Past posts have defined AI enabled SaaS. This post outlines the three types of AI enabled SaaS currently in use and some of the implications of these to your SaaS business and your customers.
Off the Shelf
The first type of AI enablement, Off the Shelf, is when an already trained and tested AI algorithm or function, typically from a cloud provider, is incorporated into a SaaS application. These algorithms are typically used via an API and little to no knowledge of AI is required to use them. The algorithms are retrained, tested and maintained by the providers. Off the Shelf algorithms are not necessarily simple, in fact, they can be very complex, and they tend to have broad use across various industries and geographies. The availability of AI algorithms suitable for this type of use is very likely to change over time but some examples of this are text to speech, speech to text (transcription), language translation, identification of certain items in photos and identification of common events in videos.
For Off the Shelf algorithms there really aren’t any special implications or requirements for a SaaS provider beyond the normal customer enablement and software maintenance requirements. Since the functionality is regularly updated, retrained and retested by the provider there are no special application maintenance requirements. And since there is not any customer unique or industry unique data used there is no special enablement, implementation maintenance or AI knowledge required. This is the simplest type of AI enabled SaaS to implement and support. Most of the AI functionality implemented in this manner will become expected baseline functionality very quickly.
The second type of AI enabled SaaS is Application Unique. One type of Application Unique AI enablement can be done by customizing standard AI functionality, which typically comes from cloud providers, using application or industry unique data. We’ll call this type of AI enablement Application Unique – Customized. Examples of this type of AI enablement include natural language processing with industry or application specific vocabulary and photo or video recognition with specific industry or application photo requirements.
There is also Application Unique – Specialized AI enablement of a SaaS application. In this type of enablement there is an AI algorithm created by the SaaS provider using application and industry unique data. To use application data across all customers will require the appropriate legal permission to use the data for this purpose. The examples of this type of enablement are numerous but include use cases such as payment fraud detection, defect detection where the definition of a defect is industry defined, or process enablement and/or measurement where the process or terminology is industry standard.
With Application Unique algorithms it is the responsibility of the SaaS provider to update the AI technology and algorithm, retrain the model, retest the model and redeploy the model. At a conceptual level you can think of this in the same way that you would all software changes however data science and MLOPS tools and expertise will required to execute the changes. In addition some special on-boarding requirements will be required.
The third type of AI enablement of SaaS is Customer Unique. In this case the AI algorithm is trained and tested using unique customer data for each customer and the timetable for deployment of the model or the retrained model is unique per customer. This type of enablement can be Customer Unique – Customized and use standard algorithms from cloud providers but use data that is unique to each customer. Examples of this type of enablement include Natural Language Processing with customer unique vocabulary and photo and video recognition functionality with customer unique image requirements. Other examples include using tools for implementing standard structures such as chatbots or various types of personalization. A SaaS application incorporating Amazon Forecast is an example of this type of enablement, the algorithm is from Amazon but it is trained and customized using customer unique data.
Customer Unique – Specialized is when customer unique data is used to create customer specific AI algorithms without using algorithms provided by someone like a cloud provider. In this situation the tools are provided by the SaaS provider and the data is customer unique. Salesforce’ Einstein Forecasting would be an example of this with Salesforce being the SaaS provider. This type of AI enablement requires both maintaining the tools used by the customer as part of the application and the development and maintenance of the customer unique model. This is the most complex situation and requires all of the identified expertise and tools for on-boarding AI Enabled SaaS applications . It also likely the strongest competitive position for AI enabled SaaS applications providing that the business is large enough to create and support the AI algorithm development tools needed and have an onboarding and maintenance process which scales appropriately for a large number of customers.
It’s important to think carefully about the type of AI enablement you have or are considering since beyond the product and competitive positions, the business and customer success aspects of each type of enablement are quite different and require different levels of investment.