City Shift Finance: Your research on operating leverage examines how different cost structures affect operating income forecasts. What did you discover about the limitations of conventional forecasting methods, particularly when revenue declines?
Nathan Mauck: The textbook operating-leverage model is elegant but, in our opinion, lacking a bit in real-world applicability. It effectively assumes costs split cleanly into fixed and variable and move predictably with sales. In practice, many costs are sticky: SG&A and depreciation rise when sales grow but don’t fall back proportionally when sales decline.
In “Predicting Operating Income via a Generalized Operating-Leverage Model,” my co-authors Sherwood Lane Lambert, Kevin Krieger, and I built a practitioner-oriented model around the components that make up operating income: net sales, cost of sales, depreciation, and SG&A. We treat cost of sales as the variable piece and adjust depreciation and SG&A for stickiness. That generalized model improved the accuracy of next-quarter and next-year operating income forecasts compared with textbook operating leverage.
The practical lesson is that the downside is where simple models are most dangerous. When revenue falls, a model that assumes costs shrink in step with sales will overstate operating income, which is exactly when leaders most need an accurate read.
City Shift Finance: When a company delivers revenue growth without a corresponding improvement in operating profit, which financial and operational factors deserve closer examination?
Nathan Mauck: Start with cost structure. If costs grew in step with revenue, operating leverage isn’t working the way the plan assumed, so look for costs that were treated as fixed but behaved as variable. Then look at mix: growth concentrated in lower-margin products, customers, or channels. Then price: discounting that bought volume at the expense of margin. Finally, capacity: growth that required new fixed commitments, such as people, facilities, or systems, ahead of the revenue they are meant to support. The underlying question is whether the growth was the kind of growth the plan assumed or that is even desired.
City Shift Finance: When actual financial performance consistently differs from forecasts, how can finance leaders distinguish between inaccurate planning assumptions and problems with operational execution?
Nathan Mauck: Look at the pattern of the misses. Planning errors tend to be systematic: consistently in one direction, across business units, and traceable to a shared assumption such as market growth, pricing, or input costs. Execution problems tend to be concentrated in specific units or initiatives, while others working from the same assumptions hit their numbers.
A useful practice is to write down, at planning time, the handful of assumptions the forecast depends on most, and check those first when a variance appears. If the assumptions held and the results didn’t, it’s execution. If the assumptions broke, the plan was wrong.
City Shift Finance: Your research on XBRL found that analysts issued forecasts sooner following financial disclosures. What does this tell us about the relationship between reporting speed, forecast accuracy and financial decision-making?
Nathan Mauck: In that study, again with Sherwood Lane Lambert and Kevin Krieger, we compared analysts’ next-year earnings forecasts before and after XBRL made financial statement data machine-readable. After adoption, analysts released forecasts significantly more often, and significantly faster, in the two weeks following 10-K and 10-Q filings. But forecast accuracy didn’t change. Faster data made analysts quicker, not better.
For finance leaders, that’s a useful caution as automation and AI compress reporting cycles. Speed is valuable when a decision is time-sensitive, but it doesn’t substitute for judgment about what the numbers mean. Improvements in decision quality come from better assumptions and better interpretation, not just faster data. The same will likely be true of generative AI: it makes analysis faster by default, and making it better takes deliberate effort.
City Shift Finance: What financial or operational indicators can reveal that an annual budget is becoming unreliable before the deterioration appears in reported earnings?
Nathan Mauck: Watch the indicators that move before earnings do: bookings and pipeline conversion against plan, pricing realization and discount rates, customer churn, receivables collection slowing, inventory building relative to sales, and unplanned headcount or overtime.
City Shift Finance: Drawing on your finance research and experience working with executives, what distinguishes financial reporting that informs major investment decisions from reporting that simply explains historical performance?
Nathan Mauck: Reporting that explains history answers “What happened?” Reporting that informs investment decisions gives some idea of the cash flow generation of the firm moving forward. While the past can be helpful in developing a forecast, you can only invest in the future. Historical reporting is necessary, but it’s built for compliance and control, not for choosing between alternatives.
City Shift Finance: As generative AI becomes more widely used in financial planning, forecasting and scenario evaluation, where do you see its greatest potential, and which decisions will continue to require substantial human judgment?
Nathan Mauck: The greatest near-term potential is in the groundwork: first-pass variance explanations, scenario generation, and stress-testing assumptions by asking the model what would have to be true for a forecast to hold, or where a plan is most fragile. That compresses days of work into minutes.
What stays human is deciding which scenario matters, interpreting what the numbers mean for strategy, and owning decisions with real consequences, like capital allocation, restructuring, and pricing. AI will also produce fluent, confident analysis that is wrong in ways that are hard to spot. The finance leaders who benefit most will be the ones who treat its output as a hypothesis to test, not an answer to accept. Personally, I have found treating the AI like a trusted colleague and keeping it engaged in conversations on what I am working on to be very helpful. The combination of the memory and outside perspective is very helpful.
