Why Companies Still Struggle to Scale AI
Why Companies Still Struggle to Scale AI
Despite the hype and hefty investments most companies are still stuck in AI pilot mode. While 92% of businesses say they’re investing in AI, only 1% describe themselves as truly “AI-mature”. Why? Because it’s not the algorithms that are holding them back. It’s everything around them.
From legacy systems and siloed data to talent gaps and internal resistance, the real barriers to AI adoption aren’t just technical, but rather cultural, economic, and organisational. Here’s what’s getting in the way.
1. People Problems, Not Machine Problems
For all the talk about model performance, it’s people and not the technology that are most likely to derail an AI project. According to research from MIT Sloan and BCG, 70% of AI adoption issues stem from people and process, not the tech itself.
A key challenge is leadership. Many executives are charging ahead with ambitious AI plans, but without clear governance, employee buy-in, or readiness across teams. In a 2024 survey 48% of staff said AI roll-outs were “tearing the company apart” as leaders moved faster than internal capabilities could keep up.
Employee trust is also low. More than half of workers fear AI will erode their roles, and only 38% say they trust corporate decisions made using AI (Forbes, 2025). But there are promising initiatives': Salesforce’s internal “Career Connect” platform, which helps employees build AI skills and transition into new roles, filled 50% of its Q1 2025 positions internally while also increasing staff engagement.
2. Legacy Tech Is Dragging AI Down
Even when leadership is aligned, the technical foundations often aren’t. Outdated IT systems, poor data quality, and integration issues continue to stall AI at the infrastructure level.
A recent survey of 500 global firms found that 68% see legacy tech as a direct blocker to AI, with nearly half still running systems more than 11 years old. The result? Projects stall or fail altogether. One European airline’s AI-powered crew rostering tool had to be shelved after €3 million in sunk costs because mainframe APIs couldn’t handle real-time data integration.
Even firms that make it past deployment face operational hurdles. MLOps, the process of managing machine learning models in production, remains a pain point. The type of AI solution whether ready-made or custom greatly impacts cost and scalability. Ready-made tools are cheaper and faster, while custom builds suit complex needs but demand more resources. With rising GPU costs and hybrid-cloud rollbacks, choosing wisely is essential.
3. The Hidden Costs of Going “All In”
AI isn’t just technically demanding, it’s financially complex. Total costs can range from $50,000 to over $1 million depending on project scope, with enterprise platforms requiring significant long-term investment (Savvycom).
McKinsey found that only 1 in 5 companies reported a positive bottom-line impact from AI two years after launch. Still, many keep spending due to “FOMO” (fear of missing out) rather than evidence of value. Without the right metrics, governance, and talent, these investments often underdeliver.
There’s also the growing risk of vendor lock-in. Switching between AI platforms or models is becoming more costly, with cloud “exit fees” rising 18% year-on-year. Some CIOs now negotiate escrow clauses just to ensure they can access their model weights if partnerships sour.
4. Regulation, Trust and the Ethics Gap
Finally, compliance and trust can’t be an afterthought. From GDPR to the incoming EU AI Act, firms face growing scrutiny over how they collect, store, and use data as well as how they ensure models are safe, fair, and accountable.
Regulatory uncertainty is already delaying launches and inflating costs. In the EU, 68% of firms say the AI Act is hard to interpret, and 40% have earmarked 2025 IT budgets for compliance. In Italy, OpenAI was fined €15 million for ChatGPT data breaches, and the Replika chatbot was penalised for failing to meet transparency standards.
Meanwhile, regulatory friction is growing. In July 2025, 44 CEOs from major European firms, including Airbus and Siemens, called on Brussels to postpone the AI Act, citing uncertainty, cost, and implementation burdens (Brussels Times). The request signals a broader tension: while AI rules are tightening, industry readiness is uneven. For many companies, unclear and evolving regulations could become a key barrier to adopting and scaling AI responsibly.
Getting Past the Blockers
The data is clear: AI success is less about picking the right model and more about building the right foundation. That means:
Creating transparent upskilling paths to bring staff along
Fixing legacy systems and data silos before scaling
Building governance into the design, not as an add-on
Running staged value proofs before going “all in”
Companies that move too fast without internal alignment risk damaging trust, wasting budgets, and reinforcing silos. But those who prepare their people, processes, and platforms before the pilot stand to gain the most.