I recently watched the movie Air. I absolutely loved it.
Note: if you don’t want spoilers, you may want to skip the next two paragraphs.
Air is a story chronicling how Nike, the underdog in those days, steals Michael Jordan away from Adidas and Converse. With the cards stacked against Nike—they had a much smaller budget than their big-brand competitor, Adidas—it was conventionally assumed that Michael was better off signing with a more established brand. That is, until Nike’s Sport Marketing Executive, Sonny Vaccaro, arms Michael’s mom with a set of strategic questions to ask during meetings with Adidas and Converse. All of a sudden, you start seeing cracks in the conventional wisdom. And the rest, as they say, is history.
The folks at Nike clearly had a brighter, more future-focused offer for Michael Jordan than folks at Adidas. But since we’re accustomed to viewing the bigger, more funded brand as the sure shot, the burden of proof is on the smaller up-start. This reminded me of the dynamics we’re seeing with Power BI.
The ThoughtSpot layup
For the last 11 years, we have been working towards our vision: allowing business users to get data insights through an intuitive interface, with minimal effort; and meanwhile, giving data teams robust control over the semantics of data, governance, and security controls. These combining forces create ideal conditions for self-service analytics.
This required a lot of deliberate investment in:
A powerful indexing system that understands and identifies the data entities in a user’s question
A machine learning system that constantly learns from both explicit and implicit user feedback
A dynamic Domain Specific Language (DSL) factory that builds a query language based on your data and metadata
We tracked the performance of LLMs like BERT, GPT-2, and GPT-3 models, but when the GPT-3.5 model came on the scene, we knew there was an inflection point in the capabilities. So, we immediately made massive investments in this direction. We realized that while LLMs have impressive world knowledge, reasoning capabilities, and SQL generation competencies, they aren’t able to answer enterprise data questions on their own. This task requires institutional knowledge, feedback, and product thinking.
This is the moment where we realized that ThoughtSpot and all the technology we‘d built over the last decade perfectly complemented the limitations of the LLMs. After about eight months of hard work, we launched a usable, enterprise-ready product: ThoughtSpot Sage.
The competitive court
As far as I know, we are the only company to build a product that empowers hundreds, and in some cases, tens of thousands of non-technical business users to answer their own data questions across the enterprise. We have been able to do this at 4 out of the Fortune 5 companies and across hundreds of other enterprises. Whenever someone brings up a competitive product with similar capabilities, I ask them if they have seen it being used in production. The answer invariably is a null set. But that doesn’t prevent large companies from selling non-existent products or demoware, muddying the waters for enterprises who want to benefit from true innovation.
It causes friction in our business, which is an indisputable problem. However, the bigger problem occurs when people try an inferior version of a product and end up walking away with the impression that “search doesn't work.” Users leave that encounter thinking LLM-based, self-service analytics tools aren’t viable for business, and that robs them of the many benefits they could have harvested from the amazing innovation happening in the industry.
One thing that Microsoft does very well is sell a roadmap to its customers: “Don’t worry about the innovation elsewhere, we will soon copy them and give you better integration within our ecosystem.”
I remember when I was working on Search—what eventually became Bing at Microsoft, one of my colleagues joyously celebrated that Windows CE had beaten Palm Pilot on the mobile devices market. I know, I am dating myself. My point is, Microsoft does a great job of being a second mover in the market by looking at what succeeded and executing their vision well.
PowerBI is a great example of this premise. It came from behind and destroyed Tableau’s dominant position in the market by building a similar product and massively undercutting them on price. But every once in a while, the Microsoft machine fails because the competition is more hungry, brimming with innovation, and nimble enough to make the move.
In the end, Windows Phone never really caught up to iPhone, and Bing never really caught up to Google. After finishing grad school, when I was deciding whether to go to Google or Microsoft, I asked the interviewer at Google: “Right now, you are ahead in the search game. But Microsoft is coming from behind with all its might, how will you compete with them?” To her credit, the interviewer said, “We are such an innovative company, they will never be able to catch up to us in search.” At that time, that answer made little sense to me. So, like an idiot, I left Google stocks pre-IPO to join Microsoft’s newly formed search division. It all worked out very well for me eventually, so not complaining.
This is what I have come to understand from these experiences: the only way to compete with Microsoft is to stay hungry and keep innovating. And that is what we are doing. We know that innovating doesn't begin and end in R&D and engineering. We know that to compete and win against Microsoft, we have to innovate in GTM too. It's easy to paint a grand vision in a roadmap, but if you look at the actual history of failed promises in self-service analytics, there is little reason to believe that Microsoft will have a better offer anytime in the near future.
Making a star
I can stand on a soapbox and say we have cracked the code, and no one else has. But I am clearly a biased party, and there is no reason for you to believe me. The burden of proof completely lies on us, the non-incumbents, to make our case. But, we’re not going to be in all the rooms where these decisions are being made. So, the only thing I can do to help you make the best decision for your business is what Sonny Vaccaro did, and I’ll do that by arming you with the right questions.
Ask the right questions:
How do you learn and adapt the institutional knowledge necessary to answer questions?
How can business users trust the answer they get from AI?
How many customers are using PowerBI Co-pilot in production?
Which customers have found success with PowerBI Q&A in production?
What is the performance like when you directly query billions of rows of data in the cloud data warehouse that is not Synapse?
How does this work when the business requires a complex schema with multiple fact tables and dimension tables?
In addition to all these questions, you should also consider why PowerBI exists. Is it strategic to Microsoft or another freebie that was designed to lure people into Azure ecosystem? If the latter, what can you expect when things go wrong and you need someone on the phone who is in it for you to succeed?
Now, the ball is in your court
After reading this, my hope is that you feel prepared for the generative AI-infused, self-service analytics onslaught ahead—that you now have the prompts you need to uncover the true intentions behind the company you’re signing over your priceless data to.
If Microsoft offers you a huge, bundled discount, and you don’t care that much about the quality of insights you get from data—by all means, go with them. But please do not go into the belly of the beast based on the false sense of security and lazy assumption that Microsoft’s strong ties to OpenAI inherently make PowerBI Copilot a good product. And most of all, don’t go because you failed to ask the right questions.
PS: If there is ever a movie about ThoughtSpot, I’d love for my character to be as goofy as Ben Affleck’s Phil Knight—I absolutely loved it!
PPS: I loved my time at Microsoft and made many lifelong friends. Some of the things Microsoft produces are incredible. For instance, we are thankful for the reliable Azure APIs for GPT models and VS Code is amazing.