If you're interested in AI then Veritone (Nasdaq: VERI $14-16) is an IPO you'll want to at least have a look at. It's a small deal led by three smaller banks - Wunderlich, Craig-Hallum and Northland. The larger firms wouldn't touch this -  it's still more like a research project than a company. [UPDATE: Under pressure from institutions the deal size has been increased to 2.85M shares and is expected to price at the high end of the range tonight. Aftermarket is harder to call based on what the syndicate is seeing now.]

Veritone is playing into what has been a red-hot trend of using "AI" as a label for almost anything in software. Whatever a company was doing before they are now doing it with "AI" and not just advanced programming. At least 50% of what is going on in these companies wouldn't pass for AI in most circles. But Veritone is aiming at harder problems that leverage much of the work going in in areas including image recognition, speech understanding, sentiment analysis, etc.

For the public markets, Veritone will be one of the first and few "pure plays" on AI technology - especially for small cap investors for whom companies like Google (GOOG) and Facebook (FB) are not fits. There are other companies that have an "AI angle" like RocketFuel (FUEL) and Criteo (CRTO) but these are still online advertising companies.

Before we get to specifics part of the reason to pay attention is that Kevin Kelly, one of those guys who really gets it right when it comes to technology, emphasized in his recent TED talk that for AI to be really productive it would require a variety of reasoning types, each with very different capabilities. That's what Veritone is aiming to do - connect and orchestrate a number of distinct reasoning engines so that they can perform notably complex and varied tasks.

An example from the Veritone IPO roadshow slides illustrates what they are trying to do. In the scene below (includes both video and audio) there is an event happening with the driver of a race car, let's say it's about a crash. Being able to "understand" the event requires analyzing the video and picking out things like logos and identifying objects, processing the audio yields the words being said by the driver and sentiment analysis helps define whether the emotions are fear or triumph. By knitting together multiple engines the analysis can be completed effectively and according to Veritone management, for low cost - especially when compared to a human.

Veritone is aiming high and is striving to be "an operating system (OS) for AI" and deliver a platform that transforms AI processing into "actionable intelligence" for enterprise customers. Today however what Veritone is delivering is a solution that handles mostly audio and video processing for the media industry (advertising). It's a very long road from here to a general purpose platform.

In other words, this is an advertising technology company that has just launched an "AI Platform" business not unlike what RocketFuel and others are trying to do.

Getting valuation right for a company like this is hard. It's too young to be able to build a credible long-term IV model. We'll let the numbers speak for themselves:

At the upper end of the range ($16) the deal size will be $46M and the market cap will be $228M.

Revenues of $8.9M for 2016 were down from $14M in 2015. The company nearly doubled G&A spending from $8M to $15M during the same period.

It's fair to ask "what are investors actually buying here?" A case can be made that Veritone is at least a play on the growth of AI-focused business applications. Management works hard to paint a picture of themselves as being the right team to pull this off but a close reading of their actual backgrounds and job history in the Veritone prospectus suggests otherwise.

In conclusion, we'd suggest investors avoid this deal and instead consider a more substantial company like Cloudera (CLDR) which already leverages data to provide enterprises with AI-based insights into their business.