Editorial
Every year, we at Deep Analysis take a time-out to reflect on the past 12 months and what we expect to happen in the year to come. Of course, unexpected things occur: COVID-19 came out of the blue and disrupted everything in 2020, and the 2024 U.S. election results were a huge surprise to many.
So, who knows what 2025 has in store? Hopefully some good — not bad — surprises. Even so, we would put good money on our predictions here to play out to some degree, and (like our 2024 predictions) we might just nail it. As always, it's important to point out that predictions are just that, a bit of fun and at best educated guesses. With that disclaimer out of the way, here are our top 10 predictions (with a bonus 11th) for 2025!
1. Services as Software (SaaS 2.0) Will Emerge
Consider this briefly: If enterprises can build agents for specific roles, why can’t third-party service companies build entire agent workforces for hire? The acronym SaaS is already in use, and the industry will have to develop something better than SaaS 2.0. Agentic Software Services, perhaps? This is coming to the market, and third-party-based AI services will emerge over the next few years. In some regards, it’s arguably the natural (or unnatural) evolution of the BPO (Business Process Outsourcing) industry.
Simply put, we mean going beyond SaaS applications and providing an entire multi-agent service that combines management and optimization. In other words, a traditional SaaS delivers tools for your workers, while Services as Software offers the tools and undertakes the work tasks through AI automation on your behalf. It may seem a bit out there and risky, but many technology firms are already exploring this, and some are even building these out for launch, so it’s going to happen. How successful they will be is an open question.
Related Article: Salesforce Brings an End to the 'Work of Work'
2. The Shift to 'Payment on Outcome' Will Lead to Some Awkward Conversations Between Customers and Suppliers
We’ve been watching the payment models for the generative-AI-derived applications that we used to call copilots and now call agents (more on that later) with interest from their inception. We predicted that the initial all-you-can-eat flat pricing was a short-term offer and that metered pricing was right around the corner, and it’s now starting to arrive.
Consumptive pricing is not new in software. Indeed, even some large vendor suites now sporting big, shiny agents atop their marketecture have had some rate-sheet-based pricing for a while. However, with agents, something more interesting is occurring, as rather than being just a straight consumptive model based on bandwidth, storage, or even instances, pricing is based on outcomes. That’s to say, if your agent interaction results in a customer being satisfied, you get billed. At least, that’s the idea.
Some definitions of what that satisfied customer might look like or how that is indicated are vague. Clicking “end conversation” on a chat window? That’s a success. Is a chat open for more than 24 hours? That’s a success. If authenticated users don’t detect a case being opened within a set time, you’d have some cause to chalk up a deflection, but that’s complex data to connect. It’s easier to bill the $2 and hope there’s an ROI in the wash, right?
3. Structured Data People Will Stop Treating Unstructured Data Like Something That Got Stuck on Their Shoes
We might be a bit sensitive about this one, but we think it’s about time that unstructured data was taken seriously by people who’ve previously ignored it.
We get it. If you’ve been delighted with the rows and columns of structured data forever, it’s hard to look at the ugly mess that makes up most of the totality of data in organizations. Our world is weird and scary. We wish it were possible just to employ brute force or magic on unstructured data to make it behave, but you’re going to have to trust that it won’t work because if it did, we would have solved its issues a few decades ago. We’re a little offended that you’d assumed we hadn’t tried that, but it’s OK. We forgive you.
Our prediction here is thus: if you take that vast amount of unstructured data in organizations seriously, you’ll find that it contains a massive amount of the sort of actionable stuff that you need to make the exciting transformative stuff — like agents — work to the full extent that you imagined they should. It might smell like exhaust fumes (because that’s often what it mostly is) and be just as hard to grab hold of, but within it are the raw materials that will enable processes to be better orchestrated, tasks to be more simply completed and ironically, everyone in the workplace to be a good deal healthier.
Oh, and one last thing. Simply jamming it all into a search index won’t work. That’s been tried a bunch of times before.
4. The Year of Orchestration
Workflow orchestration engines are not new; the concept has existed for over 25 years. Though the terminology has fallen in and out of fashion over the decades, it’ll be all the rage again in 2025. This time, though, the orchestration ambition and scale are unprecedented.
Imagine thousands of complex AI agents running across an enterprise in coordinated harmony. Everyone from Microsoft and Salesforce to UiPath and Appian is focusing on the importance of orchestration and talking about it to partners and customers as much as possible. Talking about it is one thing — pulling it off is another altogether. Orchestration engines that work in a test or demo will likely face massive challenges when deployed at a huge scale in the coming years in the wild. The agentic dream can’t come true without advanced orchestration, so the need for it and the associated challenges and bumpy road ahead will keep the topic left, right and center over the next few years.
Related Article: Microsoft's Magentic-One Coordinates Task Completion Across Multiple AI Agents
5. Information Governance Shifts Camps to Security
If there is one thing information management vendors and buyers alike can agree on, it is that information governance and compliance are tough sells. Records management is even tougher. But what if (some are asking) these functions are repositioned as part of an overall compliance and security sale?
When Microsoft moves, typically everyone else follows. A year or so back, they moved their Purview Records Management solutions out of the Microsoft 365 | SharePoint business and into the Security business. Shifting information governance and records management as a practice from information management to risk and security only fits well with some practitioners, but it makes sense. Do we expect more to follow suit? For sure. However, there will be a backlash from traditional information governance and record management professionals; it’s where things are heading. Moreover, as AI and agentic AI start to drive the narrative, information governance and compliance will be pulled into the conversation more often, whereas in their current position, they are frequently left out altogether.
6. Legacy ERP Systems Become Unpicked – Costs Are Under Scrutiny
A subplot within the 2024 market dynamics has been the saga of SAP trying to move its customer base to the cloud and, in turn, receiving significant pushback from its customers. Add to this story after story of aging ERP systems that are now under the microscope and being asked to justify their costs. Ripping and replacing is not easy, and in many companies, after decades of customization, it is nearly impossible.
Using process mining by consultants reveals that many functions and workflows within ERP are either overly complex or not used at all. The rise of agentic AI at least provides an alternative to replace hard-wired legacy systems, and there is little doubt that ERP vendors like SAP and Oracle will ride the agentic bandwagon, but they are also starting to encounter targeted attacks and credible replacement options for parts (if not the whole) of their systems. Indeed, this is the sneaky plan of many big players in the AI space. They know that legacy ERP systems are incredibly costly. Still, as these systems come to the end of their useful life, these players plan to take over elements of existing ERP operations and ultimately replace them as the system of record.
Related Article: What Does the Future Hold for ERP?
7. The Year of Agent Washing? AI Agents Will Continue to Mean Whatever Software Companies Need Them to, Until Further Notice
We noted that if 2023 was the year of generative AI assistants — that we got used to uniformly calling copilots because most of them ended up being called “copilot” — then 2024 was the year of the generative AI agents. The agents were essentially iterations of the assistants and retained their core functionality, but they added a load of new orchestration capabilities, which from time to time were claimed to be autonomous. As ever with software product categories, there’s no actual definition of an agent and how it differs specifically from an assistant. However, for many vendors, what was a copilot is now an agent.
Trying to make sense of progress can be an exhausting and baffling ordeal — one not helped by a CEO extolling the virtues of a next-generation product already in beta with selected customers and producing remarkable tenfold improvements, that turns out subsequently to have mainly been a figment of his imagination (yet made it into a well-rehearsed keynote script as well as several press conferences without contemporary correction due to reasons that have yet to be explained).
Conference
Oct
20
Gartner IT Symposium/Xpo Orlando 2025
Register
Conference
Oct
27
Gartner HR Symposium/Xpo Orlando 2025
Register
As always, we’re doing our best to break these products into logical pieces and explain how they fit together, what they might be good for, and how you can approach using them in your organization. Regardless of their nomenclature — and we’re crossing our fingers we don’t go through another mass product renaming in 2025 — we recommend you keep an eye on product release notes and cross-reference those against any on-stage hyperbole. That’s a big, free pro tip from our analyst firm to you.
8. Someone Will Notice That Desktop Copilots and Task Mining Tools Pair Beautifully and Improve the Prospects for Both at a Stroke
If you’ve ever dug into Deep Analysis's Work Intelligence research, you’ll know we’re big fans of task and process mining tools. For us, they represent excellent value for organizations when they provide insight into the tasks and processes that impact workers within the organization: how they resolve the traversal of applications, their workflows and where common bottlenecks occur (and are solved). As those with a little more experience will attest, process and transaction data is often the locked door for organizations approaching process mining. Task mining works from the workers’ desktop forward, collecting data as they go about their daily business, with those individual workstreams being combined for analysis.
It occurs to us that desktop copilots (or agents, as I suppose we’re now going to have to call them) are in a position to gather an allied collection of task data that might not yet form a standard part of analysis data, yet is identifying the creation of new task and process approaches on the fly. We’ve been somewhat cautious about the reported productivity benefits of copilot skills — “time back in your day” and all that — but combining the task mining collection agent and copilots will give you some precise working data along with tracing the adapted tasks for comparative analysis.
9. To Grow, Intelligent Document Processing (IDP) Companies Must Cross the Border
Classic IDP begins with a document and ends with the handoff of nicely formatted data to a business application. IDP vendors have mostly stayed on their side of that border, occasionally launching raids into the business app database to validate an extraction, but they have mainly focused on the very hard task of reading documents accurately.
With this core IDP value becoming increasingly commoditized by the hyper-scalers or the nimble startups, IDP companies will be squeezed hard to find more growth. They must find ways to move up the value chain by adding more functionality that previously existed only on the other side of that border. We’ve seen early examples of cross-border raids inspired by LLMs, such as “decisioning” (performing an action based on analysis of the extracted data) and “summarization.” That’s not enough value to move the needle.
IDP companies should now look for vertical use cases and add high-value cross-border functionality. Some have already transformed from “IDP” into A/R automation providers, with process automation functionality on the ERP side of the border.
Related Article: What Is Intelligent Document Processing?
10. IDP Consolidation Will Accelerate as Every Business Application Vendor Realizes It Needs a Steadily Flowing Pipeline of Unstructured Data
Too many IDP vendors are in the market, and more are joining the crowd. As business applications and major platform vendors wise up to the importance of unstructured data, we can expect to see more consolidation, particularly as many of them may come at a reasonable price.
In 2024, we saw a few acquisitions of IDP companies by application companies. One example is Eigen Technologies by Sirion, the contract lifecycle management software. However, this trend will only accelerate in 2025 as unstructured data becomes the new big data for business applications. We know for sure that many larger firms are eyeing IDP bargain acquisitions in 2025 and with funding hard to come by, it’s a buyer’s market.
11. Enterprise Blockchain Is Back on the Agenda (Again)
Industry talk revolved around enterprise blockchain seven or eight years ago, then everything went quiet. It was a victim of its hype and critical buyer confusion in conflating the scandals of the cryptocurrency world with enterprise blockchain (two very different things). However, though it’s not on the tech mainstage anymore, enterprise blockchain has gained global traction in supply chain, healthcare, government and finance. Few projects are labeled “blockchain,” but that doesn’t mean that the distributed database and immutable data technology aren’t being leveraged.
So, will blockchain be highly visible in the market in 2025? Probably not. But that doesn’t mean it is not being leveraged behind the scenes, and in a world of little trust, a dose of blockchain immutability can come in very handy.
Editor's Note: This article originally published on the Deep Analysis website.
Learn how you can join our contributor community.
