Enterprises around the world are investing heavily in artificial intelligence this year. But while spending is climbing fast, many organisations are stumbling over basic issues: untrusted data, slow policy automation, and weak governance. Recent research shows these gaps are not just operational problems—they threaten the return on investment for many AI initiatives.
Here are the facts, what leaders are doing about it, and how to make sure your organisation is not left behind.
Skyrocketing Investment, But Mixed Returns
A report from Gartner forecasts that global AI spending will reach around US$1.5 trillion in 2025. (IT Pro) That includes infrastructure, model development, AI services, and all the supporting cloud and compute capabilities. It’s a leap that reflects increasing enterprise demand for generative AI tools, data-driven decisioning, and automation. (IT Pro)
Despite the boom, many AI investments are failing to deliver hoped for outcomes. Delays, underperformance, or even outright failure are common. One reason: missing foundations. Executives often underestimate how much work remains just to get data and operations ready. (fivetran.com)
The Data Readiness Gap Is Real—and It’s Costly
Acceldata’s 2025 “AI Readiness & Data Management Benchmark Report” sheds light on what’s going wrong. Based on a survey of more than 150 senior data leaders from large companies, the findings are striking. (markets.businessinsider.com)
Key takeaways:
- Only 20% of organisations are fully satisfied with the accuracy and completeness of their data. Over half don’t even measure data trust. (markets.businessinsider.com)
- Almost two-thirds of enterprises devote more than 16% of their engineering capacity to fixing reliability issues. For many, resolving a data reliability problem takes more than a week. (markets.businessinsider.com)
- 40% report that users struggle to find or access appropriate data assets. This slows down self-service analytics and AI adoption. (markets.businessinsider.com)
- Fewer than one in ten organisations have automated more than 50% of their privacy and security policies. It often takes days or longer to apply new policies. (markets.businessinsider.com)
- Agentic AI (that is, AI able to act more autonomously or manage itself in some respects) is gaining interest. But today only about 1 in 8 firms implement any such autonomous data management. Over half plan to do so by the end of 2025. (markets.businessinsider.com)
What Leaders Are Doing—and What You Should Do
Closing the readiness gap is not optional. For organisations that get this right, the upside is huge: faster deployment, lower risk, better alignment between AI and business goals. Here are best practices emerging from the data and from Australia-focused reports (which face many of the same issues):
1. Measure Readiness with a Framework
You can’t fix gaps you haven’t identified. Use a structured framework to assess:
- Trust, Quality & Resilience of data
- Operational Velocity (how quickly data pipelines, fixes, new assets etc. can be delivered)
- Sustained Readiness, including governance, policy automation, and ability to scale without constant manual patches (markets.businessinsider.com)
Set quantifiable metrics in each domain. For example, percentage of data assets that pass quality checks, time required to remediate reliability problems, percentage of policies applied through automation.
2. Prioritise Clean, Accessible Data
Data quality remains the foundation. Clean, accurate and complete data is essential. But equally important is making sure data is well catalogued and accessible. Unstructured, disconnected silos cause friction. Organisations must invest in data catalogues, clear lineage, semantic layers. (markets.businessinsider.com)
3. Automate Governance and Policy Where Possible
Manual handling of security, privacy, and compliance slows everything. Policies that must be applied by hand or reactively are costly. Automation both reduces the risk of error and frees up engineering time. Cloud providers, data tools, or third-party platforms often offer frameworks and APIs that help here. (markets.businessinsider.com)
4. Build Agile Data Pipelines
Speed matters. Reliability matters. You want pipelines that can adapt when data sources change, models are updated, or demand shifts. Monitoring, observability, health checks, automated alerts—all these are tools that help ensure pipelines don’t break silently. (acceldata.io)
5. Upskill Teams and Align Leadership
Even the best technology can fail without competent human support. Many organisations are investing in training data engineers, data scientists, and business leaders to understand what good data looks like, to interpret outputs, to manage risk. Also, it’s critical to define who owns what: the roles of CIOs, CDOs, legal, security, operations must be clear. Business goals must be tied to KPIs that matter. (ADAPT)
What to Watch Next
As 2025 progresses, here are a few trends to monitor. Getting ahead of them can determine winners from also-rans.
- Compute Infrastructure Costs and Strategy: McKinsey estimates that by 2030, enterprises and cloud data centres will need US$6.7 trillion in capital investment to meet AI compute demand. (McKinsey & Company) That means decisions made now about architecture, hardware purchases, cloud vs on-prem will have long-term cost and performance implications.
- Agentic AI Maturation: More organisations plan to adopt more autonomous data capabilities, but many current systems are still in pilot or proof-of-concept stages. The risk is in “agent washing”, that is, overstating autonomy. Measure carefully what is real. (markets.businessinsider.com)
- Regulatory Pressure & Privacy Requirements: As governments legislate data protection and AI oversight, organisations with weak governance will face legal and reputational risk. Automation of policies, auditability, traceability of data sources and model decisions will grow in importance.
- ROI-Focused Benchmarks: Organisations that tie AI to operational KPIs—efficiency, cost savings, customer satisfaction, risk reduction—will do better than those chasing technical milestones with no business impact. Reports keep showing that many AI projects fail not because of model faults, but because the goals weren’t aligned well. (fivetran.com)
Specific Steps You Can Take Now
To move from risk to advantage, here are concrete steps any enterprise can take immediately:
- Conduct an AI readiness audit
Map out current data systems. Score them on trust, quality, resilience, velocity, governance. Use findings to build a roadmap. - Set data trust benchmarks
For instance: have 90% of data assets reach specified accuracy levels; reduce error rates in pipelines by x%; measure timeliness of data loading; define what “good enough” means per use case. - Automate policy enforcement
Use tools (internal or external) to codify rules on data privacy, security. Ensure any new data or model complies, without manual intervention. - Invest in data observability
Monitor pipelines. Detect drift. Create dashboards or alerts. Proactively locate issues. Provide transparency to stakeholders. - Create cross-functional teams
Data isn’t purely technical. Legal, compliance, business units, operations should all be involved. That improves alignment and reduces friction. - Pilot small but meaningful use cases
Pick AI projects that deliver measurable business value, not just proof of concept. Use them to refine processes, expose weak spots, improve governance. - Track ROI beyond hype
Build business cases that include total cost of ownership, ongoing maintenance, operational risk, not just initial setup. Monitor achieved benefits vs expectations, adjust accordingly.
Taking Stock
AI offers tremendous opportunity. But without clean data, speedy delivery, strong governance, and operational discipline, much of that spending may yield underwhelming returns. The reports are clear: data readiness is no longer a back-end concern—it sits at the front of the value chain.
Leaders who treat readiness seriously, tie AI investments to real business goals, and build systems that can scale will reap rewards. Those who don’t risk costly delays, blown budgets, and disappointing outcomes.
The question isn’t whether to invest in AI. It’s whether you are investing wisely—and laying the foundation so AI can deliver what it promises.
