What should be the first AI implementation?

Despite widespread enthusiasm among business leaders and technology professionals, the reality is more complex. Over 80% of AI implementations fail, twice the failure rate of other traditional technologies¹. This figure must be taken into account.

Therefore, before making any decisions, it is essential to understand how we arrived at this point and what the current context of AI adoption in organisations truly is.

The current context: between promise and reality

We live in an era where generative AI dominates the headlines and AI Agents promise to radically transform the way we work.

According to a KPMG study, the use of AI Agent pilots increased from 37% in the last quarter of 2024 to 65% in the first quarter of 2025. However, the effective implementation rate remains stagnant at only 11%².

For 96% of CIOs, AI adoption is a top priority, but most companies still struggle to move beyond the pilot phase³.

Companies such as Salesforce — which, in my case, is the company from which I consume most information and trends — already promote solutions promising to develop AI agents 16 times faster than building from scratch⁴. However, development speed must be accompanied by an effective strategy to achieve success.

The difference between organisations that succeed and those that fail lies in the approach. While many companies begin with technology, the successful ones start with the real problem they aim to solve.

Alert #1: Avoid starting with technology

“Start with the problem, not with the technology” — this advice from Travis Gibson, CTO of Big Brothers Big Sisters of America, perfectly summarises the first major alert⁴. The fascination with AI can lead organisations to look for problems to fit their technological solutions, when the approach should be exactly the opposite.

The reality is that 60% of AI pilots struggle to demonstrate a clear return on investment⁵. This happens because many companies become dazzled by technology instead of focusing on the value they can create. For many leaders, what matters is claiming to have an AI implementation more than the actual results — it is about being part of the trend everyone talks about.

Before deciding on the type of AI to implement, it is crucial to identify genuinely problematic processes within your business. AI should be seen as a tool to solve specific challenges, rather than an end in itself.

Alert #2: Not everything requires AI agents

The hype around AI agents may give the impression that all problems require autonomous agents. The reality is more nuanced. The choice between generative AI, predictive AI, chatbots or agents should be based on the specific nature of the problem:

• Generative AI: Ideal for content creation.
• Predictive AI: Excellent for data analysis and forecasting.
• Chatbots: Effective for simple programmed tasks.
• AI Agents: Ideal when autonomous decision-making is required.
   

Alert #3: Start small, but start

The temptation to immediately pursue ambitious implementations is real and natural. We have seen the same happen with many CRM implementations where big plans later encounter complexity. The most successful organisations know the right path is to begin with simple, repetitive, low-risk tasks that quickly demonstrate value and build confidence to evolve.

For example, in a company working with travel, we might focus on just one function: “Cancel my booking”. This approach allows you to:

• Test the technology in a controlled environment.
• Identify problems before scaling.
• Demonstrate value quickly.
• Learn through small iterations.
   

Alert #4: Data is the real challenge

Although it is tempting to focus on AI technology, the true challenge lies in data quality and organisation. The good news is that perfect data is not required to start; sufficiently clean and well-integrated data will suffice.

Data preparation is more than a technical issue; it is a strategic matter that determines the success or failure of implementation. Organisations with a solid data strategy already have a significant advantage.

Alert #5: Safeguards are non-negotiable

An AI agent needs to know what to do, but also what to avoid. Safeguards are essential to:

• Prevent incorrect decisions.
• Maintain regulatory compliance.
• Preserve customer trust.
• Ensure humans retain control over critical decisions.

Alert #6: Monitoring is continuous

Unlike deterministic systems, AI agents exhibit variability inherently linked to their work. This means monitoring must be more than a one-off activity; it is a continuous process. Organisations need to define clear metrics and track not only immediate results but also long-term business impact.

Alert #7: Involve teams from the start

One of the main reasons pilots fail is a lack of involvement from operational teams in the planning process. 

As Philipp Herzig, CTO of SAP, points out, the three critical factors for choosing the proper pilot implementation are: technical feasibility, end-user desirability, and business viability⁷.

In summary

AI has the potential to transform business truly, but only when implemented with the right strategy, focused on solving real problems rather than simply “claiming to have AI” somewhere in the business flow. The alerts shared are not obstacles to progress, but the foundation for a successful project.

The question is not whether AI will impact your business, but whether you will be prepared to maximise that impact when your turn comes.

Returning to the title: We may be asking the wrong question. It is not about “what should be the first AI implementation”, but rather, whenever facing a business challenge, asking:
“Could AI help solve this problem?”

This mindset shift is what will distinguish those who adopt AI with purpose and strategy from those who implement it just because it is trendy.

Contact us to determine if AI is the most suitable solution for your challenge.

References

1. RAND Corporation. "The Pitfalls of AI Deployment: Why 80% of AI Projects Fail." Research Report RRA2680-1, 2024.
2. KPMG. "Q1 AI Pulse 2025: The State of Enterprise AI Implementation." KPMG Survey Report, 2025.
3. Futurum Group. "Maximizing ROI with Agentic AI: Why Agentforce Is the Fast Path to Enterprise Value." Sponsored Research Report, 2025.
4. Salesforce. "How To Launch an Agentic AI Pilot in 10 Steps." Salesforce Blog, 2025. Disponível em: https://www.salesforce.com/blog/launch-an-agentic-ai-pilot/
5. Futurum Group. "Enterprise AI Pilot Success Rates and ROI Analysis." Research Report, 2025.
6. Salesforce. "7 Ways Agentic AI Pilots Get Stuck — And How To Move Ahead." Salesforce Blog, 2025. Disponível em: https://www.salesforce.com/blog/agentic-ai-pilot-launch/
7. Herzig, P. "What Kills an AI Pilot?" SAP Webinar Series, 2025.