top of page
< Back
Forbes - Why The Software Technology Company Of The Future Might Not Sell Software

Forbes - Why The Software Technology Company Of The Future Might Not Sell Software

This article was originally published by Forbes. For the full article, click here.


In the digital economy, data abounds. IDC’s Global Datasphere Forecast predicts that the digital economy will create more data in the next three years than was generated in the last 30 years combined. There’s enormous power in all that data, but it takes a great deal of work to unlock it. Data is complicated, and in order to make data make sense, someone needs to clean it, organize it, interpret it and put it to use. 


As Microsoft CEO Satya Nadella once said, “Every company is a software company.” That’s in part because software is the principal tool for harnessing data and putting it to work. But the value of software has changed. 


In a world of open-source software, software code is no longer a scarce good, and businesses don’t see the value proposition in proprietary lock-in to software platforms. Rather, the software technology that can deliver real value today will offer bespoke solutions customized to each business’s unique data and business needs. Accordingly, the software technology company of the future will not sell you packaged code anymore; instead, it will sell data-centric computing and access to platforms of customizable machine learning models. 


Revolutions like this have struck software before. One of the biggest changes of the last several years was the tectonic shift from licensed software products to software as a service (SaaS). Using advances in cloud-based computing, SaaS allowed companies to outsource their computing needs and paved the way for a new software business model centered on selling service architecture — a place in the cloud to get work done. And with great success: SaaS has grown tremendously in recent years, and Gartner predicts that SaaS market revenue will rise by $120 billion in 2021.


But I believe software will soon experience another major shift in the form of machine learning as a service (MLaaS). Where SaaS sold software as a cloud-based service, MLaaS will sell software as a tool for data analytics. Winning technology companies will no longer sell software with a specific, hard-coded functionality; instead, they will sell access to flexible machine learning libraries that can analyze datasets for all known and unknown future uses. 

The advent of MLaaS will change the technology marketplace. Software traditionally has very high margins because once built, a software product can be sold millions of times. Microsoft Word works the same way for countless users. But machine learning models deliver customizable solutions, and MLaaS crafts bespoke solutions to unique use cases.


Like custom-tailored clothing, these bespoke solutions provide a better product but take more time to develop. To move from static software to the ML world, users need to feed training data to models and validate learning before deploying the technology into production. ML deployed in the banking industry today, for example, largely consists of teams of data scientists experimenting with ML models in an offline “sandbox.” ML has not yet impacted industry operations. 


The best technologies don’t always win massive market share, and MLaaS will face go-to-market obstacles. Yesterday’s software companies sold to IT departments; today’s SaaS companies sell to the IT as well as the operational business lines. MLaaS will need the front-line — and often nontechnical — users to become the purchasers of the technology, completing the paradigm shift to technology for users instead of technology for the IT department. Likewise, technology companies will need to develop deep partnerships with business lines and get comfortable working together to generate, clean and process data while training and validating custom-tuned models. 


When successful, this transition to MLaaS can bring enormous rewards. This is especially clear in my field of anti-money-laundering (AML) and financial crime detection. With ML, banks can compile activity and transaction data from all their customers, augment these datasets with contextual data from open source and nontraditional sources, train models to distinguish between legitimate and illicit behaviors, and collaborate across the industry and with law enforcement, all without sharing data and protecting privacy. 


Investors will likely move to the MLaaS business because these businesses can operate with good and defensible margins. Machine learning models can also transfer efficiently — an area of ML coming to market. My companies, for example, have created two technologies that use federated learning and transfer learning methods to enable collaboration while protecting sensitive data and IP. While it may take longer to train the first instance of an ML model, once a model is sufficiently adept at classifying data, new users can easily adopt it and realize efficiencies. For example, a fraud model trained at Bank A can be moved to Bank B and, with minimal modification, meet the bespoke requirements at B that did not exist at A. 


We are at the dawn of a new era of technology where data and software abound and small teams can tune machine learning algorithms for bespoke problems. In this new era, the most efficient, defensible and successful technology company isn’t the software company of old, but the MLaaS company of the future. 

Gary M. Shiffman, PhD, is an economist working to solve problems related to human violence. A Gulf War veteran and former Senate National Security Advisor, Chief of Staff at US Customs and Border Protection, DARPA Principal Investigator, and Georgetown University professor, he founded two technology companies, Giant Oak, Inc, and Consilient, Inc. He is the author of The Economics of Violence (2020), and his essays have appeared in media outlets such as The Hill, the Wall Street Journal, USA Today, TechCrunch, and others.

bottom of page