Forecasting Randomness

Three months ago I wrote an article about forecasting seasonally trending data with a recency bias. We run an advertising trading engine where volumes fluctuate both seasonally and upon ad spend decisions influenced by ROAS and the business performance of our customers.?

Over the year end break, I have been thinking about enhancing the accuracy of our forecasting engine and extrapolating our use case to other similar business models.?

It occurred to me that any trading engine where revenue is generated by volume, and volumes are volatile, fit into the cohort: stock markets, advertising, retail (platforms, not retailers), travel, food delivery platforms (such as Uber Eats), etc.?

It also occurred to me that AI and Web 3.0 are going to move more business models into this consumptive and volumetric space, and my experience transitioning from fairly predictable revenue towards randomness will echo that of many, if not most CFOs. Consider:

Revenue (In)security

When I began my career at StatPro plc, we signed multi-year software contracts. Our company spent years developing a system, months implementing it and we could bank on 97% recurring revenue with at least 60% of that renewing beyond 12 months. Moreover we were paid annually in advance and so the business had almost perfect revenue certainty over a three year horizon.?

In 2007, we were a very early adopter of cloud computing in our sector and the business model shifted to annual and monthly subscription. Our platform was still very sticky but the moat moved from contract to product: our customers could more easily cancel and transfer to other systems but their investment in data and our product’s leading market fit secured the 97% ARR.

From 2014 when I joined Naspers, I have worked on a securities trading engine, e-commerce and advertising platforms where the revenue has become volumetric. Something in live trading markets energises me but I have noticed a steady increase in the level of competition and volatility brought about by cloud computing advancing to AI and soon to Web 3.0.

In Web 3.0 Smart Contracts further liquify revenue. In 20 years we have gone from multi-year security to monthly subscription to conditional contracts.

Acceleration

In my domain AI and algorithms have acted as a creatively disruptive accelerant, where the algorithm is largely driving markets:?

In stock trading (and advertising), complex algorithms now execute orders at high speed (thousands of times per second), precision, and volume. Whilst it has improved market efficiency by reducing bid-ask spreads and increasing liquidity, it has also led to challenges such as increased market volatility, flash crashes, and systemic risks due to automated decision-making. Human traders now focus on overseeing algorithmic systems, developing strategies, and managing risk, as algorithms handle most routine trading tasks.?

In E-Commerce, AI and recommender systems personalise customer experiences, analysing behavior, and delivering tailored product suggestions. They enhance engagement, drive sales, and optimise inventory and pricing. These capabilities are now driving, rather than responding, to consumer behaviour. Consider Temu which prioritizes items with higher purchase probabilities while cross-selling complementary products. Additionally, the platform uses behavioral nudges, such as social proof (e.g., "X people bought this today") and gamified rewards, to encourage purchases, effectively blending personalisation with psychological triggers to drive sales.

Forecasting

Which brings me to the purpose of my article: forecasting in the coming age of randomness.

For finance leaders such as myself, forecasting plays a pivotal role in positioning for success. Precise projections enable effective resource allocation, risk management, and capital planning. Consistent accuracy enhances credibility, supports market valuations, and mitigates volatility, fostering long-term growth and shareholder trust.

I personally don’t think predicting revenue to the level of accuracy required to maintain credibility and trust in my numbers is going to be possible without employing algorithms that respond to randomness - just like the human traders in financial markets, finance leaders won’t keep up with the algorithm.?

Furthermore, I’m also moving away from scenario planning and expected returns (which only work where one can predict probabilities) towards a bimodal strategy used by options traders (which are designed to underwrite randomness on the one hand with stability on the other). This approach is well articulated by Gartner and is employed by options traders to hedge unknown probabilities with opposing payoffs.

Sharing the Load

If you are a CFO or Finance Leader that is currently or about to work on business models where probabilities are turning stochastic, I would love to connect and form a group to solve these challenges together, please DM me as I will be forming a closed group of collaborators to share methods and experience.?

As a lover of technology and the energy of liquid trading platforms, I am personally excited about these trends coming to financial planning and analysis but recognise that there will be a lot of new problems to solve, without a textbook to draw from.