Prerequisites for FP&A Prediction Machines and Beyond

1. Lead the organization from gut feeling to a combined data assessed decision making. Assessment is preferred to “driven” as it implies active participation of people in the decision-making process. FP&A as facilitator is responsible to assess and understand the drivers of the plan. The application of algorithms, esp. the non-traceable ones, adds them to this responsibility. An understanding of the basics, potentials and esp. limits of statistics and artificial intelligence should be present or quickly build and shared by FP&A.
2. Linking strategy to operations via a consistent set KPIs bridges aspiration and execution. Google and MIT published a paper about marketing readiness and machine learning. Advertisement produces a lion share of interaction data earning google close to 30bn in revenue. 30bn USD per quarter. Obviously, plans to introduce machine learning offerings are elaborated. It plays nicely, the platform generates all data within. The study finds Measurement Leaders to differ from Measurement Capable & Challengers in actively developing an integrated view on the customer via KPIs while Measurement Challengers differentiate themselves by relying mainly on the gut feel in decision making (see point above). With good strategic KPIs on top and the option to drill down and through from there, the organization aligns. The valuable insight for planning is to provide this ability to drill down and through the metrics unfolding the hidden story to base actionable decisions on. This was highly significant in the Leader category with only 10% of “gut guys” Laggards seeing this a necessity. (source MIT Sloan)
3. Go beyond the financials and become a partner. The focus of the departments differs, so do the underlying data sources. Understand their requirements and integrate them into the financial planning flow. This increases the quality of the discussion and focusses them on the business drivers that may even including some of these sophisticated concepts we want to see applied in FP&A. One example may be online subscription revenues over time who are adequately represented by statistical models like “buy ‘til you die”. This black box of future subscription revenues turns the predictive and prescriptive initiatives within finance upside down. The application of AI is equivalent to an operationalizing away from FP&A. With an efficiently automated AI model, the discussion shifts to the business acumen driving the financials. (read more from Wharton)

[bctt tweet=“The caveat in the data assessed strategy imperative is the more strategic a decision, the less relevant operational data is for the narrative. “ via=“implexa“]

1. Machine Learning differs from statistics as it cares less about hypothesis, bias and confidence and more about providing operational efficiency superior to the existing process. The dominance may evolve via learning and improvement over time providing insights to the monitoring instance. Extending point one from above, the skills to implement machine learning are to be developed and shared within the company.
2. Machines require data for training, predicting and feedback. Obtaining data comes at a cost. Balancing cost and (prediction quality) benefits is a core function of our profession. The need of data acquisition can be guided with a KPI framework and strategic narrative. Reflecting on the second point, AI will drive only parts of the decision-making process. Knowledge of the diminishing return of additional data is crucial in taming the flood of data. More data is neither causal nor correlated with better. The knowledge on the thresholds of statistical usability, regulatory and competitive aspects require constant experimentation and review. This is also to be embedded in the strategy formulation process for optimized decision-making.
3. Prediction machines are efficient if they provide cheap decision making, the envisioned “operationalization” of point 3 above. Identifying the decisions costly-to-make or with potential costly effects are a priority for the use of AI. Value is created once the cost of prediction drop below the current state. Think about cost, quality and timeliness of an automated translation system in a multinational enterprise compared to having translators do the work for documents, meetings and communications.

Summary