Invoice processing with…Bing?
That is not a typo. Greg Council of ImageSource tipped me off about his recent experiments with GPT’s ability to make sense of semi-structured data without any need whatsoever for IDP. Greg’s been a long-time advocate for the use of large language models (LLM) and now the rest of us are learning why. I decided to test it myself on GPT-4, which has sneakily been hiding inside Bing Chat all along.
I asked Bing, From the following text, please tell me the invoice number, the name and address of the supplier, and the invoice total. I fed it the raw OCR text copied out of a complex multi-page, multi-line invoice. The results were perfect on the first try. Next, I instructed Bing, Please extract each line item from the text, display the results in a table, and export the results as an Excel spreadsheet. Bing performed perfectly, with one ironic exception – it could not create an Excel file. No problem, that’s coming soon with Microsoft365 Copilot.
This quite frankly blew me away and I needed a sanity check, so I messaged Greg who I know to be naturally cynical of all tech hype. Here’s what he told me: “Once Google and Microsoft introduced this capability, you could import, say, an Explanation of Benefits (EOB) form sight unseen, and it would return key value pairs and other standard data. You’d still have to normalize the output; but that’s trivial when compared to the effort it takes to configure current IDP systems for that. I brought this concept over to ImageSource and we’re working with GPT-4 now.”
Anyone else blowing up IDP with GPT?
Hyperscience Up
I finally got to meet the team from Hyperscience in a briefing earlier this week. Since its founding nine years ago the company has raised a staggering $300 million investment, by far the largest for an IDP company. Naturally I was curious to learn how they’re spending it. Product development is certainly one area getting some serious juice. They shared the product roadmap and as a former product management exec I immediately recognized the impressive level of ambition and vision. While I cannot share details, what I can say is Hyperscience plans a very big move into unstructured document processing (such as contracts, master service agreements, medical files, correspondence, etc.).
Hyperscience also reported strong growth year over year including a 100% increase in volume of pages processed, and its headcount has passed 250 full-time employees (FTEs). Its go-to-market slide was what I would expect from someone like Kofax, and a clear indication that Hyperscience has successfully transitioned from cool VC-funded AI startup into a mature business working with the top partners and winning big deals with blue-chip customers. That the company still has an “interim” CEO after 12 months is a bit odd, but all other signs point to strong growth ahead. I plan to update our Hyperscience vignette soon – watch this space.
ABBYY Down
Big Tech companies such as Meta and Google aren’t the only ones laying off workers. IDP leader ABBYY continues to reduce its workforce, with another round of layoffs this month. At its peak last year, I estimated ABBYY had over 1,400 FTEs. Now it appears the headcount is under 1,000, a 30% cut. Some of this could be due to trimming the COVID hiring bulge, as ABBYY like so many others experienced windfall growth from the massive work from home shift and over-hired.
But the Chief Revenue Officer is also gone; from my experience, such a change does not inspire confidence for sudden sales growth. Also cut was the product marketing manager for ABBYY’s proof of identity (POI) solution (which I just covered) amid rumors there has been only one sale of that to date. And let’s be fair, over the past 12 months ABBYY sales may have suffered from guilt by association because of its close connection to Russia.
Is ABBYY’s setback an isolated case, or is this the harbinger of more IDP industry cuts to come? In our IDP Market report, we predicted a shake-out because this industry is over-crowded.
The Last Capture Fortress is Crumbling
For about twenty years, the industry called what I am covering here “capture”. My friend and mentor Harvey Spencer hosted the legendary annual Capture conference where all the leading software companies and scanner manufacturers gathered – me included on several occasions. Since the great Google transformer boom of the mid-2010s, hundreds of startups have arisen under the banner of “IDP”. They don’t know or care about capture, a term that for them carries old legacy baggage.
I just received an invite to this year’s meeting and – predictably – it has been renamed The Capture and IDP Conference. Maybe the new breed of companies will feel more comfortable. At any rate, come on in, my Infosource friends, the water is warm and the pool is plenty big enough for us all.