The dataset snowball effect: How AI is changing SaaS
摘要：Above: Cloud computing conceptFrom machine learning-powered fraud defense on Shopify to Salesforce’s Einstein, over the past couple of years, SaaS industry leaders have invested heavily in artificial intelligence R&D and have rapidly acquired AI companies to give themselves a lead over the competition.
Above: Cloud computing concept
From machine learning-powered fraud defense on Shopify to Salesforce’s Einstein, over the past couple of years, SaaS industry leaders have invested heavily in artificial intelligence R&D and have rapidly acquired AI companies to give themselves a lead over the competition.
Thanks to cloud computing services democratizing access to AI, we may be on the cusp of a new age when emerging SaaS providers begin to roll out AI applications that really solve problems for consumers.
I spoke to a number of experts at this year’s SAAS NORTH conference in Ottawa, Canada about the evolving nature of AI in the SaaS industry. According to their insights, here’s what the future might hold:
First wave: Bigger players have a head start
The traditional SaaS model is based on rolling monthly subscriptions, meaning SaaS companies need to constantly improve and nurture their customer relationships to keep clients returning month by month.
Leo Lax, founder of Ottawa-based SaaS accelerator L-Spark, said, “AI is helping to reduce the manual labor that was involved in the building of the customer relationship and allowing vendors of SaaS to interface with customers in a more meaningful way.”
Over the past few years, it has only really been the larger, better-funded SaaS giants that have had the resources to hire the right talent and invest in meaningful AI R&D. But bulging bank balances alone are not enough to implement useful AI applications. The main ingredient is data, and lots of it.
Established SaaS companies with their own platforms have a head start. One of the biggest hurdles to training a machine learning system is getting access to large enough datasets. David Lennie, senior vice president of data and analytics at Shopify, explained:
“The biggest value comes from getting the largest samples as quickly as possible, and that’s more possible when you have a big network of people all doing the same kind of thing with you. SaaS companies usually have one type of solution which they are offering, and access to a market of users who can rapidly give them more data to do this even more effectively.”
Lennie argues that SaaS tools that focus on solving one particular problem, rather than being an “all-in-one” solution, do a better job of creating the right kind of data to train machine learning applications. Once companies have access to these huge datasets of “clean” data from millions of users around the world, they can start solving problems. However, Kerry Liu, chief executive officer and cofounder of Rubikloud, argues that until this point, the best success cases with AI have been internal.
“Whether it be Google optimizing their basic search, or Salesforce using Einstein to help identify the best use cases for their own account and sales managers, most of the successful applications to date have been to improve internal efficiency and internal product development,” Liu said.
However, while to date most of the applications by leading companies have been internal, they are moving in the right direction. Experts suggest that AI will soon see improved automation, personalization, voice input, and security for users.
Second wave: Cloud computing levels the playing field
Until recently, few emerging players in the SaaS industry have really used advanced AI applications. Ardi Iranmanesh, cofounder of Affinio, told me, “AI has been massively overused for marketing purposes. Many smaller companies are using basic applications like chatbots or linear regression and calling themselves AI startups.”
However, over the past few years, access to cloud computing services like AWS, Microsoft Azure, Google Cloud, and Oracle opened the door to smaller companies using more advanced applications like machine learning by harnessing “AI as a service” cloud tools.
That said, the real foot up offered by these cloud services is on the underlying compute level. Cloud computing services have changed the landscape, enabling smaller players the compute power necessary to build meaningful AI applications and deploy anywhere in the world, without needing to own any hardware or worry about data security.
This more inclusive “second stage” of SaaS AI evolution has led to the creation of a number of specialized AI SaaS companies focused on solving more niche problems, rather than the more general productivity or communication tasks solved by bigger players.
As Igor Faletski, cofounder and CEO of Mobify, pointed out, “AI has been around for awhile. What is new is that it is really being opened up to developers, and more and more small startups have access to it.”
Companies such as Beanworks and Mindbridge Analytics focus on emerging verticals and automating “white collar” tasks such as auditing and accounting, which to date have been largely ignored by SaaS giants.
Alex Corneglio, chief technology officer at EnergyX Solutions, confirms this trend. “I see a whole new level of niche products that can be tailored to very specific market personas — imagine all of the subtle and nuanced qualities that attract us to people now becoming embedded into products and services. It’s probably all coming at an ever-accelerating pace,” he said.
However, the biggest challenge of developing meaningful AI applications is getting access to proprietary datasets. In a breakout session at SAAS NORTH, David Lennie emphasized that the value of AI does not lie solely in the power of algorithms, but also in the datasets companies have access to. He cautioned that companies should fully understand how they want to use data before diving into building AI-powered solutions.
Lennie suggested that in order to overcome the AI dataset “chicken and egg” conundrum, emerging AI companies will have to share more data, and also partner with legacy companies who “have a lot of data but no understanding of how to go work with that. Maybe you can do that work for them and in exchange give some of that data.”
Iranmanesh foresees more legacy companies opening up their data to AI startups. Referencing companies like Mastercard and Visa opening up data, he argued, “While PII data laws will always be a consideration, companies are always looking to add to their bottom line, and simply storing data doesn’t solve problems.”
However, Eli Fathi, CEO of Mindbridge Analytics, countered that when dealing with tasks like auditing, it is possible to train algorithms with public data and small data samples from individual companies. Catherine Dahl, CEO of Beanworks, said accounting tasks are very repetitive, which makes them perfectly suited for training machine learning algorithms.
Third wave: Exponential growth of AI in SaaS industry
For more established SaaS companies that already collect huge flows of user and operational data, thus increasing the intelligence of their machine learning systems exponentially, we are likely to see more of a focus on solving real enterprise problems in the near future.
Forrester predicts in 2018 SaaS giants will increasingly compete at the platform level, running parts of their services on cloud computing services to deal with the increased demands for application customization and more advanced AI applications that automate a range of core business functions.
At the same time, as AI adoption increases across industries, smaller, more specialized players will have access to more clients, and thus more data sets to hone their AI. Lennie suggested that if smaller companies focus on solving specific problems one by one, this will give them the ability to move on to adjacent problems and develop their platforms to the level of SaaS giants. He said, “You can rinse and repeat that model. If you rinse and repeat enough times, you will end up with a very developed platform.”
Faletski predicted that the next wave of widespread AI adoption for SaaS tools will be pushed forward by companies like Amazon, Microsoft, and Google investing heavily into hard AI R&D and battling to build ecosystems on their platforms and become the biggest “AI as a service” provider. This will further open the door to smaller players using cloud AI applications or developing their own algorithms and taking advantage of the incredible scale offered by cloud services.
Liu agreed that “the Big Five understand that the more AI applications there are on the market, the more cloud compute options there are for them. Large-scale tech companies are absolutely incentivized for businesses of all shapes and sizes to adopt AI as a solution to enterprise problems, because if they do that, they can deploy that data and those applications on an elastic compute platform which can scale infinitely.”
If the great minds of the SaaS world are to be trusted, it seems that the snowball effect of AI evolution in software companies is well underway.
SaaS giants have developed smart platforms that are growing in power exponentially, and cloud services have leveled the playing field for smaller niche players. Liu argues that people underestimate the speed of acceleration. While some predict it will take 10-15 years for all Fortune 500/1000 companies to adopt AI SaaS products into their core business functions, he predicted it will happen within the next 5.
Craig Corbett is chairman and editor of one of Eastern Europe’s largest English language tech publications, 150Sec.