Rethinking AI Failures: Think Small to Think Big - A Free-Market, Joint Enterprise Model for Innovation

Before we get into more doom and gloom let's turn the growing AI Flops trend into a collaborative success shall we: Think Small to Think Big by blending small product based AI innovators, enterprises, and independent consultants. By fostering collaboration among these diverse stakeholders, instead of the usual faces bagging all the business, we will have a free-market joint venture model. Free market competition is good, it fuels the sort of innovation that got us out of the caves!

For those of us still batting this means much needed competitive advantage, or just levelling up the playing field - on time time, on budget, with better outcomes to unlock the full potential of AI and drive innovation across industries.

The Rand Report dives into the reasons behind the current high profile AI failures, identifying five key root causes:

1. Misunderstanding the problem: A lack of clarity about the problem the AI is meant to solve can lead to misaligned solutions.

2. Data issues: Insufficient, poor quality, or biased data can hinder AI model performance.

3. Misaligned objectives:  Optimizing AI models for the wrong metrics or failing to integrate them  into existing workflows can lead to ineffective solutions.

4. Lack of expertise: Building and deploying AI solutions requires specialized skills that might be lacking within an organization.

5. Communication breakdowns:  Poor communication between technical and non-technical stakeholders can  result in misaligned expectations and project derailment.

The RAND report serves as a crucial reminder that successful AI implementation requires more than just technical prowess. It underscores the importance of clear problem definition, robust data management, alignment with business objectives, skilled personnel, and effective communication.

However perhaps the crucial constant to consider is whether the reliance on big consulting firms for AI projects might be a contributing factor to the high failure rate. Their modus operandi is not to cut to the chase or find the cure, but to focus on maximizing profit for as long as possible (NHS, Government et al). This overshadows the specific needs and internal expertise of individual departments within any form of enterprise. For me adopting a "pureplay" approach, where each department has a say in the development and management of its own AI solutions, will offer several advantages:

• Closer alignment with business objectives:  Each department has a deeper understanding of their own challenges to  tailor credible AI solutions more effectively to their specific needs.

• Empowerment and ownership: Giving departments greater autonomy fosters a sense of responsibility and encourages internal innovation.

• Cost savings: Building AI capabilities in-house could reduce reliance on expensive external consultants.

• Agility and flexibility: Smaller, department-specific AI teams can be more agile in responding to changing business needs.

However, this approach also presents certain challenges so perhaps utilising the talents of smaller or independent consultants working hand in hand with AI product focussed companies is a better 'hybrid' approach :

• Expertise and skill gap: Smaller departments might not have the necessary AI expertise in-house.

• Dovetailing of effort: Without a level of central coordination, departments might end up reinventing the wheel and wasting resources.

• Scalability: The  three-way AI product-advisor- enterprise makes it easier to manage the  transition to enterprise wide AI projects effectively.

Overall, this balance between the two approaches might be ideal. For complex, enterprise-wide AI initiatives, leveraging external expertise could still be beneficial. This turns the 80% chance of an AI Flop into collaborative success at a fraction of the cost of the big consultancy approach. However there is another ingredient the budget and power hungry top players want you to ignore - SLMs.

Small or Specialized Language Models (SLMs) trained on internal enterprise data, rather than general Large Language Models (LLMs) like Copilot, presents a compelling strategy to reduce hallucinations, avoid data sharing and the constant organisational anti-AI sentiment, thereby increasing the success rate of AI projects within enterprises.

Let's break down the benefits:

• Reduced Hallucinations:  Training an SLM on specific internal data ensures the model has a  deeper understanding of the company's terminology, processes, and  context, leading to more accurate and relevant responses. This reduces  the risk of generating nonsensical or misleading outputs, a common issue  with general LLMs trained on vast and diverse data.

• Enhanced Data Security & Privacy:  By keeping the data within the company's infrastructure, it reduces the  risk of exposing sensitive or confidential information to external  models or third-party vendors.

• Targeted Solutions:  SLMs can be fine-tuned for specific tasks or departments within the  enterprise, enabling them to provide more specialized and effective  assistance, further minimizing errors and misinterpretations.

• Mitigating Anti-AI Sentiment:  By demonstrating that the AI models are built upon and operate within  the company's data ecosystem, it fosters trust and transparency,  potentially easing concerns among employees about AI replacing their  jobs or making decisions without human oversight.

Overall, adopting a more tailored and domain-specific approach to AI implementation seems to offer a promising pathway to address some of the core challenges highlighted in the RAND report and pave the way for greater success in AI projects within enterprises.

If you wish to explore more cost effective and innovative strategies to enhance the value and reliability of AI while promoting user acceptance and ethical considerations why not think smartR AI and our SCOTi® solution. We are an AI product focused company who work with our advisor and enterprise stakeholders to deliver on time and on budget. We also pre-train only on your data so the only one mining your data is you!