Next Era: Enterprise AI Applications

10 Jul Next Era: Enterprise AI Applications

I’ve been wondering for a while what might be next for enterprise software. Whether a small private or large public company, where should you invest our time and money?

Maybe looking into the past can give us some guidance. Enterprise software has gone through three distinct eras. In the 1st era, infrastructure software companies emerged like Microsoft and Oracle, which focused on programmers. Software developers used Microsoft Visual Basic and the Oracle database to build custom workflow applications for the enterprise throughout the 90s. By the late 90s the 2nd era of enterprise software began with the creation of packaged on-premises enterprise workflow application. Companies emerged including PeopleSoft, Siebel, SAP and Oracle. These applications focused on automating key workflows like order-to-cash, purchase-to-pay or hire-to-fire. Enterprises didn’t need to hire programmers to develop these workflow applications, they only needed to buy them, implement and manage them. The 3rd era began in the 2000s with the delivery of packaged workflow applications as a cloud service. Examples abound including Salesforce, Workday, Blackbaud and ServiceNow. This 3rd era eliminated the need for the enterprise to hire operations people to manage the applications and has accelerated the adoption of packaged enterprise workflow applications. While you could still hire programmers to write a CRM application, and operations people to manage it, why would you?

Let’s now switch our attention to analytics, which is not focused on automating a process, but instead on learning from the data to discover deeper insights, make predictions, or generate recommendations. Analytics has been populated with companies specializing in the management of the data (e.g. MongoDB, Teradata, Splunk, Cloudera, Snowflake, Azure SQL, Google Big Query, Amazon RedShift ); companies dedicated to providing tools for developers or business analysts (e.g., SAS, Tableau, Qlik and Pivotal) as well as software for data engineers including formerly public companies such as Mulesoft (acquired by Salesforce) and Informatica (acquired by Permira).

Furthermore, thanks to the innovations in the consumer Internet e.g., Facebook facial recognition, Google Translate, Amazon Alexa, there are now 100s of open source software and cloud services available which provide a wide array of AI and analytic infrastructure software building blocks. For those interested in geeking out, here is a brief introduction. Some of this technology will be dramatically lower cost. Consider today for about $1000 I can get 1,000 servers for 48 hours to go thru a training cycle to build a machine learning model.

I’m going to use the label AI to refer to the entire spectrum of analytic infrastructure technology, and also because it sounds cooler. Today we are largely in the 1st era. The software industry is providing AI infrastructure software and requiring the enterprise to hire the programmers, ML experts to build the application as well as dev ops people to manage the deployment. This is nearly the same as the 1st era of enterprise workflow software.

If we’re to follow the same sequence as workflow applications we need to move beyond the 1st era focused on developers and start building enterprise AI applications.

So what is an enterprise AI application?

Enterprise AI applications serve the worker not the software developer or business analysts. The worker might be an fraud detection specialist, a pediatric cardiologist or a construction site manager.

Enterprise AI applications have millennial UIs and are built for mobile devices, augmented reality and voice interaction.

Enterprise AI applications use historical data. Most enterprise workflow applications eliminate data once the workflow or the transaction completes.

Enterprise AI applications use lots of data. Jeff Dean has taught us with more data and more compute we can achieve near linear accuracy improvements.

Enterprise AI applications use many heterogeneous data sources inside and outside the enterprise to discover deeper insights, make predictions, or generate recommendations and learn from experience.

A good example of a consumer AI application is Google Search. It’s an application focused on the worker, not the developer, with a millennial UI and uses many heterogeneous data sources. Open the hood and you’ll see a ton of infrastructure software technology inside. So what are the challenges of building enterprise AI applications?

1. The nice thing about transactional or workflow applications is the processes they automate are well defined, and follow some standards. Thus, there is a finite universe of these apps. Enterprise AI applications will be much more diverse and serve workers as different as the service specialist for a combine-harvester, a radiologist or the manager of an off shore oil drilling rig.
2. The application development teams will be staffed differently. Teams will have a range of expertise including business analysts, domain specialists, data scientists, data engineers, devops specialist and programmers. With such a wide array of cloud-based software even programming will look different.
3. Finally the development of these analytic applications will require a different methodology than was used to build workflow application. In workflow applications we can judge whether the software worked correctly or not. In enterprise AI applications we’ll have to learn the definition of a ROC curve and determine what level of false-positives and false negatives we’re willing to tolerate.

Some companies are emerging to serve the developer including Teradata and C3 as well as the compute & storage cloud service providers, Microsoft, Google and Amazon. While there is plenty of room for creating custom enterprise AI applications, the true beginning of the next era will be the emergence of packaged AI applications. There are beginning to be some examples. Visier, founded by John Schwartz, the former CEO of Business Objects, has built a packaged applications focused on the HR worker. Yotascale has chosen to focus on the IT worker who is managing complex cloud infrastructure. Welline built a packaged enterprise AI application for the petro-technical engineers in the oil & gas industry using the Maanaplatform. Lecida, founded by some of my former Stanford students, is delivering a collaborative intelligence application for workers who manage industrial (construction, pharma, chemical, utility..) machines. They are using AI technology to make machines smart enough to “talk” with human experts, when they need to. Those models are built in less than 48 hours using a ton of software technology.

In order for data to be the new oil, we need to begin the next era and start building custom or packaged enterprise AI applications. These applications serve the worker not the software developer or business analysts. The worker might be a reliability engineer, a pediatric endocrinologist or a building manager. Enterprise AI applications will have millennial UIs built for mobile devices, augmented reality and voice. And these applications will use the oceans of data coming from both the Internet of People and the Internet of Things to discover deeper insights, make predictions, or generate recommendations. We need to move beyond infrastructure to applications.