Seven lessons from partners actually delivering AI projects
Christine Horton speaks to partners to uncover why so many AI projects stall at the pilot stage and what it really takes to deliver at scale
Everyone in the channel’s talking about AI but in reality, fewer are delivering it at scale.
Over the past two years, partners have invested heavily in AI practices, launched advisory services and helped customers experiment with everything from Microsoft Copilot to custom agents.
Yet while interest remains high, many organisations are still struggling to move from proof of concept to production.
According to research from the Global Technology Industry Association (GTIA), 97 per cent of IT solution providers are using AI internally, but only around a quarter could be described as genuinely AI-driven, with a documented strategy, formal governance and dedicated leadership in place.
“The pipeline’s full. It’s the conversion that’s the problem,” explains Carolyn April, VP of research and market intelligence at GTIA.
The challenge is becoming familiar to partners across the industry.
While customers are increasingly convinced of AI’s potential, many remain unsure where to start, how to measure success and what is required to scale projects beyond early experimentation.
Stop starting with AI
Andy Crossley, CTO at Oakland, Softcat’s data, AI and automation consultancy, says many organisations are still approaching AI from the wrong direction.
He argues that many organisations are still caught up in what he describes as “AI for AI’s sake”, driven by boardroom pressure to demonstrate progress rather than a clear understanding of how AI supports business objectives.
“We spend quite a lot of time trying to challenge people,” he tells CRN UK.
“’Why do you think you need AI for that? That’s just good analytics or basic machine learning, or actually, just look at the data you’ve already got and use it.’”
Arran Speding, team lead for data, automation and AI at Softcat, says the first step should be understanding why a customer is pursuing AI in the first place.
“We always try to get as deep an understanding of the organisation as we can.
“What are the drivers? Who is pushing it in the business? What is the outcome they’re trying to achieve?”
Rather than treating AI as a standalone initiative, Speding argues that partners should focus on the role data, automation and AI can play in solving an existing business challenge.
“That’s not an AI problem. That’s a business problem,” he says.
“Then we apply: ‘what role does data play in solving it, what role does automation play, and what role does AI play?’”
Elsewhere, Ensono’s chief AI officer Jim Piazza believes the most successful AI projects usually start smaller and more operationally than people expect.
“They do not begin with a giant transformation programme. They begin with a painful workflow, a measurable outcome and a team that is ready to change how they work,” he says.
For partners, that means focusing less on tech demos and more on helping customers identify inefficiencies, bottlenecks and repetitive processes where AI can create measurable value.
AI projects: data remains the biggest obstacle
While large language models continue to improve, all the partners point to the same challenge: data.
According to GTIA, data readiness remains the most consistently underpriced and underscoped element of AI projects.
“Customers underestimate the state of their own data, and partners underestimate how much cleanup it takes before AI can deliver anything meaningful,” says April.
Data may exist, but it’s often scattered across multiple systems, inconsistently labelled or poorly governed.
As a result, organisations can build compelling demonstrations but struggle when they attempt to deploy AI against real operational data.
“‘Our data is ready.’ Usually, it is not. It may be available, but not clean, connected, contextualised or governed well enough for production AI. If you think your data is good enough, you’re wrong,” says Piazza.
For partners, that creates a significant opportunity.
Rather than leading with AI tools, the most successful partners are increasingly leading with data assessments, governance reviews and infrastructure readiness engagements.
Why AI pilots get stuck
For many organisations, proving that AI works isn’t the hard part - it’s scaling it.
While the focus often falls on models, platforms and tools, partners say the real barriers are usually elsewhere.
“Technology isn’t really the blocker when it comes to bringing AI to life. It’s everything that sits around it,” says Speding.
That “everything” includes data quality, governance, operational readiness, security, change management and the ability to measure outcomes.
It is also the reason so many promising AI pilots fail to progress into production deployments.
Crossley describes production deployment as a “hockey stick” curve, where the complexity and cost increase dramatically once an organisation attempts to operationalise a successful pilot. Many organisations also struggle because they have not established a baseline.
“If they solve that problem, what’s it worth?” says Crossley.
“They don’t know because they’re not measuring the process as it is today, let alone what it could be.”
That lack of measurement makes it difficult to build a compelling business case for broader rollout.
Advania UK’s CTO Chris O’Brien thinks organisations also underestimate the importance of governance, security and compliance as deployments scale.
“A proof of concept can look compelling in isolation but scaling it across real business processes means dealing with risk, controls, accountability and integration in a much more disciplined way,” he says.
AI projects and the ‘people problem’
AI projects also often fail because organisations focus too heavily on technology and not enough on people.
GTIA identifies change management as one of the most overlooked aspects of AI adoption.
“Partners bring the technical chops but not the change management or process design depth. The implementation works but adoption doesn’t follow,” says April.
O’Brien says he sees the same issue.
“AI adoption depends heavily on the kind of leadership mandate people are given. This is a significant change in how work gets done.”
The organisations seeing the strongest results are investing in role-based training, practical use cases and ongoing support rather than simply deploying tools and expecting employees to embrace them.
Piazza argues that AI should be positioned as a tool that augments employees rather than replaces them.
“The biggest wins often come from helping experienced people move faster,” he says
The channel’s AI opportunity
As AI platforms become increasingly commoditised, partners are discovering that the real value lies elsewhere.
GTIA believes data preparation, governance and organisational readiness represent some of the biggest opportunities over the next two years.
“Deploying tools and platforms? That’s getting commoditised fast,” says April.
“The value sits upstream and downstream of the technology itself.”
For partners, that represents a shift from selling technology to helping customers operationalise it.
“The MSP opportunity is evolving from managing technology estates to helping clients operate in a fundamentally different way, where humans and AI agents work together safely and effectively,” says O’Brien.
Seven lessons from partners actually delivering AI projects
1. Start with the business problem
Successful projects begin with a specific challenge, workflow or inefficiency, not a technology discussion.
2. Measure the process before you improve it
If customers cannot quantify the current state, they will struggle to prove ROI later.
3. Treat data readiness as a project
Poor data quality remains the biggest obstacle to production AI deployments.
4. Build governance early
Governance, security and compliance should be designed into projects from day one, not added before go-live.
5. Expect production to be harder than the pilot
A successful proof of concept proves the technology works. Production proves the organisation is ready.
6. Keep people at the centre
Role-based training, leadership support and change management often determine success more than the technology itself.
7. Focus on outcomes, not tools
The channel opportunity is shifting away from product selection and towards implementation, operationalisation and ongoing optimisation.