Don’t let “data perfection” kill your AI initiative

We’re delighted to share our blog with Lou Simon, for this extra-special, guest feature post. This blog builds and expands on the points raised in this post, recently published on Forbes.

AI adoption is not keeping pace with the desire to implement AI in organizations, which has never been higher. There are valid reasons that organizations are hesitant to trust AI; security, compliance, and ethics. But the easiest to point at is the lack of quality data. The assumed knowledge around AI is that it is worthless without good data. “Data is the fuel for AI” or “Bad data is junk food for AI” are two soundbites that perfectly capture the issue.

But what if this assumption was wrong? Could organizations get a huge competitive advantage by taking a different stance on AI, using imperfect data, and taking an agile approach that helps them manage their trust issues? The benefits are:

• Operational efficiencies
• Learning about how to implement and scale AI
• Retaining top talent who will drive innovation

In the rapidly evolving landscape of AI, businesses often find themselves caught between the promise of groundbreaking technologies and the realities of their existing data infrastructure. Many enterprises struggle with the concept of “data perfection,” a belief that data must be flawless before embarking on AI initiatives.

Rethinking thinking

The traditional mindset in many enterprises is that AI projects require a pristine, well-organized dataset to be successful. This belief stems from the complexity of AI technologies, which often depends on vast amounts of high-quality data to produce meaningful results. However, achieving data perfection is not only difficult but also unrealistic in the dynamic environment of today’s businesses. The endless pursuit of perfect data can stall AI projects indefinitely, preventing companies from realizing the benefits of AI.

We’ve seen a revolutionary approach that directly addresses this challenge by enabling organizations to get started with AI without waiting for perfect data. This pragmatic strategy allows businesses to begin exploring AI’s potential and start generating value in a matter of weeks.

The three key principles behind the approach are:

1. Start with Imperfect Data: You need to recognize that your data will never be perfect. So just start with the data you have today. By leveraging advanced technologies like machine learning, natural language processing (NLP), and retrieval-augmented generation (RAG), we’ve seen that you can extract meaningful insights even from incomplete or messy datasets. This allows businesses to get up and running with AI in as little as 2-3 weeks for simple use cases, without the need for extensive data preparation or cleansing.
2. Start small. Iterate quickly: You need executive management that understands AI is a journey, not a destination. This agile approach emphasizes continuous learning and refinement, allowing AI models to improve over time as more data becomes available. This iterative process not only enhances the accuracy and relevance of AI solutions but also enables organizations to scale their AI capabilities gradually and sustainably.
3. Practical, Business-Oriented Use Cases: Rather than focusing on theoretical applications, you need to identify high-impact use cases that align with business goals. Whether it’s optimizing sales processes, enhancing customer service, or streamlining operations, it is possible to design to deliver tangible results that drive business value.

Back in 2015, I wrote about the potential of AI, which is now achievable. But, seeing the adoption challenges, I have since spearheaded a consulting approach around the 3 principles. I have operationalized the principles in a platform that enables customers to quickly validate and test AI in different use cases with incomplete data sets against different prompts and language models. The platform enables them to drive continuous improvement by AB testing different parameters; prompt, weighting, choice of LLM, data volumes, and formats. And from there, they can expand the scope into even more valuable use cases. 

This approach provides a low-risk, easy on-ramp to AI, enables an ROI to be built as customers get more engaged and confident in using AI, and sets the groundwork for scaling AI’s adoption inside the organization. It is unlocking the limitations of AI trust and poor data quality. It is best understood through a customer example. Based on the huge competitive advantage the customer is getting they understandably don’t want to be named.

Real-world example

A global manufacturer of industrial products couldn’t see a compelling use case for AI. During the use case brainstorming phase of the project, they identified a challenge with passing support cases from Tier 1 to Tier 2. The knowledge articles and product schematics were inconsistent and out of date. Therefore they couldn’t be trusted without input from the “grey beards” with 20+ years of company experience. This made scaling support teams time-consuming, and a high turnover of staff made it even more challenging.

The initial AI project analyzed support tickets, knowledge articles, and product schematics to provide better responses to open tickets. What was surprising was how good the responses were once the prompts were fine-tuned against the best LLM. The “agile” approach and AI platform made it easier to iterate each of the parameters and see measurable progress toward the optimal solution.

Enthused by the early success, the scope of AI was increased to cover more complex support tickets. One example was a customer with obsolete equipment that needed to replace an actuator. The AI support recommended using a part for a newer machine, rewiring the panel, and shaving down the actuator to make it fit. These step-by-step instructions were blind tested with one of the grey-beards, which completely independently came up with an identical solution. 

The benefits of employing AI were:

• New support team onboarding reduced from 3 months to 1 week
• Support teams felt empowered and staff turnover was reduced
• An established approach for scaling AI inside the organization

Final word

Based on the conversations I have with executives who are hesitant about AI, the power of this approach can be summarized as “You wouldn’t wait until your operational processes and HR policies were perfect before you hired an intern. And you wouldn’t give the intern responsibility for decisions without oversight and support. Think of AI as that intern.”