AI, where do we go from here?

Your organisation’s AI benefits have been delayed; we apologise that your transformation journey will take longer than planned…

We’re now past peak in the AI Hype Cycle with most organisations deep in the “Trough of Disillusionment”. A significant proportion of organisations have invested time and resources in AI while benefits remain elusive. Where do we go from here?

AI will be transformational, but there are clear reasons why it’s yet to deliver within a B2B context. Understanding these reasons and preparing your teams now for AI adoption will be key to future success:

1. Data: The power of LLMs is dependent on the vast data sets they’ve been trained on. This data is overwhelmingly consumer vs. business focused. Business data sits firmly behind firewalls (for good reason), resulting in models that are designed almost exclusively for consumers.

Action: Prioritising B2B use cases, then organizing and digitizing your data to address them is a prerequisite to AI value. Invest now in data governance and architecture to be ready.

2. Context: Large data sets establish context. AI performs a lot less well when applied to subject matter-specific tasks where they lack that context. Most businesses exist precisely because they have discovered and exploited a niche in the market. To add value, AI models need to understand these very specific environments. This requires thorough business process mapping and an understanding of how decisions are made within your organisation.

Action: “Tuning” models with the broader context of your organisation is key to optimising outcomes. To do that, you first need to map existing business processes and document how decisions are made.

3. Confidence vs. Competence: AI is a people-pleaser. Its tendency to return answers with 100% confidence, only to correct itself when challenged is less of an issue in the consumer world but can’t be applied to business-critical uses cases where trust is non-negotiable. Understanding the subset of use cases where AI may be appropriate and then thoughtfully determining how AI can support human decision-making is key to successful adoption.

Action: Selecting use cases where AI can add value vs. where deterministic models are more appropriate is key. In parallel, determining the level of human oversight and input required up front will result in better outcomes.

4. Black Box vs. White Box: Most LLMs are “black box” with opaque decision-making processes. That’s’ fine for recipe suggestions but doesn’t work for businesses that operate within regulatory frameworks that require them to demonstrate how decisions are made. Deciding which model(s) to use and where those models are hosted aren’t just technology decisions – they will influence which use cases you are able to address vs. those that are out of scope.

Action: Documenting decision-making processes first and embedding these parameters and guardrails into model design is key to regulatory compliance. Embedding these considerations in model selection and deployment approach will ensure that you can address a broader range of target use cases.  

5. Estimating vs. Knowing: LLMs are probabilistic vs. deterministic. That enables them to make informed estimates of what the likely answer to a question might be. The current architecture of LLMs means they can never “know” anything with 100% accuracy.

Action: Ensuring sound judgement by embedding humans in the loop is critical for most B2B use cases. Conducting an inventory of uses cases to understand the degree of autonomy that’s appropriate before deploying AI will lead to better outcomes.

While returns on AI investments so far have been disappointing, the potential upside remains significant. Now is the right time to prepare for AI adoption. This means taking time to understand and prioritise use cases and document current decision-making processes before making conscious choices about how to automate those processes with a combination of people and AI.

Organisations should resist the urge to lead with technology and instead lead with the problems they are trying to solve. Only when these are understood should decisions be made about how to address them. Some will require AI and some won’t.

At EA we’re helping our clients thoughtfully work through these challenges. By deploying our Business Process Diagnostic, we’re able to document our client’s existing business processes, identifying areas of inefficiency which are then addressed using a combination of structural change, role clarity, and business process simplification. Once that’s completed, we selectively deploy AI to deliver automation that drives quantifiable impact.