top of page
Search

Decision Intelligence in Modern AI (pt.1) – The True Driver of Value

  • petervermeulen9
  • Feb 2
  • 7 min read

AI integration and the ways in which AI generates real value are not new. They have been built, tested, and refined over the last 80 years — and the fundamentals are not changing anytime soon.

 

That statement may rustle a few feathers, particularly among newcomers and start-ups positioning AI as entirely uncharted territory or as a means of wholesale human replacement. History tells a different story. AI has been embedded in business processes since the mid-20th century, and the lessons learned from those decades of implementation are not only still relevant today, but more critical than ever.


So why focus on Decision Intelligence rather than AI integration? Because in practice, they are largely the same thing. AI has worn many labels over the years, but its true power has never been about machines “thinking” — it has been about helping humans learn, reason, and make better decisions. The heart of modern AI is decision intelligence, and if we want AI that genuinely drives value and integrates into day-to-day work, this is where the focus must remain.

 

AI only delivers lasting value when it augments human decision-making. Implementations that ignore this principle tend to fail.


The most successful AI systems are not external, novelty tools, but capabilities designed to work alongside us — enriching our work, improving judgement, and allowing humans to focus on what we do best. This is why anyone considering AI adoption must first understand the foundational principles and hard-won lessons of decision intelligence.


In this article, I introduce Decision Intelligence as the cornerstone of modern AI implementation — exploring its definition, its history, and the pitfalls we are repeating today. In future articles, I will outline a repeatable Decision Intelligence framework and show how generative and agentic AI expand, rather than replace, our decision-making capabilities.

 

What is Decision Intelligence?

Decision intelligence (or decision science) is a practical discipline that advances decision making by explicitly understanding and engineering how decisions are made and how outcomes are evaluated (Gartner, n.d.).


What does this mean for us? Well, without being too reductive it’s worth realising that for a business or an individual, value is driven by expert decision making:  “Should we increase/decrease the price of our product?”, “How can we best market ourselves to reach customers/clients?”, “What is the future going to hold for us and how do we act now?”

 

“Value is driven by expert decision making”

 

The list goes on of course, but ultimately a business’ performance is based on their experts making these decisions, and if the decision making process can be improved with deeper insights, guidance or automation then this has a direct line to cost saving and revenue generation. Of course there is much, much more that goes into business performance but decision expertise is the main driver.


This is where decision intelligence steps in. If we can understand the business operations, the customers/clients, the services provided and the markets we can directly improve our decision making.

 

The History of Decision Intelligence & AI


A major movement that created the foundations for decision intelligence came during and after WW2 when ‘Operational Research’ (or operations research) was first coined (Britannica, n.d.). This is a branch of applied mathematics that deals in the development and application of analytical methods to improve management and decision making.


Some of the early applications include the development of radar systems and applied decision making in war scenarios (game theory), but later proved its worth in organisational settings such as systems and training in the military and convoy systems to reduce shipping losses etc. From the clear benefits that arose from operational analysis, qualitative management and decision intelligence, Operational Research practices were moved to business applications post war with constant development and refinement. Optimisation and Machine Learning fundamentals were created and with the invention of computers, Operational Research could now solve hundreds of thousands of problems assisting in human decision making and generating huge value.

 

This ultimately set the foundations for AI applications from then until now and surprisingly, whilst capacity and tools have changed, the fundamentals stay the same. AI was invented to assist in human decision making and its best applications today still hold to this fundamental idea.


With ‘AI’ as a concept now encapsulating optimisation, machine learning, deep learning, statistical modelling and now generative solutions, we are in a particularly good position to generate value, but only if we do it the right way.

 

Why Are Current AI Implementations Going So Wrong?

What we’ve learned from all this is that if we build the right tools to augment our decision making and improve the accuracy and efficiency of our day-to-day operations whilst keeping accountability, transparency and guardrails, the sky is the limit! That being said, why is it that we are not seeing this consistently? Why is it that 95% of all AI projects created by businesses remain do not progress from the pilot stage? (McKinsey & Co, n.d.).

 

95% of all AI projects do not progress from the pilot stage.

 

Right now, what we are finding is that AI is being implemented without a decision focus. What I mean by this is that fundamental best practices designed in the last 80 years of decision intelligence are not being followed:

  • AI models are created without a plan on how they will assist end users.

  • No integration plans of how AI solutions fit into workflows.

  • No incentives to use the solutions or fear of the end solution replacing humans.

  • No accountability – if the AI model recommends the wrong solution, who vets and is ultimately accountable?


Some real-world examples of these are currently all over the news (and there will be a lot more of this to come). Remember the lawyers using GenAI to write case documents? There were no controls or accountability so the LLM did what it does best and invented a bunch of realistic sounding cases that never actually happened. Accountability was assigned pretty quickly when the lawyers were sentenced by the courts. Ultimately, though the AI created the legal documents, the lawyers were accountable for their content.


Another great example I got from my neighbour. He visited an Apple Store to get a quote for replacing his damaged iPhone and the staff there who are specially trained in technical repairs were at the time being forced to use a new AI system. This tool works by receiving a photo of the damaged phone, then using image recognition and deep learning, recommends the overall value of the phone and its replacement costs. A great bit of technology - implemented terribly (even by a tech leader). The staff felt undervalued with their years of training supplanted by a new system. They were also in fear of being replaced by this forceful implementation, therefore they would take pictures of the damaged phone in such angles, that the damage did not look as bad, and the clients would get a higher refund/lower fix costs. A minor AI rebellion that will cost the company millions (and more) all due to fundamental mistakes in AI integration. The list goes on but since this is an article I will try to contain myself…

 

How Should We Implement AI? – Lessons From Decision Intelligence


There is a huge framework of how to implement AI safely, securely and ensure its built to deliver value and solve actual problems. In the next article (part 2) I will delve deeper into this framework, build out ways in which AI should  be implemented and provide case study examples of where things worked well and where mistakes were made.


Since this is just an introduction we’ll posit some key things to think about before getting too much into the detail. If you are considering AI for your business, here are some main decision intelligence learnings to think about before you progress further:


Understand Revenue Driving Processes and Tasks

First off you must understand the business operations, especially those that are driving value for the business, the best areas for AI implementation are what they have always been:

·         Price Optimisation/Revenue Management – Optimise the revenue coming into the business be it pricing for goods or services or other decisions.

·         Supply Chain/Inventory Management – Optimise operations to reduce costs/ loss of customers/clients.

·         Digital and Marketing Spend – How to best spend money to make money -improve brand reach and image.

·         Customer and Product Insights – Understanding your customers and products so you can adapt to best serve and expand your customer base.

If you can understand what drives value in your business and what decisions help/hinder this, you can increase your revenue and operational efficiencies.


Augment Current Decision Makers

How can you augment the current decision makers and improve the efficiency and validity of their choices? If you understand their day-to-day operations, you can see which processes can be improved, what insights or modelling will help them react quicker or increase decision validity. Whatever is built, it should serve the end users, therefore they need to be involved throughout the AI design, build and rollout phases.


Ensure Accountability

Throughout the process of AI building and integration there needs to be assigned and agreed accountability, who ensures that the solution is fit for purpose? Who ensures that end users are trained and happy with the solution? Who is in charge of testing its validity? etc… Without these agreed at the start of the project the holes will be felt later on.


Learn and Adapt

Test current business assumptions – you may learn something is more/less important than originally thought, which may make fundamental improvements to your business operations. Maybe competitor pricing is less important than you first thought, maybe there is a demographic of customers that do not value a product as much as you estimated… Science is about learning and adapting to what’s learned - AI systems need to be the same.


Start Small

Most AI projects that succeed start small and work their way up. Change management is a fine art and requires attention and commitment to implement change that is beneficial and lasts. Start with some simple insight models or automation for a limited scope (partial product or customer base), measure the improvements, learn from mistakes and generate trust and faith incrementally in the systems you build.

 

Right, we’ll leave it there for now. This is an area I feel passionate about due to it being a pillar of my work for the last decade in implementing AI solutions for businesses.

If you wish to learn more privately please feel free to contact me for a free impartial consultation and we can discuss the correct way for you or your business to implement value driving data and AI solutions.

 

Thanks,

Pete Vermeulen

Founder – Head of ML & AI Solutions

Plum Tree Solutions

 

 

 

References

Britannica. (n.d.). History in Operations Research. Retrieved from https://www.britannica.com/topic/operations-research/History

McKinsey & Co. (n.d.). The state of AI in 2025: Agents, innovation, and transformation. Retrieved from https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai

 

 
 
 

Comments

bottom of page