How Not To Demo With AI

This month Yellow Radio attended Incorta’s ERP conference in SF, as part of the aiXplain team. For context, aiXplain provide a class-leading GenAI platform that enables clients to use any LLM model, from any source, so that organisations can build and run LLM-enhanced applications with no vendor lock-in and easy model substitutability.

Incorta, as a leading ERP platform, had chosen aiXplain as the underlying AI service provider, and at this conference Incorta were showing the results of this partnership to their existing and future partners and clients.

During the conference they provided a live demo, with a client CFO on stage, able to ask questions. These were complex, comparative questions that the Incorta platform, powered by aiXplain, was able to answer well. As GPT-4o had just been launched, I followed along in real-time in the audience, entering the same questions in the app on my cellphone. Not only was ChatGPT now able to answer these questions and provide the same answers, it was able to provide downloads of the data - exceeding the capabilities of the demo.

The audience were impressed and the team were delighted at the success of the demo. I then privately spoke to the team, pointing out that the demo could have been achieved with a $20/month subscription, they were horrified. But the demo could have easily been a demo of unique value - by showing Incorta able to answer questions that required unique, private data that only they held. Had they done this, ChatGPT would have been unable to provide the same value.

Increasingly as foundation LLM models become more capable, there are now only two sources of defensibility:

1. The use of private data that no frontier model has, and
2. The use of complex supply chain or distribution channels

In this scenario, Incorta actually have both - but failed to demo or explain this.

The moral of this story is simple: always consider how the market is evolving, focus on a unique value proposition - and execute with speed.