Many organisations have been experimenting with AI. They've deployed pilots, trialled new services, built proofs of concept and started exploring agentic AI. As those early initiatives move closer to everyday use, attention is shifting from AI's potential to its practical implications. For CFOs, CIOs and CEOs, the priority is understanding how to scale AI without creating an unpredictable and difficult-to-control cost base.
Datacom Consulting Director, AI Business Solutions, Mark Horrocks, says there is a big misconception that AI is just like traditional software.
"Software licences are relatively predictable. AI is an operational investment. As adoption increases, consumption increases and organisations need to be ready for that."
Unlike conventional software, where costs are usually fixed through per-user licensing, many AI services can introduce consumption-based charging. Organisations pay not only for access to AI capabilities but also for how heavily those capabilities are used. As employee adoption grows and AI becomes embedded in business processes, costs can rise alongside usage.
This shift means AI spend must be managed differently from traditional IT budgets.
AI pilots often begin with a small number of users and a limited range of use cases. Costs appear manageable. The real challenge emerges when those pilots succeed.
As organisations move from experimentation into operational use, they start introducing AI into customer interactions, internal workflows and agentic processes. At that point, consumption becomes far more difficult to predict.
The challenge is not simply how many people use AI. It is understanding what the AI is doing, how often it is being used, what data it accesses, and which services sit behind it.
An employee using Microsoft 365 Copilot through a fixed per-user licence creates a relatively predictable base cost. Microsoft 365 E7 brings Microsoft 365 E5, Microsoft 365 Copilot, Microsoft Agent 365 and Microsoft Entra Suite together in one integrated offering. However, organisations using consumption-based AI services, advanced agent capabilities or external integrations may also introduce variable costs that require active oversight.
This is why organisations need to think about business outcomes before selecting technology.
"The key is to start with outcomes, not technology," says Horrocks. "Understand the business opportunities you're trying to address and the results you're trying to achieve, like the ROI, or the business impact before you decide which AI tools to use."
There is another misconception emerging in AI adoption: the assumption that higher spending produces better results.
Horrocks argues that cost and quality are not directly linked.
"Higher AI costs don't necessarily produce better AI outcomes. We often see organisations wasting tokens through context bloat, feeding models more information without improving the quality of the answer."
Context bloat occurs when users provide excessive or irrelevant information to a model. While this increases the amount of data processed, it does not necessarily improve output.
Multi-turn conversations can create similar challenges. As AI systems retain previous context, each interaction may become more expensive because the model continues to process information from earlier exchanges.
Reducing this waste is not only a technology issue. Organisations also need to educate staff on how to use AI effectively, including when to start a new prompt, how much context to provide and how to avoid feeding models unnecessary information. Better prompting habits can help reduce bloat, lower consumption and improve the quality of outputs.
Horrocks also highlights a growing phenomenon he calls "token maxing".
"Busy AI doesn't necessarily mean productive AI. Organisations can end up generating lots of prompts, chats and interactions that increase costs without improving outcomes."
In other words, high usage metrics should not be confused with business value. Leaders should focus on outcomes, productivity improvements and operational benefits rather than simply measuring AI activity.
One of the most practical lessons for executive teams is that AI cost management should begin well before a solution reaches production.
"The best time to forecast AI costs is during the design process," says Horrocks. "Once you understand what the agent will do, what data it will consume and how often it will be used, you can start estimating consumption."
This requires organisations to evaluate AI in the context of real business scenarios.
Executives should ask:
What measurable business outcome is this AI use case designed to achieve?
What data will it access?
How frequently will users interact with it?
What models and services will it use?
What systems and APIs will it need to call?
What is the expected cost per user, task, workflow or completed action?
How will the organisation monitor quality, adoption, security and return on investment?
Answering these questions early allows organisations to estimate consumption before they scale adoption. Modern platforms increasingly provide tools and usage estimators to support this process, making forecasting more achievable than it was only a year ago.
For many organisations, the answer to AI cost management is not technical. It is operational.
"A strong AI governance strategy underpins everything. It defines who owns the agent, who monitors it, who manages the service and ultimately who is accountable for AI spend."
As AI adoption grows, ownership becomes increasingly important. Business units may own individual agents; IT teams may monitor usage and performance, and executives may be accountable for overall spend and business outcomes.
Many organisations are now appointing dedicated AI leaders or establishing AI councils to oversee strategy, governance and investment decisions.
Horrocks also recommends maintaining two key inventories: an AI service inventory and an AI agent inventory.
"Every organisation should maintain visibility of what services they're consuming, what agents exist, who owns them and what they cost."
Combined with monitoring, usage controls and spending thresholds, these governance mechanisms help prevent unexpected cost increases and support more sustainable AI adoption.
"The good news is there are now tools such as Agent 365, Kosmoy or the M365 Admin portal that can help organisations monitor usage, track consumption and see where costs are coming from. That visibility makes it much easier to manage AI spend before it becomes a problem," says Horrocks.
Datacom’s own approach is grounded in the same principles it recommends to customers. As a Customer Zero for Microsoft 365 E7, Datacom is applying Copilot, agents, governance and value measurement within its own environment, then translating those lessons into repeatable customer patterns.
When asked for one piece of advice for CEOs, CFOs and CIOs preparing to scale AI, Horrocks's answer is simple.
"Plan. It's not glamorous and it's not where the excitement of AI sits, but planning is the key to scalable AI Plan how it works, where it works and how it both supplies and consumes data. Organisations that plan effectively and have high-quality data get to see valuable returns."
The organisations that achieve the greatest value from AI will not necessarily be those that spend the most. They will be the ones that align AI to business outcomes, forecast consumption before deployment and establish governance before adoption accelerates.
As AI becomes an increasingly important operational capability, successful organisations will need to do something that sounds surprisingly traditional: plan, monitor, measure and then plan again. In the rush to adopt AI, disciplined cost management may prove to be one of the most important competitive advantages of all.
Datacom combines deep local expertise with global AI capability to deliver people-first AI solutions to enterprise organisations and government departments.
