When I started doing Data Analytics, we used SQL to extract data, then fed the results into Excel and produced the Pivot Tables and Charts that were required. The focus was on understanding what had happened. This looking backward approach then led to people attempting projections: “Last month we saw sales of £5000, so this month we can expect to see sales of £6000”. Logic and reasoning were often missed. It may be that there was a promotion running that was not present in the data, or it could be based on seasonality. The point was that when you looked at the decks produced in those times, the reasoning behind projections was often missing.
This logic gap is what Data Analysis has been trying to bridge for decades now. The panacea, of course, is to have Prescriptive Analysis – that is to say, something that can look at past performance and not only predict future performance but give a path to achieve that. What we fail to realise is that this is an anathema. If it were possible to achieve this level of analysis, then the bookies would be full of multi-millionaires.
In the real world, we must understand more about the “What Happened” or Descriptive Analytics. These are based purely on what has happened in the past, so they are seen as fixed or “locked in place”. The person at the bookies will look at the past performance of a horse and the conditions in which it performed, so we can see that Descriptive Analytics are vital. The acts of Predictive Analysis – forecasting future performance based on historic data – must always be based on that past performance, Descriptive Analysis.
Prescriptive Analytics are seen as the holy grail of Analysis, where you can use your analysis of past performance to explain exactly what actions should be undertaken to achieve the desired result, so a 5% increase in sales Month over Month. All too often at present, we are seeing it being presented as something that is within the gift of the new version of [insert tool name here] now with AI. This could not be further from the truth. By all means, the new version of the tool may have a historic backlog of data to evaluate and an increasingly robust set of algorithms to apply against it; however, does that mean that it is correct for your organisation? Sadly, the answer is no.
Truth one about a new AI tool is that it will be at its worst on Day One, meaning that for your business, if you are an early adopter, you will be getting the least accurate version of the AI. It will have been “trained” on assumed correct data up until that point, and only when it starts to be used will it be using data that is definitely from the correct use case. The results must also be fed back into the learning engine, or the tool will not be learning. Privacy concerns are blocking that feedback loop in premium AI suites, while the breadth of free services may prevent learning from being truly effective.
Microsoft’s own Copilot for Power BI (ha) is a prime example of this, as developers are being told to simplify models to be closer to the textbook Kimball Schema that the engine is more familiar with. This suggests that the real-world data was too varied for the core tool to understand and learn from. Learn is an interesting term, as this version of Copilot is not supposed to have the feedback loop that would be needed to fully update the training of the model.
Truth two is that we cannot afford AI. The cost of AI is on the increase, with people using a significant amount of cloud CPU to perform searches. Looking back at Microsoft Copilot for Power BI (HA), again, the anecdotes shared seem to suggest up to 1% of capacity performance is used for “conversation”. That is not a problem if you have a single user of your F64 environment, but the reality is that most people will have 200+ to justify the cost. If everyone interrogates the AI at peak times, 200% utilisation may become a problem, either financially or in terms of performance.
It would be easy to assume that we at Geordie Consulting are against AI tools, but the truth couldn’t be further from it. We use AI regularly within Geordie Consulting, and Copilot Studio is something we use to build internal tools to support our operations. However, AI is being pushed on our customers like some kind of magic tool that will save their businesses, while also doing something suitably heroic. The reality is that unless you start by ensuring your Descriptive Analytics are perfect, you risk imperfections being amplified as you progress through Predictive Analytics and into Prescriptive Analytics.
Geordie Consulting works with clients to first maximise the data utilisation across their organisation to better understand the Descriptive Analytics and identify the drivers. This then provides a more solid and stable foundation for us to bring Predictive and then Prescriptive Analytics into your business. Our sole objective is to maximise the value you get from your data. Use your Business Intelligence to build intelligent solutions for your Business.
