Embedded AI for Analytics: What You Should Consider

For those B2B SaaS companies reading this article: how many of you have been asked for an AI strategy?

Not all SaaS have sufficient AI expertise to build these types of data products. In most cases, the best bet is to find a third-party provider who understands the capability and nuances that surround it, and work with them to embed AI into your product. Embedded AI for analytics is an exciting use case, but there are a heap of considerations one has to make before committing to this capability and a particular provider.

In this article, we cover the key considerations for those thinking about AI for analytics.

‍

1. Data Privacy and Security

‍

Many organizations don’t want to pass their data to LLM providers, like OpenAI. They understand it’s necessary to pass their database schema, but there’s often no need to pass the body of data.

With Vizzly, you can run the query engine (semantic layer) on your infrastructure. Nor Vizzly or the LLM provider will process customer data. Of course, the sequential question must be: ‘how can you optimize accuracy without the corpus of data?’. Let’s dive into that!

‍

2. Utilizing Vizzly's Model without Data Transfer

‍

For the model to be effective without passing the body of data to the LLM provider, one needs to supplement their data schema with metadata describing what each dataset and each field is. There is some (and unavoidable) manual intervention required here, but the value is proportional to the input. Ultimately, you need high accuracy for the AI to elicit genuine user value. Read more about Vizzly’s query engine here.

‍

3. Conversational User Experience

‍

With an embedded AI provider, yes, you won’t need to build the APIs or train the model, but you also won’t need to think about constructing and optimizing the frontend. Conversational user experience (UX) is a new area of design that comes with a variety of nuance.

For example, with Vizzly, the user receives:

• Dynamic suggestions to help understand what's possible with the AI
• Escape hatches should they want to revert back to a traditional interface
• Opportunities to send feedback to the Vizzly model

Historically, NLP in analytics has been used to generate one-off queries; with Vizzly AI, the user can iterate on the response in the two-way conversation.

‍

4. Seamless Integration

‍

However you decide to include an AI capability for analytics inside your SaaS app, it should be native to your product experience and interface. It shouldn’t feel like a feature tacked on to the end of your product in an effort to keep up with technology trends.

In Vizzly, our first application of AI is to provide a conversational interface alternative to our data panel, where the user is expected to build the query for the data view. They can toggle seamlessly between both interfaces based on their preference. Of course, Custom Reporting must be enabled to expose this functionality.

‍

5. Continuous Improvement

‍

This consideration applies to any non-core competency: by outsourcing to a third-party, of course, you can accelerate time-to-value, but you’re safe in knowing the vendor is going to continuously build on what they have today. Vizzly is always releasing new features. If you’re interested in staying up to date with the team and product, be sure to sign up to our newsletter.

‍

Conclusion

‍

Embedding AI for analytics requires thoughtful consideration. Partnering with Vizzly streamlines integration, addressing data privacy while enhancing model efficacy. Vizzly's conversational UX ensures a cohesive user experience, integrating AI seamlessly. Continuous improvement, seen in Vizzly's enhancements, remains crucial. Stay updated; join our transformative journey in AI-enabled analytics or reach out to us for early access!