It is no secret that AI has serious potential to transform businesses. For instance, making operational processes more efficient through enhanced fraud detection, or accessing real-time customer insight to accelerate product development. However, more often than not AI initiatives are failing in and around the Proof of Concept (PoC) stage, with research showing 46% of AI projects are scrapped between proof of concept and broad adoption. Failure is often caused when businesses start with the technology and look for a problem to solve, rather than identifying a real business challenge first. The key here is developing a shift of mindset, from tech-first thinking to delivering measurable value.
The risks of overthinking AI
The most successful initiatives start with a clearly defined challenge, such as reducing invoice processing time by 50%, detecting fraud in under 30 seconds, or improving first-contact resolution in customer service. Once the problem is clear, leaders can choose the right tool, whether AI or something simpler.
Don’t put all your eggs in one AI basket
When used correctly AI is a highly effective and reliable tool, but it can’t be treated as a silver bullet. Some challenges may only demand simpler traditional software solutions, or a more basic form of automation. For example, a small retailer wanting to reduce the number of basic customer service calls might find more value in a simple FAQ chatbot or searchable knowledge base as opposed to developing a more complex generative AI based solution.
When implementing AI, businesses must understand what AI can and cannot do, weighing it against cost, complexity and maintenance requirements. Leaders should benchmark AI’s potential against alternative solutions and ensure they have the skills, infrastructure and data quality needed for success. This way, AI is chosen for where it can truly add value and not because it’s the most fashionable option.
By comparing AI against simpler approaches and making the decision on business value rather than hype, organisations avoid wasted investment and can focus resources where they’ll deliver the biggest return.
Adjusting the approach from concept to value
Instead of going all in on PoC, which mostly only demonstrates technical feasibility, businesses may benefit from a Proof of Value (PoV) approach. PoV shifts the focus from merely determining if AI works to evaluating whether it delivers measurable business impact.
One key advantage of a PoV mindset is consistent business alignment. This means AI initiatives are aligned to real business outcomes rather than just technical possibilities. On top of this, the PoV approach encourages value-driven decision making, which enables decision-makers to evaluate the tangible benefits of AI, such as reducing invoice payment cycles from 30 to 10 days, or automated compliance checks cutting onboarding time by 50%. Together this creates a stronger buy-in as it proves value instead of feasibility, which provides senior leaders with a tangible reason to support full-scale adoption.
Delivering a successful AI proof of value
In terms of delivering a successful, accurate and reliable PoV, there are certain best practice measures that businesses can adopt once the problem statement has been identified. The first stage is evaluating the technology fit, and assessing whether AI is actually needed, or if a simpler solution can deliver the same results. It is during this initial planning phase where ethical approaches should be considered. This includes ensuring fairness, transparency, and accountability in AI systems, and ensuring projects mirror corporate values and broader social value commitments, such as sustainability, diversity, and responsible innovation.
On top of this, one of the most important steps in the process is ensuring data quality. AI solutions that use low-quality data, which may include duplicate records or incomplete logs, will not succeed compared to AI solutions using high-quality data sets that are complete and in a standardised format. This is because low-quality data risks potential inaccurate results, bias and hallucinations. Having high-quality data is especially important during the pilot, measure and iterate phase. This is because businesses can run small-scale pilots with clear success criteria, measure results rigorously, and refine the approach before full-scale deployment. Having high-quality data here is essential as it will provide accurate results that help inform the full deployment.
In parallel to the assessments and pilots taking place, it is crucial to ensure early alignment with stakeholders. This is because buy-in from both technical and business teams can ensure their needs will be met. This is especially the case for those who will be using the solution as this helps to create smooth integration, onboarding and usability. Once all these components are aligned the testing of the AI solution can proceed. This is the most important part of the process as it is the opportunity to demonstrate it will deliver the expected benefits at the anticipated scale.
This PoV approach provides increased focus in identifying relevant AI fits, guaranteeing data quality and alignment with relevant stakeholders. By incorporating these steps with traditional approaches to AI projects, such as relevant staff training programmes or phased rollouts, help organisations will become better equipped to address the challenges that often undermine AI initiatives. This is especially critical at a time when strategic investment in AI is more important than ever.
Long lasting value is the ultimate goal
AI can provide a competitive advantage if it’s applied strategically. Organisations that resist the temptation to rush towards AI for its own sake, and instead focus on solving real business problems, are far more likely to see more productive and efficient outcomes. By prioritising a thoughtful, problem-first strategic approach, businesses can set themselves up for AI projects that not only pass the PoV phase but deliver lasting value.
Neil Gladstone
Neil Gladstone is Data and AI Practice Director at Sopra Steria, working with organisations across industry and government to deliver efficient and ethical Data and AI solutions. Before joining Sopra Steria, Neil held senior leadership roles at Roke and Siemens Rail, and led major technology transformation programmes at the BBC.
