Through Dicker Data’s AI Accelerate roundtable series, we brought partners together in Melbourne, Brisbane, Perth and Auckland to unpack what’s really standing between AI ambition and deployment.
One message came through clearly: customers aren’t struggling to find AI use cases anymore. They’re struggling to get them into production.
Drawing on the challenges partners shared around the table, we’ve identified the biggest barriers slowing AI projects down, and the five gates organisations need to navigate to move from pilot to real-world deployment.
Most organisations are no longer struggling to find AI use cases. The challenge is moving them into production.
The biggest barriers aren't technical. They are ownership, governance, trust and cost.
Data sovereignty is about control and visibility, not just the physical location of data.
AI consumption costs behave more like a utility than a traditional project expense, making planning and monitoring essential.
Five key gates stand between an AI pilot and production: governance, security and data readiness, cost, infrastructure strategy, and production readiness.
Twelve months ago, almost every AI conversation I had with Dicker Data partners started with the same questions: Is this real? Where do we start? What's the use case?
Today, those conversations look very different.
Many partners already have customers experimenting with AI. They've identified opportunities, built pilots and seen promising results. The challenge now isn't proving AI can work. It's getting those pilots into production in a way that's sustainable, secure and commercially viable.
Over the past few months, Dicker Data's AI Accelerate team hosted roundtables in Melbourne, Brisbane, Perth and Auckland, bringing together partners, vendors and AI specialists to discuss three recurring themes:
What stood out wasn't concern about the technology itself. No one questioned whether the models were capable enough.
The obstacles were ownership, cost, trust and organisational alignment.
"Nobody was stuck on the technology. The blockers were ownership, cost, trust and who is in the room." - Amir Kalil, AI Practice Lead, Dicker Data.
Across every city, the answer to questions around sovereignty was largely the same: it depends.
For larger organisations and highly regulated industries, sovereignty remains a non-negotiable requirement. For many mid-market organisations, however, it often ranks behind business priorities such as cost, speed and ease of deployment.
What was most interesting was how participants defined sovereignty.
Several groups observed that data stored locally isn't necessarily data that is properly governed. Simply knowing data resides in Australia doesn't eliminate risk if organisations lack visibility over who can access it, how it is used or where it flows once AI systems interact with it.
In practice, sovereignty is increasingly becoming a question of control rather than postcode.
Some discussions extended beyond data to the AI models themselves. If a business becomes dependent on a single model provider, continued access becomes a strategic consideration. Sovereignty isn't just about where information resides. It's also about maintaining the ability to access the services critical to business operations.
Perhaps the most common concern raised wasn't offshore storage at all. It was employees using unapproved AI tools outside established governance frameworks.
This one split the room. Plenty of partners aren't having the cost conversation with customers at all. Others are deep into it.
The difference was almost always whether the partner had used AI heavily inside their own business. The ones who had watched their own consumption move month to month, and had the surprise invoice, now raise cost at the design stage rather than at the first renewal.
The trap is that consumption behaves like a utility but gets quoted like a fixed-scope project. Agents make it worse. A person asking a question is one call; an agent finishing a task can be dozens, each re-reading everything so far. The pilot bill tells you very little about the production bill.
The advice from the tables was consistent. Instrument the pilot. Match the model to the task instead of sending everything to the biggest one. Show customers cost per query against the value of the output and build budgets and alerts in from day one. Partners who wrap that into a managed service are turning an awkward conversation into recurring revenue.
It also ties back to sovereignty. Once volume is big and steady, owning or reserving capacity starts to win on the maths, and that's where the cost and control conversations meet.
"The pilot bill tells you very little about the production bill." - Amir Kalil, AI Practice Lead, Dicker Data
The consensus from the roundtables was clear:
When discussing why organisations struggle to move beyond pilot projects, most people expect to hear about technical skills shortages or immature technology.
Instead, virtually every roundtable landed on governance.
An AI pilot can demonstrate genuine value, but if nobody clearly owns the initiative, understands the risks or accepts accountability for outcomes, progress stalls.
"You can have a great use case, but if nobody can say who owns it, what data it touches and who's accountable when it gets something wrong, it stalls." - Amir Kalil, AI Practice Lead, Dicker Data.
One of the strongest observations came from Brisbane.
Many partners are still positioning AI primarily as an IT conversation. However, the business problem, business value and business budget often sit elsewhere.
Several attendees argued that discussions should start with the business owner rather than the technology team. The most successful engagements focused on operational outcomes, such as reducing service desk workload, improving employee productivity or accelerating customer response times.
Some organisations are formalising this approach through AI steering committees or business-led governance teams that ensure the right stakeholders are involved from the outset.
Underpinning all of this is data readiness.
Many mid-market organisations still struggle with fragmented, unclassified or poorly governed data. While data quality projects rarely generate excitement, they often determine whether AI initiatives progress beyond a proof of concept.
The discussions across all four cities pointed to a consistent framework.
I increasingly see the journey from pilot to production as five gates that every organisation must pass through:
Governance: A clearly defined business owner, agreed objectives and documented boundaries for how AI can be used.
Security and data readiness: Appropriate access controls, quality data, classification standards and a strategy for managing shadow AI.
Cost: Visibility into consumption, clearly defined budgets and measurement based on business outcomes rather than token usage alone.
Infrastructure strategy: A deliberate decision about where workloads will run, whether that's public cloud, sovereign AI infrastructure, GPU-as-a-service or on-premises environments.
Production readiness: Defined quality thresholds, human oversight where required, success metrics and measurable business value against a baseline.
For partners, these gates represent more than implementation milestones.
Many mid-market organisations don't have dedicated AI governance teams, FinOps resources or security specialists. They often rely on a trusted partner to guide strategy, governance, optimisation and adoption.
That's where long-term value is created.
There's also a growing regulatory dimension to consider.
From 10 December 2026, organisations covered by the Australian Privacy Principles will face additional privacy transparency obligations under the Privacy and Other Legislation Amendment Act 2024. Organisations may need to explain how computer programs use personal information when making, or substantially contributing to, decisions that could significantly affect an individual's rights or interests.
For many organisations, AI governance is no longer simply a technology discussion. It's becoming a risk, compliance and business management discussion as well.
Your customers have already found their AI use cases.
The real opportunity now lies in helping them operationalise those use cases responsibly, predictably and at scale.
The partners who can guide customers through governance, data readiness, cost management and production deployment won't just help deliver AI projects. They'll become long-term strategic advisers.
That's exactly the role Dicker Data's AI Accelerate team is designed to support.
If you'd like help moving a customer from pilot to production, we'd love to start that conversation.
Frequently Asked Questions
What are the five gates between an AI pilot and production?
The five gates are governance, security and data readiness, cost management, infrastructure strategy, and production readiness. Organisations that successfully move AI into production typically have clear ownership, trusted data, budget controls, an agreed deployment model, and measurable business outcomes.
Why do AI pilots stall in mid-market organisations?
Most AI pilots don't stall because of technology limitations. They stall because of unclear ownership, poor governance, data quality issues, cost uncertainty, or a lack of executive sponsorship. In many cases, the conversation never reaches the business decision-makers who own the budget and expected outcomes.
What does data sovereignty mean for AI?
Data sovereignty in AI is about control, visibility and accountability, not just where data is stored. Organisations need to understand who can access their data, how it is used by AI systems, where it is processed, and what governance policies are in place to protect it.
How should partners manage AI costs for customers?
Partners should treat AI consumption more like a utility than a fixed-cost project. Best practice includes monitoring usage from the pilot stage, matching models to specific workloads, measuring cost per business outcome, and implementing budgets and alerts before workloads scale into production.
What's the difference between an AI pilot and an AI production deployment?
An AI pilot is designed to test a use case and validate potential business value. A production deployment has defined governance, security controls, operating procedures, budgets, monitoring and measurable outcomes, allowing the solution to be used reliably at scale.
Why is governance important for AI adoption?
Governance establishes ownership, accountability and guardrails for AI usage. Without governance, organisations often struggle to progress beyond pilot projects because no one is responsible for managing risks, overseeing decisions or measuring business outcomes.
What changes under the Privacy Act from December 2026?
From 10 December 2026, organisations covered by the Australian Privacy Principles will need to provide additional transparency where computer programs use personal information in decisions that could significantly affect an individual's rights or interests. This increases the importance of AI governance, documentation and oversight.