The Year in Review - AI in the real world

The diginomica team documented far too many AI projects for me to summarize here. Instead, I picked projects that collectively illustrate the most important concepts and tactics for AI results. The goal? A narrative that gives us more clues to AI success on the ground - and the pitfalls to avoid. 

Improving society one integration at a time - three Boomi end users talk tech for good

McBride suggested one thing he would have done differently is to be more data-focused to help with the shift towards AI-enabled technology services: 'We're beginning to see a push towards having to make your data AI-accessible and making sure that you're working in a data-driven environment. Now we're on a big push. Other digital leaders should focus on that area.'

Why? AI readiness became a major theme for enterprises that want to achieve real benefits with AI - and avoid landing in one of those dreaded "AI fails to deliver" reports. Exhibit A: these Boomi tech-for-good use cases by Mark Samuels. These three organizations gained data/analytics benefits, with AI futures in mind. If you can move ahead on data and integration projects, and stack up results on the way, then AI readiness is becoming an organizational practice. 

What does 'context for AI' really mean? Six enterprise leaders on why process intelligence makes AI work

Pfizer now follows a strict mandate: lead with process, not with technology. Riordan explained that Pfizer won't automate processes until they're standardized and understood, because: 'All we're doing at that point is adding a lot of costs to try and automate a broken process very quickly.' It's a discipline that has driven Pfizer from 58% order automation two years ago to 85% today, with a target of 95% by end of next year across 200+ markets globally.

Why? Data platforms/analytics projects are not the only path to AI readiness. Arguably, process improvement - or, in Celonis lingo, "process intelligence" - is an equally important play. Cue Derek's post from the Celosphere 2025 event, where six enterprise leaders stress the importance of process quality. Pfizer reinforces the storyline: stacking up big results is a great way to build AI momentum. Also note the emphasis on "context" - a crucial concept that can separate AI project success from out-of-the-box LLM gambles. 

Workday Rising 25 - hot topics customers are watching (and a couple they should be)

It is scary - and I think it should be scary if you're not willing to invest in yourself. We've challenged our employees to say: if you think you're going to be doing the same job in five years, you should be scared.

Why? At Workday Rising US in September, crucial AI lessons surfaced. One: SaaS is hardly dead; it's a strong ingredient for "AI readiness." When you factor in the complexities of context engineering, model performance, and risk management - not to mention agentic tool calls - consuming embedded AI via trusted vendors makes sense. As these Workday customers told me, automation gains via SaaS data standardization with Workday set them up well, e.g. Medidata's 73% revenue attribution, with fewer people in finance than they had four years ago. But as this pull quote from Medidata's CFO indicates, communicating your AI/skills strategy to employees is a must. I don't necessarily agree with all the skills/job messaging I heard. Nonetheless, it's far better to spell out AI's job impacts clearly - and how it lines up with your corporate culture. We're starting to see differences in this messaging from customer to customer, and that's a good thing. Consume mature AI tools, but own your strategy.

Oracle AI World 25 - AI adoption and project realities - Oracle customers share their field stories

Choctaw Nation has already adopted 40+ generative AI capabilities, and they are looking towards Oracle's AI Agent Studio as well. We asked the Choctaw Nation team: of the gen AI features adopted so far, which have had the biggest impact? Both were in HR. Crow:  'Performance review AI assist - the usage of that was phenomenal, and in Goal Creation as well - we saw a huge uptick in goal creation.'

Why? Though customer adoption of agentic AI is nascent, we did see gen AI use cases in action in 2025. For Oracle customer Choctaw Nation, progress around skills and performance reviews is notable. These are not low risk areas for AI, so it was fascinating to hear Choctaw Nation of Oklahoma explain how they are implementing these sensitive use cases. The most inspiring project? An oral language preservation project, supported by Oracle AI. My talk with Children's Hospital of Los Angeles emphasized two huge themes in enterprise AI, including compliance by design. To me, that's why industry-specific AI collaborations are where the biggest value lies.

IFS Connect 2025 - Industrial AI meets AI agents, but how are customers deriving value?

WISE allows us to simulate changes to planning and scheduling in real time and immediately assess the impact. It’s like having a sandbox inside production. And with [IFS] Copilot, our field technicians will soon have instant access to our technical knowledge base - cutting onboarding time in half and capturing the expertise of our most seasoned employees.

Why? "Industrial AI" is coming into focus, as customers learn how to embed AI into mission critical workflows. As I've noted, customers are achieving value through a blend of process visibility, old school AI, proper use of new AI tools, and some type of iterative model for industry change. Now that TOMRA is running on IFS Cloud when they want to improve a process, they have a lot of different ways they can do that. It's not some magical AI agent that makes things happen: it's everything that the AI agent is built upon. 

Ocado CEO Tim Steiner's tough love for clients - you need to learn to operate differently in an online grocery world

Today an Ocado CFC (Customer Fulfillment Center)  is capable of fulfilling a 45 to 50-item order with around 10 minutes of labor. We estimate the equivalent process in the store network would take about 70 minutes. So, it's an hour of labor saved for a single order. And this is at a time when labor costs and labor availability are one of the major pressures facing retail supply chains worldwide.

Why? I'm aggravated by the robotics discussion in 2025. Enterprise vendors generally downplay strong robotics stories, in favor of agentic keynotes. But as Stuart's Ocado piece demonstrates, savvy enterprises are addressing labor shortages and turnaround time via today's industrial robots, not cyborgs. Put humanoid robots on stage all you want - that doesn't mean LLMs transform robotics. Like autonomous vehicles, robotics is on a different advancement timeline, but it doesn't mean they aren't advancing. Let's see more industrial robots getting things done in 2026, and less humanoid keynote misadventures.

SAP Connect 2025 - inside an early Ask my Payslip rollout: how PostNL balances fast AI adoption with risk management

Technology-wise, we can do it today, but we want to be really sure the way that we present information is correct, not just because now it's factual information, but now I've got to break that down into the language part of it, and the perception of people on the language... We had some documentation before that was split on different departments and levels, and we're just incorporating all that information now into the system.

Why? Even with proper context/data, LLMs are fundamentally probabilistic. It's imperative for companies to design their use cases appropriately. Two more crucial concepts: degree of accuracy tolerance, and degree of agentic autonomy. Both vary by customer and use cases, but as I see it, the vendors that provide the customers with the most flexibility (e.g. an "autonomy toggle" per role/workflow), will fare the best. During this standout interview with PostNL's Jack Naudé, he shared their stance on these topics as they move into an Ask My Payslip rollout with SAP SuccessFactors. Naudé and I didn't totally agree on how much you can control LLM output; I included that debate. As customers evaluate, our argument is probably more instructive than agreement.

Salesforce Connections 2025 - putting Agentforce to the test with travel services customer Engine 

While we've been rolling all this out, we've really been keeping a handle on: what is our customer satisfaction scores with these interactions? In the travel space, there's always going to be these high-touch, human-facing needs - the lightning strikes scenario. 
The AI agent has allowed us to remove the less complex stuff, or remove the contacts that customers are okay with, through a seamless, quick self-service interaction, but then just leaves more space for that 'lightning strikes' scenario where a customer needs to call in.

Why? No 2025 AI project review would be complete without Salesforce, where Agentforce seems poised to take over every other Salesforce terms. Initially, I wasn't crazy about doing a use case review on AI agents for travel; i can't think of many areas where losing contact with human support can be worse. But I was very surprised - in a good way - by how Salesforce customer Engine designed their use case with exactly these concerns in mind, and got results. Another Salesforce agentic AI use case standout: Stuart's How one financial institution is shaking off the limitations of legacy chatbots - Boston Credit Union's agentic journey so far. 

The CCE 2025 AI review - how do we account for the gap between vendors and customers on agentic AI?

We've been increasing these layers of abstraction with process, with more advanced technologies and so on. And I think that with agents, we're going to see the same thing. So AI agents are going to come in on the ground floor, and they're going to replace a lot of entry-level tasks. But what is also going to happen is your first level humans are going to start to think more like managers. 

At Constellation Connected Enterprise 2025, a panel of AI practitioners brought this year's themes to a head. This quote captures much of the tension - and possibility - of agentic AI in the workforce currently. Another panelist challenged the notion that AI and humans are interchangeable: "I think we need to combine the strengths of a human being with the strengths of machines." Yes, we ended the year with a considerable gap between agentic keynote messaging and customer reality. But as we see from these use cases, that gap is closing a bit, and know-how is growing. Of course, that doesn't solve the question of how each of us should respond to these shifts. I wrote about the skills imperative and quest for purpose amidst machines, but that's just another stake in the ground for 2026. Note: for more on AI project do's and don't, check my Enterprise AI: year in review video appearance with Andreas Welsch.