Hybrid Platforms Trends Series - AI - Strategy and Governance 

In part one, we covered a view from the bridge and some high-level trending conversations. We also discussed the complexity of managing an organisation's vast plethora of AI use cases. When we look at that full stack of requirements, only some, if any, organisations could claim to have on-book experts in all areas. This is why we believe an ecosystem of expert partners is the way to ensure speed and excellence of delivery. 

The Partnership Approach - Deeper Insights  

One area in which a partnership can help accelerate outcomes is to utilise specialist AI expertise. 

When we look at the entire stack of requirements discussed in part one, the breadth of skills needed is vast, and usually each area requires specialist knowledge. Without this knowledge, there is a risk that projects will stall or overrun, reducing the time-to-value or resulting in a missed competitive advantage.  

We noted this at a recent customer visit. Their project was over six months behind schedule due to the need for more critical domain expertise in the specific field of AI. The customer had a partner, but they lacked the specialist knowledge to accelerate the outcome.  

This is why, here at CDW, we embrace the partnership approach and believe that combining our core strengths with specialists allows us to help you get the results as fast as possible. When building custom AI outcomes, we recently announced our partnership with Deeper Insight, bringing over a decade of experience spanning Predictive and Generative AI use cases. The team at Deeper Insights bring a unique blend of advisory management consulting and AI domain expertise that aligns with the CDW's view of solving challenges through technology.  

One of the fantastic frameworks that the Deeper Insights Team has developed is how to drive and ensure the successful adoption of AI outcomes within your organisation. These four core requirements are then wrapped up by the platforms that CDW provides.  

Problem Definition: 

The core of this framework is the problem definition and how we ensure it aligns with business outcomes, has buy-in from key stakeholders, and manages any associated risk. Jointly, we see too many ‘cv’ AI projects that were doomed to fail before they started. A clear AI strategy is critical to ensuring any investment will return a measurable return. 

Data 

We all know that data is the core fuel for any AI project. Still, most organisations lack the tools to identify that data and often the skills or time to qualify its suitability to support the desired use case. Combining CDW data frameworks, which can discover and provide contextual access to data, with Deeper Insights' ability to assess its quality (i.e. signal v noise) lets us rate the chances of success at pace.  

Learning Method 

In part one, we discussed predictive vs. generative AI, which became a large part of the Learning Method conversation. The domain expertise within the Deeper Insights team understands these requirements from real-world experience. It will ensure the most appropriate approach to solve your challenge for the lowest cost and complexity.  

Expert Knowledge 

The most forgotten part of the AI journey is the expert knowledge that your people bring to the outcome. Your AI must learn to be effective, and someone must teach it. The best people to do this are the experts inside your organisation. We can drive faster value and mitigate adoption challenges if we align with those experts throughout the journey. 

Platform 

The final part of the puzzle is the platform. Should this be SaaS, Cloud or in a data centre? There is no simple answer to this, as many factors like data availability, performance, cost, and sovereignty will have an impact.  

The learning method chosen and the scale of the project will dictate components like compute, networking and storage. Do we need to train a new model or implement a RAG solution? Where does the inferencing need to take place? What response times does the system need to adhere to?  

All these factors can lead to a web of complexity that threatens the speed of innovation and the value it will bring. Luckily, CDW augments the DI team with decades of experience deploying AI and HPC platforms, allowing us to cut through the noise and define the most effective architecture for the use case. We will dig into some of these considerations in part three. 

Governance  

One area of AI that is at the top of both CDW and Deeper Insights' minds is AI governance. There are many aspects to this conversation, some of which are summarised below: 

• Ethical Principles 
• Legal and Regulatory Compliance 

• Risk Management 

• Transparency and Explainability 

• Monitoring and Evaluation 

• Human Oversight 

• Training and Education 

• Innovation and Continuous Improvement 

At its core, we will return to the age-old trio of People, Process and Technology (PPT) and how we build an operating framework to ensure AI is adopted at the speed needed without breaking core legal or ethical guardrails. The new laws coming into effect around the globe make navigating the implications challenging, and the potential fines will make GDPR look insignificant. This is not just a problem for organisations building custom AI solutions but for all leveraging AI in its many forms. We believe it is critical to every organisation's AI journey. 

Enforcing and monitoring a solid policy and process will also take good tooling. Ensuring all use cases are captured enterprise-wide, compliance is tracked, and benchmarking against global legislation is not a task for an Excel tab!  

Avoid the news 

One outcome of setting a robust AI governance framework is avoiding being the next news story about AI gone wrong. Googling will reveal many stories of AI issues, but the DPD incident shows how a lack of governance and controls can lead to unexpected outcomes. The risk to reputation is hard to ignore and highlights why building a well-governed private AI solution will be the final route for many organisations.  

We discussed different model sizes in the first part and learning techniques earlier in this part. One thing to consider is using smaller models with a more limited scope for specific use cases. The main problem that can happen is that the model deployed is capable of far more than it was being used, opening the risk of a jailbreak and threat actors using it for unintended purposes. This will also link to our third part when discussing the compute power needed to run these models. 

 Summary: 

When we combine domain expertise from partners like Deeper Insights with our heritage in Platforms, Data and Applications, we can orchestrate outcomes in timescales that would otherwise not be possible. The three core conversations (Strategy, Governance, and Platform) must be addressed in parallel to avoid innovation and speed being halted by an unexpected roadblock. Navigating this journey is vital to unlocking the full potential; this is why we developed our journey to success, as shown below, allowing organisations to map current capability and deliver tangible outcomes.