The thirty-fourth episode of the podcast that you know and love as "We Love Ugly Data!" is out; available in audio form everywhere you get your podcasts from and additionally in video form via YouTube (which we've embedded below). This month, it's time for the annual Deep Analysis predictions, so Matt is joined by both Alan and Dan to discuss a handful of those in our extensive predictions report and catch up on how well those made 12 months have panned out now that they've made contact with reality.

In this month’s episode:

Dan: 2025 Review and 2026 Prediction

First up, Dan discusses his 2025 prediction, "To grow, intelligent document processing (IDP) companies must cross the border. He's giving himself a tick here, as he's seeing many vendors pushing the edges of both the vertical (industry) and horizontal (business function) markets that they've grown out of. For 2026, Dan is backing the prediction "The unstructured data gold rush will finally begin", which follows on from his 2025 prediction (and also Alan's 2025 prediction recapped below) that the growth in specific vertical and business functions for AI-derived technology like IDP and the broader AI agent market means a need for access to a lot more business data (and all the jeans, shovels and buckets that requires).

Alan: 2025 Review and 2026 Prediction

Alan starts with a review of his 2025 prediction "Structured data people will stop treating unstructured data like something that got stuck in their shoes" and gives himself a partial tick, in that he believes that there is general move from the big application vendors to recognise unstructured data is really useful for context, but also at the same time that it's really complicated to pick out the different data types within that big unstructured pile (and that's not just a technical challenge). For 2026, he's pitching "Edge computing will re-emerge as a strategic imperative"; specifically, that access to AI-derived applications increasingly need to take advantage of on-device processing, clever caching technology, etc., to enable remote workers (and related use cases) to operate smoothly.

Matt: 2025 Review and 2026 Prediction

Finally, it's Matt and he's keen up provide an update on his 2025 prediction "The shift to “payment on outcome” is going to lead to some awkward conversations between customers and suppliers". He explains that the business of developing consumption-based pricing models for AI agents has become increasingly complex, moving beyond the success criteria (especially Salesforce's per-conversation pricing, which remains but is joined by other options). In general, determining the economics of AI agent use remains in the early stages of being made easier to calculate. That notwithstanding, Matt is giving himself a tick here. For 2026, Matt is backing "'Doorstep adoption' of AI will be exposed as a counterproductive farce" as his chosen prediction; that vendors enforcing the bundling of AI tools into renewals doesn't make their adoption real. Matt suggests that it's akin to having a trailer (or a caravan if you prefer) welded to the back of your car without your permission, and then a bill arriving for the job for the excellent utility it provides.

https://www.youtube.com/watch?v=LulxRfzf8jk

Related Links for Series 4 Episode 12

Here's Alan's introductory blog for this year's predictions: It’s Time to Separate the Wine from the Hype.
The download page for the full awards

Support the show

The thirty-third episode of the podcast that you know and love as "We Love Ugly Data!" is out; available in audio form everywhere you get your podcasts from and additionally in video form via YouTube (which we've embedded below). This month it's time to discuss the annual Deep Analysis Innovation Awards, so Matt is joined by both Alan and Dan to discuss this year's winners; Infrrd and LatticeFlow AI (but as ever, show notes are included just below the embedded video).

In this month’s episode:

Preamble: Why do we have Innovation Awards?

First up, Alan talks about why Deep Analysis has annual Innovations Awards and why they are different to many other awards that are out there. Firstly, you can't apply for them, and if you win one there's no fee to pay, fancy dinner to attend (and buy a table at) or winners speech to rehearse. Also there's no set number of winners each year either; the team discusses which briefings wowed them and meets the awards criteria; Does it solve a problem? Does it apply ingenuity? Does it add value? Does it show flexibility?

Winner: Infrrd

Dan introduces the first award winner, Infrrd. Having initially annoyed him with their marketing choices, he was won over with the number of patents that have been awarded for their research in document AI and their development team deserves recognition for that work. Additionally, Dan was impressed with a demo involving Autocad drawings and data extraction, retrieval and document understanding (and awarded them a mini award of their own in the Intelligent Document Processing Market Analysis 2025-2028 report for best demo). Finally Dan was also impressed with Infrrd's work on mortgage loan files; a notoriously difficult use case. Alan adds that IDP in general is featuring in virtually every briefing that Deep Analysis receives right now, as it is a vital part of how unstructured data is made available for AI and agentic technology to use and it could be said it's what is driving many of the predominant AI use cases and is fundamental to LLMs themselves.

Winner: LatticeFlow AI

Matt then talks about second 2025 award winner, LatticeFlow AI. One of the features of 2025's discourse has been that AI projects are not making it out of pilots and into production. One of the reasons that is happening is that often the hurdle that is compliance, risk etc are not baked into the development from the beginning. LatticeFlow AI is focused on ensuring that an organizations AI applications meet both internal (think GRC; governance, risk and compliance) and external (regulations, like ISO, EU AI Act etc), with a focus both on internal data quality as well as how well individual models are likely to work in conjunction with it. Alan adds that that it's another areas that underlines the increasing gap between demos and production realities; as soon as the word governance people tend to switch off, but it's a vital part of ensuring that organizations are insulated against inevitable legal challenges coming down the road for organizations that don't realise its importance.

Postamble: What are we looking for in 2026?

To wrap things up, the team nominate what they hope they'll be seeing in 2026 that will end up with them pitching briefings in for the nominations pot in 12 month's time.

Dan: Hoping to see better support for tabular information in IDP products; there's lots of start-ups working on the

Support the show

The thirty-second episode of the podcast that you know and love as "We Love Ugly Data!" is out; available in audio form everywhere you get your podcasts from and additionally in video form via YouTube (which we've embedded below). As usual, it’s three topics in (just a touch over) 30 minutes, and the show notes are included just below the embedded video. It's Matt and Alan this month for a Salesforce-heavy episode, discussing their acquisition of Apromore, some of the thoughts from the pair sparked by the recent 2025 edition of the Dreamforce conference, and wrapping up some of the fall tech conference trends.

In this month’s episode:

Topic 1: That’s Apromore!

First up, Matt and Alan discuss Salesforce's recent acquisition of process analysis vendor Apromore, reminding themselves that Salesforce was an investor in the company (as well as there being a co-selling partnership). As is pointed out, our pair had both thought that there was a good fit for an acquisition (and had said so), but also got into the details of what using such a platform means practically (hint, it's not magic and needs process skills to operate if effectively, but it's a potentially huge on-ramp for Agentforce). They then go on to point to other business application ecosystems where similar investments have been made (and Matt acknowledges that he'd forgotten about SAP somehow when writing the analysis of the deal). Finally, there is some pointing at Deep Analysis's "Work Intelligence" research as a way of better understanding how all this might fit together.

Topic 2: Agents of Dreamforce

The big annual Salesforce Dreamforce conference took place recently, and whilst Matt watched from home in the UK, Alan was on the ground in San Francisco. They discuss how Salesforce is a bellwether vendor in the business application market, and whilst there was a host of product announcements, it was the changing messaging around Agentforce that got them most intrigued. Matt's product and adoption-focused blog post and Alan's one from the conference floor are discussed, with the general opinion being that it's much harder to deploy agents that were being made apparent 12 months ago, and the messaging is starting to reflect the steps required to get on track.

Topic 3: Fall Trends

Finally, Alan's got a couple of thematic trends to discuss from his travels at a bunch of business application conferences this fall/autumn and briefly discusses them; firstly, the underwhelming response from attendees to announcements and - if you ask them - the audience is hot for IDP.

https://www.youtube.com/watch?v=xEd5UHaeNCs

Related Links for Series 4 Episode 10

Work Intelligence Market Analysis 2025-2030 - our full market analysis of the Work Intelligence marketplace, including process analysis.
Work Intelligence: Improving Processes by Balancing Human Intelligence and AI - our initial 2022 report on Work Intelligence as a concept, which you can download for free.
Salesforce acquires Apromore to give Agentforce a process analysis on-ramp - deal anal

Support the show