Improving Restaurant Forecasting with a Driver-Based Process

Episode # 161: Restaurant forecasting can feel impossible when demand changes fast and yesterday’s playbook stops working. Sales might be steady while traffic quietly drops, or a shift to delivery lifts revenue but creates new labor, inventory, and margin problems. We want forecasts that explain what is happening, not just a number that misses with confidence.

We walk through a driver-based restaurant forecast built on what leaders can manage: traffic, average check, and mix. That means getting painfully clear about definitions like transactions vs guest counts, then showing how pricing, discounts, add-ons, day part, and channel mix move the check. We also dig into why forecasts fail in the first place: treating stockouts and closures like normal demand, assuming last year is automatically comparable, and letting different teams run on different assumptions. If you only measure total sales accuracy, you can miss the real problem and never learn from forecast error.

From there, we talk practical AI for restaurant demand forecasting. A useful machine learning approach can forecast transactions by restaurant, day part, and channel, then estimate check and menu mix to project item quantities for purchasing and prep. Inputs like POS history, menu changes, promotions, loyalty data, weather, and local calendars matter, but testing matters more. We share how to backtest models under realistic conditions and why you should compare AI results to both your current manual forecast and a simple seasonal baseline.

If you want a restaurant forecasting process that improves staffing decisions, reduces waste, and makes promotions less of a gamble, this is for you. 

Episode outline:

1. Why a dynamic forecast matters more than ever,
2. Build the forecast from its critical drivers,
3. Develop a more robust forecasting process and continue to improve it. 

Please connect with me on:

1. Instagram: stephen.mclain
2. Twitter: smclainiii
3. Facebook: stephenmclainconsultant
4. LinkedIn: stephenjmclainiii

For more resources, please visit Finance Leader Academy:  financeleaderacademy.com.

Transcript

Copy Transcript Copied

Why Restaurant Forecasts Feel Hard

Stephen McLain 0:04

Restaurant forecasting is difficult because demand can shift quickly, and a sales total by itself rarely explains what is happening. Revenue may be on plan because menu prices or average check have risen even as customer traffic weakens. Or a change in product, channel, or day part mix may lift sales while creating different staffing, inventory, and margin needs. In this episode, we'll break restaurant forecasts into the drivers leaders can manage, like traffic, average check, and mix. We'll look at how weather, household finances, promotions, local events, and operating conditions affect those drivers. Then we'll discuss why forecasts miss, where AI can improve the process and how to test models so that better predictions lead to better decisions. Please enjoy the episode. Welcome to the Finance Leader Podcast, where leadership is bigger than the numbers. I am your host, Stephen McLain. This is the podcast for developing leaders in finance and accounting. Please consider following me on Twitter, Facebook, Instagram, and LinkedIn. My usernames and the links are in this episode's show notes. You can also follow Finance Leader Academy on LinkedIn. Thank you. This is episode number 161, and I'll be talking about forecasting in the restaurant industry and I will highlight the following topics. And three, develop a more robust forecasting process and continue to improve it. Financial historian and educator Peter Bernstein said forecasts create the mirage that the future is knowable. We are tackling forecasting this week in the restaurant industry. We know that forecasts are often best guesses for what we believe may happen in the next period, the next quarter, and even the next year, but there are so many variables to consider even when looking at last year's performance. There are economic variables, historical variables, ever-changing trends in dining and tastes, and health choices. And there are generational variables that continue to shift. We walk through how to build trust with shared facts without forcing everyone into the same system. You'll hear what shared facts actually look like in practice, clear metric definitions, authoritative sources, consistent timing, known limitations, and named data owners. So please listen if you have not already. Enjoy.

The Hidden Reasons Forecasts Miss

Stephen McLain 3:08

How can we make a better forecast? Let's consider why our forecasts often fail in the first place. Forecasts miss when teams assume the past will repeat without adjusting for changed conditions. Now last year's comparable week may have had different weather, school dates, prices, promotions, hours, or local events. Treating a stockout or temporary closure as ordinary demand can also train a model or a manual process to expect the wrong sales level, and if finance, operations, marketing, and supply chain use different assumptions, the forecast can be internally inconsistent even before the month begins. Now many organizations track only total sales accuracy. Errors at busy locations or peak day parts can be hidden when other errors offset them at the company level. Without ever tracking over or under forecasting, investigating exceptions and assigning owners to improve the next forecast, the process records misses but does not learn from them. I

Where AI Can Improve Accuracy

Stephen McLain 4:20

want us to consider the application of AI to assist in better forecasting. AI can help analyze many interacting signals across locations and also time periods. A practical model might forecast transactions by restaurant, day part, and channel, estimate average check and mix, and then project item quantities for purchasing and also for prep. Inputs could include point of sale history, menu and price changes, promotions, loyalty or reservation data, channel mix, weather, and local calendars. A 2024 study conducted with a large US restaurant chain examined how internal and external data and different machine learning approaches could support demand forecasting. Its practical lesson is to test data and model choices across different market conditions rather than assume that the most complex model will always perform best. If your restaurant is relying on last year's sales plus a growth target, it may be time for a more useful forecast. McLean Solutions can help your team connect traffic, average check, menu mix, and local demand drivers to better decisions about staffing, purchasing, and growth. Reach out to McLean Solutions to schedule a strategy session and strengthen your forecasting process. Now

Why Dynamic Forecasting Matters Now

Stephen McLain 5:48

let's talk about forecasting in the restaurant industry. Number one, why a dynamic forecast matters more than ever. Forecasts often fail because the process starts with a sales target or a simple percentage increase instead of a clear set of operating drivers. A top line estimate may be adequate for a high-level budget discussion, but it will not tell a manager whether to adjust staffing, prep, or purchasing. A second common problem is inconsistent or incomplete data. Transactions may be missing a channel, item identifiers may have changed, and promotions, closures, stockouts, or service interruptions may distort the historical baseline. Restaurant leaders need forecasts that keep pace with a consumer environment that is uneven across locations and also customer groups. As one current indicator, the National Restaurant Association's August 2026 tracking survey found that 51% of operators reported higher same store sales year over year, while 43% reported higher customer traffic and 46% reported lower traffic. That gap is a reminder that sales growth does not necessarily mean more guests are coming in. A forecast should help leaders tell whether results are being driven by traffic, check growth, price, or a shift in what and where customers buy. Affordability pressures also matter to the forecast. In the association's third quarter 2026 survey, 40% of consumers said they were relying more on discounts and value promotions, while 34% said they were ordering fewer add-ons such as dessert and beverages. Those behaviors can affect both traffic and average check, and their impact may differ by market, customer segment, and restaurant format. A few points to always consider sales growth and traffic growth tell different stories. A forecast should support decisions about labor, food purchases, pricing, and promotions. The same economic lift can affect restaurant formats and customer groups differently.

Separate Traffic From Average Check

Stephen McLain 8:15

Number two, build the forecast from its critical drivers. Start with a clear definition of traffic. Depending upon the format and the available data, the operational measure may be guest visits, covers, or transactions. Now for a transaction based view, a simple starting identity is net sales equals transactions times net sales per transaction. If leaders use guest counts instead, they should pair them with spend per guest. The important point is to define the measure consistently across locations, time periods, and also channels. Now average check is not one driver. It can change with menu pricing, item mix, party size, add-ons, discount use, day part, and also channel mix. A chains blended average check can move simply because more sales came through delivery or a higher check day part. Forecasts should therefore show how much of expected sales come from volume and how much comes from check. Then explain what is expected to change within the check.

Add Weather Economics And Local Events

Stephen McLain 9:34

Number three, develop a more robust forecasting process and then continue to improve it. Historical sales provide the starting point, but they are not the full forecast. Calendar patterns such as day of week, holidays, school breaks, and seasonality need to be adjusted for changes in opening hours, closures, remodels, menu availability, and promotional calendars. Pricing and promotions should be reflected explicitly. A promotion can draw incremental guests, shift demand from another day or product, or change the net check through discounting. External drivers should be chosen based on the restaurant's customers and trade area. Weather is a good example. Now rain, heat, snow, or severe conditions may change dine-in demand, drive-through volume, and delivery in different ways. The effect can vary by market, day part, and format. So leaders should test local patterns instead of applying one universal weather adjustment. A restaurant demand study has examined forecast using menu level sales and also weather data illustrating the potential value of weather inputs while also showing while item level predictability must be checked. Economic conditions can also influence visit frequency, check size, and channel choice. Useful signals may include local employment, wage trends, household income, and disposable income conditions, fuel prices, consumer sentiment, and competition for food spending from grocery stores. Other local factors can matter just as much like concerts and sporting events, tourism, school calendars, construction, road access, nearby employer schedules, competitor openings, and major community events. National economic data can help establish context, but local results should determine how these signals translate into a particular store's forecast. Now I've previously worked with a national restaurant chain. Forecasting has evolved significantly since then, but I do remember that we did work together closely as a team to develop the initial annual forecast, which I believe in doing. Anytime you collaborate as a team, when it is done properly with involved leadership and very good guidance, you will get a better and more accurate product. Now for action today, take the next four-week forecast and identify whether the largest expected sales changes come from traffic, average check, or mix. Choose one location or day part where an external factor may be affecting demand. Then compare the forecast with a simple baseline before changing the process.

Quick Actions And Better Model Tests

Stephen McLain 12:39

Please subscribe to the podcast on the platform you are currently listening to, and also subscribe to my weekly email. When you subscribe to the email, you will receive a free guide about developing your finance leadership. It's filled with many tips and strategies to grow your leadership. Thank you. Now today I talked about forecasting in the restaurant industry and I highlighted the following points. Number one, why a dynamic forecast matters more than ever. Number two, build the forecast from its critical drivers. And three, develop a more robust forecasting process and then continue to improve it. A restaurant forecast becomes more useful when it explains the business in operational terms. Start by separating traffic from average check, then examine how mix, pricing, promotions, weather, household finances, and local conditions may change each driver. AI can help identify patterns across that information, but only if the data are reliable. The model is tested against realistic conditions, and managers can act on the result. AI is most useful when it turns the forecast into an understandable operating signal. Leaders should be able to see what changed the prediction, such as a holiday shift, a weather event, a promotion, or a recent traffic pattern, and whether the change affects expected sales, staffing, or item demand. Predictive models can generate the numbers, generative AI can help summarize exceptions or let managers ask questions about the forecast. Those outputs should still be checked against the source data and operating context. Model testing needs to mirror how the forecast will actually be used. For example, when backtesting weather-driven demand, use the weather forecast that would have been available when the decision was made, not the weather that actually occurred. Compare models with the current manual forecast and a simple seasonal baseline using rolling, time-based tests.

Wrap Up And Next Steps

Stephen McLain 14:47

Track error and directional bias at the levels managers use, such as location, day part, channel, and item, not just in the aggregate. Now I hope you enjoyed the Finance Leader Podcast. If this episode helps you today, please share with a colleague and leave a review. Please check out FinanceLeaderAcademy.com for more resources and for ways that I can help you and your team. And now go lead your team and I'll see you next time. Thank you.