New Research from Tractica on Enterprise AI Shows Deployments Getting Results


A new study on how AI is being used in large organizations by Tractica, a market research company focused on human interaction with technology, analyzes nearly 300 different use cases to create its market forecast.

The report finds that the number of proof-of-concept and pilot programs continues to grow, and larger scale commercial deployments of AI technology are being publicized by enterprise organizations around the globe.

Tractica forecasts that worldwide revenue from deployments of AI software, hardware, and services will increase from $14.9 billion in 2017 to $23.6 billion in 2018, a year-over-year increase of 58%.  The market intelligence firm’s analysis finds that the top 10 industries for adoption of enterprise AI are as follows:

  1. Business Services
  2. Government
  3. Healthcare
  4. Automotive
  5. Advertising
  6. Retail
  7. Finance
  8. Aerospace
  9. Media & Entertainment
  10. Telecommunications

“Results from early enterprise AI deployments are quite strong in improving operational efficiencies, reducing expenses and enhancing the resolution of data analytics,” said principal analyst Keith Kirkpatrick of Tractica. “Much of the success of AI is due to the fact that most tasks currently delegated to AI technology are data-driven and therefore easily measured or benchmarked. When AI technology is deployed, even during a small pilot program, the benefits can quickly be demonstrated and proven by looking at the performance data.”

In the longer term, Kirkpatrick sees AI market growth moving beyond data analytics use cases to include applications for computer vision and natural language processing capabilities.

Not every use case described by Tractica may prove out, but executives in almost every industry need to consider the potential impact AI will have on their respective business models, bottom lines, and careers.

Companies that successfully use AI will be reticent to publicize results. They may intentionally or unintentionally sandbag their competitors and make marketing difficult for their vendors.

It is clear that organizations are focused on integrating AI, based on the number of pilot programs, proof-of-concept demonstrations, and commercial deployments of AI technology already being publicized by enterprise customers around the globe. Much of the success of AI is due to the fact that most tasks currently delegated to AI technology are data-driven and, therefore, easily measured or benchmarked. When deploying AI technology, even during a small pilot program, the benefits can quickly be demonstrated and proven by looking at the performance data.

A significant barrier to adoption is the requirement that most AI systems require statistically valid, clean, and accurate data. As with any information system, bad data will result in bad assumptions and predictions, but with AI technologies, such as ML and DL, clean data will be even more critical. Although there is some data, such as images of cats, that is freely abundant on the internet, other data is much rarer and harder to find, and will limit how and where AI can be applied.

As primitive a model as they are, Deep Learning systems do represent a radical departure from past computer systems. The data that they are given determines their structure, not the instructions from a programmer. This turns traditional design systems on their head. In a conventional system, a human instructs a program what to do with data. In a DL system, the data instructs the program about what humans should do.

Here are Selected excerpts from the report,  “Artificial Intelligence for Enterprise Applications”:

Shortage of Talent: Google, Facebook, and other big tech firms are paying engineers proficient in DL more than $400,000 a year. Those with the most experience can earn incomes over seven figures. Companies seeking to compete directly with the technology giants, such as in the automotive sector, are going to have to pay top dollar to attract this talent.

Resistance: Many upper-management types resist incorporating AI, given the challenges related to defining the proper role(s) for AI within the organization, the time and effort related to selecting vendors to help implement, test, and evaluate proof-of-concept projects, and the time and cost issues involved with department or company-wide AI programs.

Summary of Barriers: AI projects face some substantial barriers to commercialization, because they are:

  • Easy to oversell
  • Hard to understand
  • Unusually controversial
  • Subject to many limits in technology
  • Challenged by adapting to change

Yet, the projects will move forward because they are producing tangible results.

Finally, another barrier  to accepting AI has much to do with corporate culture and people. By its very nature, AI likely will replace many job functions that are often deemed repetitive, lower-value, or, in some cases, simply more efficiently handled by a machine than a human. There is often significant opposition to incorporating new technology within an organization from those who might either be replaced by that technology, or those who may be forced to learn new skills or take different types of positions within the organization. Moreover, incorporating change of any type is often met with resistance, simply because it upsets the status quo within an organization.

Here are some example use case projects and companies involved:

Drone Collisions: Neurala, a Boston-based company specializing in AI, is tackling the problem of drone collisions with the help of DL technology. The company trained its software by feeding it video images of potential collisions from the Microsoft Flight Simulator. Neurala’s software notifies drone users and operators whenever it recognizes similar, real-time images from a single camera mounted on the drone.

This technology is being applied to support a variety of use cases, including image detection, segmentation, and classification, as well as to support navigation, search, change monitoring, and research, and to identify, find, and track specific types of objects.

Predictive Maintenance: Airbus is working with EasyJet to provide predictive maintenance capabilities for its fleet of more than 200 aircraft. Airbus is using EasyJet fleet data in conjunction with data from other carriers to improve prognostic tools and predict when parts need to be replaced, ultimately helping carriers like EasyJet improve fleet performance and reduce maintenance costs. Microsoft recently announced a product suite designed to monitor aircraft and predict the remaining useful life of aircraft engine components, based on analyzing large public data sets from past aircraft engine life performances.

The ability to predict failures before they happen and systematically address them helps increase safety and reduce mishaps, delays, and costs associated with broader downtime in the event issues are not preemptively identified and addressed.

Livestock Management: BovControl, a 5-year-old startup, aims to create the “internet of cows.” Farmers enter cow data (e.g., weight, birthdate, medication, vaccinations) and connect the app to the monitoring device they use to track the animals (e.g., smart collars, ear tags, etc.). Then the app uses AI to analyze and make predictions about each cow, predicting due dates for pregnant cows, milk production or anomalies in production, medication, vaccination needs, etc. The company is also expanding features in meat sourcing and provenance, compliance adherence, export, inventory, and integration with other farm management systems.

Auto On-road Customer Service: From Tesla to Ford to Toyota, just about every auto manufacturer is working on developing channel strategies for delivering customer service to drivers and passengers. While driving or riding in connected cars, will drivers welcome customer support from manufacturers, hyper-local marketing from brands, or personalized alerts via context-aware virtual assistants? Automated on-road customer service will also likely tie in with personalized services available in cars. A key question is who will support or co-develop which business models. Will the answer be the manufacturer, dealership, network service provider (NSP), insurance provider, advertiser, technology giant, or city? The market is too nascent to determine a clear winner, although manufacturers and NSPs are collaborating closely. The market is still too early to determine which model will be most successful.

Building Automation: PointGrab is a company that provides sensing hardware and software that use DL and CV embedded into IoT devices for edge processing. Specifically, the company uses object tracking algorithms for background modeling, novelty detection, motion estimation, and nonrigid object detection, coupled with proprietary ML classifiers and training pipelines to support learning and modeling of office/work space management, staff planning, retail analytics, and occupant safety to track movement of building occupants, and to drive energy savings, smarter allocation, and cost savings for commercial environments.

Chatbot-based Ecommerce: Brands including Whole Foods, Pizza Hut, Disney, and The North Face, have developed their own chatbots, available on their own mobile apps, websites, short messaging service (SMS), and messenger apps alike. The North Face, for instance, built an expert personal shopper (XPS) bot that helps match specific customer needs with specific products. The company recently reported that customer engagements with the bot averaged about 2 minutes in length and the platform had a 60% CTR for product recommendations.

Travel Services: WayBlazer is developing chatbots as voice-enabled agents to aid travel agents with super specific results based on inquiry. It uses supervised and unsupervised DL, NLP, and image processing to mine vast amounts of data, running sentiment analysis across text reviews, tagging images, activity content, and other data. From an end-user perspective, a query such as “we want a romantic weekend getaway” is served by inferring properties that have been tagged and force ranked as “romantic,” then tailoring recommendations for the individual based on IP address, user ID, and geographic coordinates.

Energy: PowerScout is using DL to analyze satellite data to detect and determine homes that would be likely candidates for solar panels due to their positioning and exposure to light. This helps optimize sales and marketing costs associated with targeting the right potential buyers. The company has trained two neural networks to determine: 1) whether a home already has solar panels (or not), and 2) whether nearby vegetation would hamper installation or energy generation efforts. It is also developing an e-commerce site in order to use this data to let users run feasibility and estimated value and returns on solar panel installation for their homes. Based on “solar worthiness,” the service then matches potential buyers with local installers, and offers tailored financing options. In the future, PowerScout hopes to use this data to optimize community solar sales by suggesting installations wherein multiple residents could take advantage.

Find more information about the report here:  “Artificial Intelligence for Enterprise Applications”.

— By John P. Desmond, AI Trends Editor