Why Many AI Startups Are Consultancies Posing as Software Businesses

There are plenty of reasons generative artificial intelligence hasn’t taken off with businesses as much as pundits expected, including high costs and errors (i.e. hallucinations). One less-discussed reason is that conversational AI is simply hard to use. Big companies outside of tech can’t often hire the kind of technical talent needed to make AI software cheap enough to use at scale or to connect the AI to their private data so that it can handle tasks specific to their needs. 

As a result, startups in the field have found revenue hard to come by. And one group of startups has resorted to exploiting the ease-of-use issue by acting like consulting firms rather than traditional enterprise software businesses, VCs tell me. (Back in February, I wrote about a failed AI software startup that decided consulting would be a better fit.)

One investor I spoke with has met several startups in recent weeks that frame themselves as “data curation startups," helping customers generate data from large language models to help customize other AI models for specific tasks or domains like fraud detection. Such data are referred to as synthetic.