The Year in Review - the year in Partner Thought Leadership

If 2024 was the year AI went mainstream, 2025 was the year that enterprises started asking harder questions. What does it actually take to make AI work? Who is accountable when agents act on their own? And in the scramble towards automation, what happens to the humans? 

The answers coming out of this year's partner thought leadership share a common thread – technology alone isn't enough. Whether the topic is agentic AI, data streaming, or supply chain resilience, the most relatable pieces argue that success depends on foundations – clean data, clear governance, and a workforce that is ready (and supported) to adapt. The hype cycle has shifted to implementation reality and with it, a more grounded conversation about what AI can and can't do. As we get to the end of the year, settle in and read some of the highlights.

Acumatica: Reliable AI requires data prep, governance and training – but where do you begin?

Before the agents can act, the data has to be ready. John Case makes the case that most businesses – especially SMBs – are held back not by a lack of AI ambition but by fragmented systems, inconsistent data, and unclear governance. His four-step approach (unify, cleanse, govern, train) is practical rather than aspirational. It's grounded in a simple reality – AI amplifies whatever data quality issues already exist.

The Dukathole Group example brings this to life – a brick manufacturer that unified data across six divisions and used real-time dashboards to manage acquisitions, expand into construction, and create over 450 jobs. As Case puts it:

Taking care of data first is essential. AI amplifies the importance of data integrity, hygiene and governance. Reliable AI outcomes depend on high-quality data, and businesses that prioritize these foundational steps set themselves up for success.

Atlassian: What's the ROI of AI? You won't know until you've learned how to make the most of it

Tal Saraf opens with an uncomfortable truth: it's too early to measure true AI ROI, and the obsession with proving value may be getting in the way of realising it. His argument is that now is the time for experimentation and cultural groundwork, not spreadsheet justifications.

The shift he's advocating is to stop treating AI as a tool you use occasionally for quick answers. Start treating it as a teammate. That's a bigger change than it sounds – it means moving from "summarize this page" to genuine collaboration on complex planning and decisions. Atlassian's own research backs this up:

Strategic AI collaborators experience greater benefits compared to 'simple AI users.' These collaborators not only save up to a full workday each week but also experience enhanced work quality and are more inclined to reinvest their time in acquiring new skills.

One piece of practical advice is to set up Slack channels for AI news, encourage employees to share quick videos of how they're using AI, and create safe playground environments for experimentation, and there's lots more to dig into from the research findings.

Blue Yonder: Tariffs vs sustainability – data, networks and technology can help companies thread the needle

Saskia van Gendt tackles what may be the most complex strategic challenge of the year: how do you maintain sustainability commitments when tariff disruption is forcing rapid supply chain reconfiguration? The answer isn't straightforward, and van Gendt doesn't pretend otherwise.

The relationship between tariffs and energy gets knotty fast. Shifting production away from China might reduce carbon intensity if the destination grid is cleaner – but Canada's renewable-heavy grid comes with potential tariff penalties. Onshoring to the US could lower logistics emissions, but increased manufacturing demand might bring coal plants back online. And fashion companies face an additional constraint – the recycled and bio-based materials they need often cluster around the overseas facilities they're being pressured to abandon.

Where technology helps is in managing the complexity:

Data and analytics can integrate data on sustainability factors such as carbon emissions and waste levels with supply chain data on facilities location and logistics and transportation. This can help companies make difficult decisions that involve multiple factors and complex trade-offs.

Celonis: Resilient digital twins – essential in the age of AI and automation

I like Prof. Wil van der Aalst's analogy here. Just as self-driving cars need live traffic data, not just static road maps, digital twins of organisations need real-time information to be useful when conditions get volatile.

His concept of the "resilient digital twin" is speaks to a common problem - AI trained on historical data can't handle unprecedented disruption. The solution combines process intelligence with live external data feeds and – crucially – human oversight for situations that fall outside the model's experience:

No matter how sophisticated your training data and AI models are, disruption, emergencies and complex automation use cases require human intelligence. AI cannot know how best to respond to extreme or new situations that have never been experienced before.

The practical path he outlines is helpful too – start with basic process mining, add feedback loops, then enable automated triggers. It's a good counter to the temptation to jump straight to full autonomy.

Certinia: Operationalizing Customer Success – how to build a successful and sustainable CS organization

Todd Kisaberth's piece tackles a problem that is familiar to many – customer success teams that operate in silos, disconnected from the rest of the organisation. His argument is that CS principles need to be woven into every department's workflow, not confined to a dedicated team.

The methodology of defining objectives, segmenting customers, mapping journeys and designing support models is methodical but not mechanical. There's a real emphasis on clarifying which team owns which touchpoint – and eliminating those moments where customers get conflicting messages from different parts of the business.

One finding that caught my eye: over half of CS organisations are now monetising their activities. As Kisaberth notes:

Premium offerings such as personalized training, advanced analytics, or priority support services not only enhance the customer experience but also create new opportunities for CS to be a revenue-generating function.

Confluent: The future of data streaming – five key predictions for 2025

Richard Timperlake positions 2025 as the year data streaming moves from leading-edge to mainstream. His argument rests on compelling ROI evidence: 64% of businesses where data streaming is foundational report achieving or anticipating up to ten-times return on investment.

But he's honest about the barriers. Cultural resistance – the "why change what isn't broken?" mindset – remains significant. And he names a specific friction point:

For some organizations, batch processing models continue to dominate, as DSPs are often viewed as a costly technical upgrade rather than a catalyst for lucrative business transformation.

He shares a useful lens of cost efficiency, growth, customer experience as a way to look for where data streaming adds value. And his prediction that 2025 could mark "the zenith of cloud adoption" – driven by regulatory constraints, repatriation pressures, and geographic restrictions – is still relevant for the coming year.

IFS: The 3 pillars you must get right to succeed with agentic AI in the industrial enterprise

Christian Pedersen cuts through the agentic AI noise with a practical take of data, technology, and culture. None of these is optional, and the interdependencies matter.

On data, he's blunt about the state of most industrial environments – "scattered across siloed legacy systems, hidden in unstructured formats or trapped in infrequently updated spreadsheets." The remedy requires designated data stewards, standardised formats, and security embedded from the start.

The cultural piece is where many deployments stumble:

Even the most advanced AI agents will stumble if people refuse to embrace them. Concerns about job displacement, a lack of AI literacy and mistrust of automated decision-making can stall even the best-planned deployments.

I particularly liked his framing of AI agents as new hires who need more check-ins at the beginning. It's a helpful way to set realistic expectations about the path to autonomy.

Oracle: AI agents promise a productivity revolution – here's what leaders must get right

Chris Leone opens with Robert Solow's 1987 observation that computers were everywhere except in the productivity statistics – and draws the parallel to generative AI today. The promise is vast but the measurable impact, so far, is not. His case for AI agents centres on their ability to reason, adapt, and act rather than just execute scripts. The examples are solid - Ford converting 2D sketches to 3D models, Sonos recalling past customer interactions for troubleshooting, AtlantiCare helping doctors navigate medical records.

The governance gap he shares is worth noting:

Regulators in the UK, Singapore, and Hong Kong have made it clear that boards are responsible for overseeing AI risks. In the US, corporate case law is evolving in ways that could hold directors accountable for AI oversight. Yet today, only 17% of companies report board-level AI governance.

Planful: Beyond the AI hype – 3 practical FP&A use cases you can use today

Steve Welsh gives a masterclass in pragmatism. His piece doesn't promise transformation; it offers three specific use cases where AI is already delivering value in finance: predictive forecasting, anomaly detection, and natural language insights.

The Rocket Software example is a good illustration of using AI-generated forecasts to guide hiring decisions at scale without adding finance headcount. Luis Martinez, Sr. Manager of FP&A, explains:

We can lean on Planful AI to rationalize our hiring pace by creating upper and lower bounds or pointing out where something looks odd. That helps us have a meaningful dialogue with the business, know where to ask questions, or see where someone accidentally added an extra zero.

The broader survey data is telling too – 60% of finance leaders use AI daily, but only 52% apply it directly in FP&A. There's still a gap between adjacent use and core workflow integration.

Pure Storage: Technologies, spending, and sustainability – will 2025 mark a turning point?

Patrick Smith surveys a landscape of uncertainty – geopolitical tensions, economic caution, environmental pressures, cyber threats – and asks whether 2025 will be the year that forces more disciplined technology choices.

His prediction that AI projects will face stricter scrutiny reflects a shift from experimentation to accountability:

Spending rationalization will become essential, anticipating potential fatigue from the technological frenzy and facing increased pressure to demonstrate tangible returns on investment.

The sustainability tension he identifies is real: rising AI and storage demands are temporarily sidelining energy reduction initiatives, even as sustainability remains a stated strategic priority. And like Timperlake, he suggests 2025 could mark "the zenith of cloud adoption" – regulatory constraints like DORA and cost pressures are changing the calculus.

Sage: The power of an end-to-end construction cloud platform

Julie Adams tackles an industry with particular complexity: remote teams, field operations, financial data, labour shortages, and supply chain volatility all competing for attention. Her argument for cloud platforms rests on four pillars: real-time connectivity, AI-powered automation, security, and smarter bidding. The invoice error scenario she describes makes the case for integration better than any abstract claim could. Picture this: an accounts payable clerk processes an invoice for 1,000 cubic yards of concrete instead of 100 v a potential $585,000 overpayment. The cloud platform catches it immediately by cross-referencing field tickets entered through the superintendent's mobile app. No phone calls, no email chains, no month-end surprise.

On AI automation, she's realistic about the scope:

Your accounting team can process hundreds of invoices automatically, with built-in error checking that catches mistakes before they become problems. Project managers can generate detailed progress reports in minutes instead of hours.

Her recommendation to start small – perhaps with a single cloud-based tool while keeping existing systems in place – acknowledges that transformation doesn't happen overnight.

Salesforce: It's time to talk about "AI literacy"…literally

Peter Coffee took on a deceptively simple question: what does "AI literacy" actually mean? The EU AI Act's Article 4 mandates it, but the definition remains contested.

His concern is that a minimal interpretation – the ability to write a prompt and read a response - won't be enough. The UNESCO definition he cites is more demanding: "a means of identification, understanding, interpretation, creation, and communication."

The historical parallel he draws caught my eye:

In 19th century Russia, the nobility and the military officer class routinely spoke French to each other, using the Russian language only to deal with servants and with rank-and-file soldiers.

Coffee warns that if AI becomes a language of exclusivity, broader "literacy" may be more illusory than real. People may overestimate their understanding and fail to notice when their interests aren't being served.

Samsara: Autonomous AI looks set to shape the future of fleet management

Philip van der Wilt documents the rapid adoption of AI dashcams in fleet management – and the results are striking. Lanes Group cut mobile phone usage by 92% in seven months. Countrystyle Recycling saw violations drop by 85%.

But the real story is what comes next – agentic AI that can make decisions and act without human intervention. Van der Wilt outlines three areas of near-term progress – autonomous decision-making for infrastructure monitoring, predictive maintenance that automatically schedules repairs, and hyper-personalised customer interactions. The vision of collaborative AI networks is ambitious:

In the supply chain, for example, everything from inventory management, order processing, dispatch, routing, and fulfilment could be completed autonomously with minimal human input.

He's candid about the governance questions this raises, particularly in Europe where data security and privacy concerns are shaping a stricter regulatory stance.

ServiceNow: AI is eating the world – why the AI revolution is good for business

Brian Solis makes the case for AI as a great equaliser - enabling small retailers to deliver personalisation on par with e-commerce giants, and healthcare providers to improve outcomes with precision diagnostics. His distinction between iteration and innovation is worth sitting with. Iteration improves what exists; innovation creates something new. With AI, automation is iteration. Augmentation - humans and machines collaborating toward outcomes neither could achieve alone – is where innovation lives.

He uses Steve Jobs' comparison of early TV to radio (filmed on a camera) to make a point that resonates:

What happens when a new medium enters the scene, we tend to fall back into old media habits.

Solis notes that ServiceNow is doing the same with AI - automating yesterday's work rather than reimagining what work could be.

UiPath: 93% of IT execs are eager to implement agentic AI – but have they considered governance?

Cameron Mehin opens with survey data that captures the current state: 93% of IT executives are interested in agentic AI, 37% already use it, but 56% cite IT security as their primary concern.

His concept of "controlled agency" speaks to this tension directly. The methodology limits what agents can access, maintains human oversight, and implements security measures like encryption, access controls, and regular audits. The result:

Under this model, employees control the entire framework, while the agents are still able to work autonomously to make decisions and learn over time to improve their outputs.

The implementation advice is sensible: start with an internal, smaller-scale process with limited financial or security impact, build familiarity, then expand. It's a journey, not a patch-up project.

Workday: A vision for humanity in the next era of work

Kathy Pham steps back from the technical details to ask a broader question: as AI reshapes work, what happens to the distinctly human? Her answer draws on Workday research showing that ethical judgement, relationship-building, and empathy remain the most valued competencies - now and in an AI-driven future.

The trade-off she explores between efficiency and experience is subtle but important:

Take, for example, the ubiquitous use of GPS. While AI-powered navigation gets us to our destinations with precision, it may also rob us of the remarkable discoveries typically found off the beaten path.

Her call to invest in critical thinking, emotional intelligence, and cultures that value human connection isn't a rejection of AI. It's an argument for what organisations need to consciously preserve as automation expands.

Zoho: The enterprise is lacking digital maturity – decision intelligence can help, but how do you get there?

Raju Vegesna introduces "decision intelligence" as the synthesis of AI, business intelligence, and contextual intelligence. He maintains that enterprises often fail not because they lack advanced tools but because those tools don't mesh with existing workflows.

The Virtuoso example illustrates the point – a luxury travel advisory that migrated from Microsoft Dynamics to a unified system, gaining improved coordination, better collaboration, and higher-quality contextual data. He proposes matching computing power to task complexity - and this is a useful counter to the tendency to throw maximum AI horsepower at every problem. As he puts it:

It's also far less efficient to, for example, pull a wagon with a tow truck when a bicycle would suffice – in a business setting, especially at the enterprise level, inefficiencies with technology accumulate.