You have probably sat in a budget meeting where the AI line item swallowed everything else. The cloud migration that would cut your infrastructure spend by thirty per cent gets pushed to next quarter. The ERP upgrade that would retire three legacy systems loses its champion. The data-quality programme that would make your AI models trustworthy never makes it past the spreadsheet. Everyone wants the generative capability, but no one wants to fund the foundation that makes it work.
Cognizant's research confirms what many technology leaders already feel: AI spending is cannibalising IT modernisation budgets, and the trade-off threatens the return on both. When organisations pour capital into large language models and copilot licences without first addressing technical debt, data silos and process fragmentation, they build on sand. The AI tools perform poorly because the underlying systems cannot feed them clean inputs or operationalise their outputs. The modernisation work that would unlock that performance sits unfunded, creating a cycle in which neither investment pays off.
The pattern shows up in three ways. First, infrastructure budgets shrink as AI compute demands grow. Cloud spend that once funded application re-platforming now covers GPU clusters and inference endpoints. Second, talent budgets shift toward prompt engineering and model fine-tuning, leaving fewer resources for the architects and data engineers who would redesign core systems. Third, project timelines stretch as teams try to deliver AI pilots on top of platforms that were never designed to support them. The pilot succeeds in a controlled environment, then fails at scale because the production systems cannot handle the load or the integration complexity.
The risk is not that AI spending is wasteful. The risk is that it becomes wasteful when it runs ahead of the modernisation work that would make it effective. A generative assistant that helps customer-service agents draft responses delivers little value if the CRM system cannot route the conversation to the right queue or surface the customer's purchase history. A forecasting model that predicts demand more accurately than a spreadsheet delivers little value if the supply-chain system cannot adjust orders in response. The AI capability is real, but the organisational capability to use it lags.
IT modernisation is not a legacy concern. It is the work that turns data into a usable asset, processes into repeatable workflows and systems into platforms that can absorb new tools without breaking. When that work is deferred, AI projects inherit the dysfunction of the systems they touch. Models trained on inconsistent data produce inconsistent predictions. Automations built on brittle integrations fail when upstream systems change. Dashboards that aggregate insights from multiple sources cannot reconcile conflicting definitions of the same metric.
The organisations that extract value from AI are the ones that treated modernisation as a prerequisite, not a distraction. They cleaned their data lakes before they trained models on them. They decomposed monolithic applications into services that could be orchestrated by intelligent agents. They standardised APIs so that new AI tools could plug into existing workflows without custom middleware. That groundwork is expensive and unglamorous, but it is what separates a proof of concept from a production capability.
Learning and development leaders face a parallel squeeze. Budgets that once funded upskilling in cloud architecture, data engineering and agile delivery now fund AI literacy workshops and prompt-engineering bootcamps. Both matter, but when the balance tips too far toward AI-specific skills, teams lose the foundational capabilities that make AI projects succeed. An employee who can write a sophisticated prompt but cannot design a data pipeline or map a business process will struggle to deploy AI in a way that scales.
The answer is not to choose between AI training and modernisation training. It is to teach them together. Show people how to assess whether a system is ready for an AI layer. Train them to recognise when technical debt will undermine a model's performance. Build capability in the disciplines that AI amplifies rather than replaces: data governance, process design, integration architecture and change management. The organisations that do this will spend less time troubleshooting failed pilots and more time scaling the ones that work.
AI spending will continue to grow, and it should. But when it crowds out the modernisation work that makes AI effective, it risks becoming another technology investment that promised transformation and delivered disappointment. The leaders who resist that outcome are the ones who treat infrastructure, data quality and process redesign as enablers of AI, not obstacles to it.
Sources:
.png)
The WorkFlow podcast is hosted by Steve Glaveski with a mission to help you unlock your potential to do more great work in far less time, whether you're working as part of a team or flying solo, and to set you up for a richer life.
To help you avoid stepping into these all too common pitfalls, we’ve reflected on our five years as an organization working on corporate innovation programs across the globe, and have prepared 100 DOs and DON’Ts.
