It is fascinating to watch how quickly the hype of ChatGPT and generative AI (GenAI) is transforming into features in enterprise B2B software. This week I spoke to Tom Wilde from Indico Data, who led me on a deeper dive into the IDP company’s GenAI product roadmap. Indico was founded in 2014 as part of the 3rd Wave (machine and deep learning), but at the core it’s really a 4th Wave company working on large language models (LLMs) from the start. Co-founder Alec Radford published a seminal generative AI paper way back in 2015; so they are not new kids on the LLM block, they really know what they’re doing with GenAI.
Since the ChatGPT launch roiled the software market last year, the Deep Analysis team has consistently advised a cautious approach to enterprise adoption because of the inherent problems with LLMs: hallucination, incorrect answers, poor math skills, privacy, and security, among others. We have urged vendors, integrators, and end users to ignore Mark Zuckerberg’s “move fast and break things” ethos and instead “move slowly and don’t break anything – yet!”
We also pointed out that LLMs need guardrails and boundaries placed around them by product vendors before they can be reliable enough for business use. It seems that Indico has done just that in its new GenAI implementation. The company has carefully integrated OpenAI’s GPT-4 (via the Microsoft AzureOpenAI offering) into its existing product stack. As a result, GPT performance is controlled by Indico’s reputation for delivering safe, secure, and scalable AI.
This makes Indico one of the adults supervising the GPT playground, helping enterprise business leaders implement GenAI’s “magic” into their operations without the unruly side effects. The timing couldn’t be better. Enterprises are now forming SWAT teams to prioritize potential use cases for GenAI and will soon be ready to evaluate product offerings.
Safety First
Safety starts with Indico’s commitment to AI explainability which we covered in our 2021 vendor report. Indico has added prompt performance tracking metrics so the process owner can monitor how well the prompts are performing. To address data privacy and security concerns, Indico offers a private tenant implementation with data encryption and no data leakage.
Scalability is a lesser-known GPT problem but it’s a potential deal-breaker for many enterprise use cases at volume. According to Indico, AzureOpenAI (and by extension, GPT from OpenAI) is not yet practical or cost-effective for high volume use cases. Indico’s approach is to use GPT-4’s user-friendly skills to create the models, workflows, and summaries, and then for production, use its own discriminative AI models which are already battle-tested in high-volume transaction process environments. There are also some issues with rate limiting and the service level agreements the cloud LLMs can currently offer. To overcome that, Indico offers customers a 99.95% runtime SLA.
Beyond IDP
Adding GPT-powered search and query and document summarization to its product stack means Indico can now move beyond IDP and into the realm of unstructured content analytics and business intelligence. But adding new features doesn’t guarantee market fit or competitiveness; Indico will encounter entrenched specialists in those disciplines.
I asked Tom why he thought Indico could compete in the analytics space. He described Indico’s strategy using a river-lake analogy. The document intake use case is like a river with a steady stream of content, and Indico’s customers trust it to process this flow. That river empties into a lake and the lake contains all of the customer’s enterprise unstructured data.
Tom thinks the current customer base will be amenable to extending Indico use for fishing in the lake, so to speak. As evidence, he offered the example of an existing insurance customer who wanted to update their actuarial risk tables using data locked inside hundreds of thousands of policy documents. The customer estimated it would take years for their people to manually read through everything. With Indico’s new GenAI tools, the job can be completed in about two weeks.
Are We There Yet?
Not quite. As with many other GPT-themed product announcements we are covering, at the time of this report Indico’s GenAI product enhancement is not yet released. Beta testing has just commenced with a select group from its customer base.
What is our advice for business owners and enterprises looking to reframe their automation strategies? The 4th Wave of IDP is going to be transformative, and it will happen much faster than previous hype cycles such as Web 3.0 or blockchain. One need look no further than Microsoft’s CoPilot AI pricing announcement yesterday to realize that the product tsunami is imminent. Every business that uses software to automate information and processes must work out a strategy now on how it will use GenAI. Having no strategy, or ignoring it, are not viable options.
In conclusion, allow me to repeat our mantra on LLMs, GPT and GenAI: “move slowly and don’t break anything – yet!”