AI Use Cases – Why AI and Business Planning should be Happy Bedfellows

Overview

I have always been of the use that the most powerful use cases for the introduction of AI should be predictive in nature. Of course, many, many forms of AI apply different techniques to derive the probability of a particular outcome, and thus deliver predictive outcome, in many cases, the exact extent to which the propensity to take a particular action, as actually taking the action, isn’t nearly as predictive as one hopes.

There are, as a result, a few key implications of this.? First, it signifies that the size and frequency of the error necessitates judgement, and thus, at this point at least, human intervention. Second, it skews the application into production of use cases significantly away from problem solving areas where the challenge is both arriving at the right inputs as well as determining the right set of AI techniques to acquire the right outcome on a repeatable basis.

These conclusions haven’t of course stopped many new firms being funded to tackle problems where the need to quantify human decision making is high, for example in areas like client acquisition and retention, but it has raised in me a high level of scepticism toward marketing claims that promise real performance improvement, rather than the possibility of changing process productivity, and in turn, a theoretic improvement in the odds when it comes to recognizing the “right signs and signals”, rather than the actual outcome itself.

Challenges with BP Forecasting

I have been led over the past week to ponder this as a result of work I am undertaking with trying to build a cohesive business plan for the next 5 years in order to establish a true baseline for one of my clients.? Generally speaking, building such a plan is usually oriented around applying assumption on top of well established trends in terms of sales, cogs, margins, and well understood operational expenses, alongside a relatively well developed data set of future business prospects and conversion objectives.? Inevitably with such an approach, one discovers as the visibility toward the future becomes more fuzzy, hopes and biases, when it comes to future performance start, to become features imposing their own aspirations on the plan. When there are too few of these expectations expressed, external investors start to wonder about the transformational opportunity associated with different contributing strategies, while if there are too many, investors might be put off by too many risks and unknowns. Thus, it is as much art as science to arriving at the right balance, and with it, both a both achievable and believable outcome.

AI as a tool to Overcome Issues

AI has been raised as a possible way to arrive at a more robust forecast, but a lot of what I have seen often seems to rely on the past being a strong indicator of the future as well as suggesting that similar attributes can be relied upon to produce similar results.? I have some familiarity with the pro’s and con’s of building models in this way, and often have recognized that indicators that signal positive as well as negative sentiment, esp. when consistently referenced, carry significant predictive weight, but unfortunately, in my experience, they don’t often do a great job in spotting the transformational moments early or often enough.? Thus, it would seem to be the case that if these types of indicators become the core inputs for an AI model, the likely outcome will probably be only moderately better than the more basic and na?ve method.

How might we rectify this? I believe the answer lies in extending the nature of the data set so that it can bring into the modelling inputs that can express the influence of human behaviour as well as decision taken that clearly reflect the intersection of rational and emotive responses.? The inputs that arise from this aren’t always about taking action, but also capture inertia, continuity, and regression which are also legitimate responses to both risk and opportunity.?

Predictive Power through Behavioural Data Scorecards

Thinking in this way, particularly when building a business plan from a bottom/up perspective, allows one to think about each item as embodying a quantitative signal to represent the different states that could be associated with any aspect of the future income statement, and related to it, the other key deliverables required for business planning. As an example, to accompany a particular financial projection of both contracted and non-contracted revenues, one could additionally introduce measures related to the ambitions, and guiding principles of leadership and key stakeholders including gatekeepers, the quality and breadth of the dialogue, and the process steps that a client actually goes through when determining if it will be investing or not, and how it will then allocate funds.

Much of this type of unstructured data or derived judgement has always been theoretically available to make use of in the forecasting process, but more often than not has been inaccessible, partially because of where it resides, but also because organizations made little effort to acquire them in a coherent or consistent way. This might seem purely like a technology issue, but it also has to with the overarching understanding of “data as a fuel”, with both a lack of purpose as well as incentive. Fortunately, the advancement in computer vision, natural language processing, and semantic modelling through LLM, now offers the promise to not only identify, extract and structure at scale content of this design, but asl new forms of processing that can score sentiment, identify inference, and measure the difference forms of response to risk and opportunity.

Conclusion

For people with my strategic and forward planning role, this represents a potential game changing scenario for business planning, but will only happen when everyone involved in finance, operations, and commercial affairs, not to mention, overall leadership, also gets on board with the program and supports the type of investment needed in the entirety of the data fabric.? These types of programs take a couple of years to start generating visible ROI in a number of different domains, but based on where AI modelling and data science is heading, the area of business forecasting and alongside it strategic capital management may well prove to be the two most fruitful domains for SMEs and corporates to pursue and deploy into their planning process.

?

?

?