia in North America
Problem they solve: Industrial companies waste resources on unplanned downtime. Many plants still deploy employees to walk around and collect vibration data every month, which is inefficient and often misses quickly developing problems that could force them to shut down equipment and/or suspend operations.
How they solve it: Petasense provides multi-parametric wireless sensors that communicate to cloud-based software, which then uses machine-learning algorithms to calculate a machine’s health score. The software predicts future problematic events, preventing equipment failure. If a problem is pressing, real-time alerts and diagnostics are sent to plant personnel.
Petasense also collects information on the relations within process data, so, for example, when operating conditions change, Petasense can see what the impact is on machine health.
Competitors include: GE, Emerson, SKF, Pruftechnik and Fluke
Customers include: Silicon Valley Power, JLL, Cushman & Wakefield and Stanford CEF
Why they’re a hot startup to watch: Petasense has secured seed funding, and, more importantly, has won over named customers in Silicon Valley Power, JLL and others. Co-founder and CEO Abhinav Khushraj previously led strategy and product for Nokia in North America. Petasense’s employs a dual hardware (multi-parametric wireless sensors) and software (machine-learning-based analytics backend and apps) approach. Whether or not a vendor needs the entire sensor-to-cloud-to-app stack remains to be seen, but Petasense certainly makes it easy for industrial customers to take the plunge with analytics at the edge.
What they do: Build AI tools that enable on-device, edge-based data collection and analytics
Year founded: 2013
Funding: $400,000 in seed funding
Headquarters: San Francisco, Calif
CEO: Noah Schwartz, who was previously the Assistant Chair of Neurobiology at Northwestern University
Problem they solve: Deep Learning promises to revolutionize many fields, but it has hard limits that, thus far, are throttling back its potential. To be effective, Deep Learning requires huge amounts of data, bleeding-edge hardware and a team of expensive data scientists to implement and manage the system.
“In terms of performance, Deep Learning models are very effective and powerful if you are solving a narrow problem, but once built, they are inflexible and fragile, and their inner workings are completely hidden from view,” noted Quorum AI CEO Noah Schwartz. Deep Learning also runs up against specific limitations when applied to IoT, particularly where data privacy, security and control are issues, and especially when the device must send data to a centralized cloud server for processing.
How they solve it: Quorum AI provides an AI engine and AI platform toolkit that help developers build portable AI for IoT edge devices. Developers can embed Quorum AI’s software directly on their devices or plug Quorum AI directly into their applications, enabling the devices to collect data, learn from that data and provide insights in real time.
The platform toolkit enables developers to wrap their AI-powered devices in special protocols, allowing them to communicate with one another in order to generate more complicated insights. For enterprises, the tools also enable the devices to optionally publish/subscribe to centralized models in order to bootstrap and/or standardize performance.
Quorum AI gives developers the option of eliminating dependency on the cloud, which alleviates the bandwidth, privacy and latency issues that are common when using traditional cloud-based AI methods.
Quorum’s AI tools not only help IoT devices learn and respond to events in real time, but also enable devices to become more personalized as they learn from their own unique contexts, rather than seeking guidance from a universal model stored in the cloud. Quorum AI devices can engage in peer-to-peer learning rather than having to transfer data to a centralized AI server or data warehouse and then using it to generate a model.
Competitors include: IBM Watson, Google Cloud, Microsoft Azure (via the acquisition of Bons.ai), AWS and Qualcomm
Customers include: At the time of publication, Quorum AI did not have any customers willing to go on the record.
Why they’re a hot startup to watch: Quorum AI has $400,000 in seed funding and no named customers to date, so the clock is ticking for them. CEO Noah Schwartz has an intriguing background, coming to this space from Northwestern University, where he was the assistant chair of neurobiology. Quorum provides a toolkit that allows developers to embed an AI engine directly into their devices and enables devices to communicate amongst themselves in a personal area network, with each device communicating with and learning from other nodes in the network.
What they do: Develop AI software that analyzes IoT data to help manufacturers solve productivity and quality challenges
Year founded: 2012
Funding: About $50 million. Investors include GE Ventures, E.ON Group, Mitsui, Jump Capital, Mercury Fund, Draper Nexus, Pritzker Group and TekFen Ventures.
Headquarters: San Francisco, Calif.
CEO: Jon Sobel. He formerly served as General Counsel for Tesla and Yahoo.
Problem they solve: Manufacturers struggle to make optimum decisions quickly. When dealing with problems that emerge on the plant floor, any indecision or delay in decision making could be costly.
In manufacturing, data variety (due to thousands of sources) is far greater than in other IoT use cases, and according to research from Morgan Stanley, the sheer quantity of data is also larger than anywhere else. Traditional analytics tools can’t cope with either the variety or volume.
How they solve it: Sight Machine software uses canonical data models and AI to ingest, integrate, and map massive amounts of heterogeneous data into operational models. The canonical data models represent any machine, line, facility, supplier, part or batch that the manufacturer specifies. Once modeled, data is then systematically and continuously analyzed.
By standardizing the manufacturing models and following a data-first approach to decision making, Sight Machine enables manufacturers to automate data ingestion in a rapid, highly repeatable manner. The standardized model allows manufacturers to create downstream applications that immediately leverage the modeled data.
Analytical techniques include advanced inferential statistics, machine learning and AI, all of which are applied to generate manufacturing-specific insights. Within its platform, Sight Machine analyzes and visualizes data, so results can be viewed via a browser on any connected device.
Competitors include: PTC/ThingWorx, C3IoT, Uptake, IBM Watson, Seeq and Tulip
Customers include: GE, WestRock, Nagase, Inteva, Fiat Chrysler, Komatsu and Fujitsu
Why they’re a hot startup to watch: Sight Machine has the deepest pockets in this roundup, backed by $50 million in VC funding. CEO and co-founder Jon Sobel was previously with Tesla and Yahoo, while co-founder and CTO Nathan Oostendorp and co-founder and Chief Data Scientist Kurt DeMaagd previously co-founded Slashdot.org. Finally, the customers Sight Machine has accumulated are impressive, including GE, Fiat Chrysler, and Fujitsu.
Link: SWIM.AI
What they do: Develop edge intelligence software
Year founded: 2015
Funding: $10+ million. In July 2018, SWIM.AI announced that it had closed a $10 million Series B funding round led by Cambridge Innovation Capital plc (CIC), with a strategic investment from Arm, and further participation from existing investors Silver Creek Ventures and Harris Barton Asset Management.
Headquarters: San Jose, Calif.
CEO: Rusty Cumpston, who previously co-founded Sensity Systems, which was acquired by Verizon in 2016. Before that, he served as CEO for CloudShield Technologies.
Problem they solve: As edge devices and data volumes grow exponentially, enterprises are swamped with data, so much so that it’s difficult to extract value from the data flood. That data, however, holds insights that could transform production, boost efficiency and streamline operations.
Being able to analyze data and predict system performance on the fly puts enterprises ahead of the game, but legacy analytics tools can’t manage the volumes and disparate types of data edge devices generate.
An even trickier aspect of the problem is that in order to improve many processes or to avoid downtime, certain events must be interpreted in real time, or the value vanishes for anything other than historical trend analysis.
How they solve it: SWIM.AI has developed an edge software stack, SWIM EDX, that is designed to efficiently process streaming edge data to quickly generate business insights.
SWIM EDX enables businesses to reduce, analyze, learn and predict from what SWIM.AI refers to as “gray edge data.” SWIM EDX combines edge computing, machine learning and self-training through digital twins, all running locally on edge devices working in a mesh/fabric architecture.
With SWIM, these insights can be shared automatically with other applications, and industrial decision makers can access them on the fly from any device with a browser.
Competitors include: Microsoft Intelligent Edge
Customers include: Trafficware
Why they’re a hot startup to watch: SWIM.AI is backed by more than $10 million in funding. CEO Rusty Cumpston previously co-founded Sensity Systems, which was acquired by Verizon in 2016. Trafficware is a named customer, and SWIM.AIuses what it calls “digital twins” – self-training digital twins, that is – that model the behavior of real-world devices and systems in order to enable predictive analytics from the edge.
