AI is everywhere.
Is it working for your business yet?

Most AI projects fail due to organisational issues, not technical ones: unclear goals, poor data quality, misaligned expectations, and attempting to build internally without the right expertise. The failure rate, between 80% and 95% depending on the study, is roughly double that of non-AI technology projects.

One finding stands out: vendor-led and partnership-based AI projects succeed approximately 67% of the time, while internal builds succeed only about 33% (RAND Corporation, 2024). The difference isn’t the technology; it’s the expertise, focus, and realistic scoping that external specialists bring.

Well-designed AI implementations typically show measurable value within 3-6 months. Poorly scoped projects never pay off, regardless of timeline. The difference isn’t patience; it’s whether you started with a clear, bounded problem.

The honest math includes costs vendors don’t emphasise: ongoing maintenance often runs 50-60% of initial investment annually (industry benchmarks). Training takes time. Integration is almost always harder than quoted. The real break-even for a successful project is usually 6-12 months.

If you use AI for hiring decisions, credit assessments, customer service automation, or any system that significantly affects individuals, the EU AI Act likely applies, even if you’re not a technology company and even if you’re based outside the EU but serve EU customers.

Penalties are substantial: up to €35 million or 7% of global annual revenue for prohibited practices (EU AI Act, Article 99). However, the Act includes specific provisions for SMEs: regulatory sandboxes, reduced documentation requirements, and proportionate fees.

By August 2026, complete six core steps: inventory all AI systems in your organisation, classify your role for each (provider vs. deployer), determine the risk level of each system, review vendor contracts, establish an AI governance framework, and train relevant staff.

The timeline is tighter than it appears. Prohibited practices are already enforceable. August 2026 activates the bulk of remaining requirements including full high-risk system compliance. Start your AI inventory now; you cannot assess compliance for systems you haven’t identified.

For most small and mid-sized businesses, buying or partnering delivers significantly better outcomes than building internally. Research found that purchased solutions succeed approximately 67% of the time, while internal builds succeed only about 33% (RAND Corporation, 2024).

Building makes sense in narrow circumstances: when AI is your core product, when you have genuine in-house expertise, or when your use case is highly specific. For most operational AI, the real question is which partner understands your industry and stays accountable after launch.

Effective AI training combines tool-specific skills with clear data handling rules: what information can be shared with AI systems, what must never be entered, and how to verify outputs before using them. Most companies underinvest here; 58% of employees using AI report receiving zero security training.

Training doesn’t need to be extensive. A 30-45 minute module covering approved tools, data classification, and incident reporting handles the essentials. The security risk isn’t the technology; it’s untrained users sharing confidential information with external AI services.