Turning internal data into better labour schedules
FARE Conference 2017 brought foodservice operators, suppliers, executives and practitioners together in Dallas from 26 to 28 June. The programme connected restaurant, convenience, healthcare, education, recreation and other segments through educational sessions, keynote speakers, supplier exhibits and networking. For busy operators, the useful conversations were often about a familiar pressure: getting enough people on the floor when demand arrives without carrying idle labour costs through quieter periods.
Using internal data to predict peak labor hours and schedule efficiently reframes that pressure as a planning problem rather than a guessing game. Transaction timestamps, covers, order channels and menu mix can reveal when a service period needs more capacity. Those signals become more valuable when connected to employee skills, availability, agreed hours and the rules governing each roster. The outcome is a schedule that supports service and pays for itself, not a spreadsheet built only to reduce one number.
Conference registration offered access to that exchange in one place. Operators could compare experiences with peers, ask suppliers practical questions and hear how ideas from the keynote stage translated into operating decisions. The themes reflected in attendee testimonials centred on practical learning, peer exchange and access to suppliers. That makes the subject especially relevant to anyone managing people, budgets and customer experiences across multiple sites.
Australian operators face distinctive conditions. Melbourne’s MCG fixture crowds and Melbourne Cup week can shift demand sharply, while a quiet Monday arvo can look very different from a school holiday in Brisbane. Most full-time roles are built around a 38-hour ordinary week, and Saturday, Sunday and public-holiday work can attract award or agreement penalty rates. Roster language is familiar—arvo, break, swap and knock-off—but compliance with the Fair Work framework is formal. Local knowledge must inform the model without replacing it.
Why peak hours deserve a better forecast
A weekly sales target cannot show where a bottleneck begins. A café may need extra hands during a short breakfast rush, an aged-care kitchen may face predictable meal-service peaks, and a convenience site may rely heavily on the overnight window. Restaurants can also lose service capacity when the right skill is absent, even when total headcount looks adequate. Forecasting peak periods therefore means estimating workload by role and time block, then comparing it with deployable capacity.
Forecast accuracy should not be judged in isolation. A prediction that adds half an hour of unnecessary cover every day may look harmless in a percentage report while producing a substantial annual cost. The opposite error creates queues, missed meals, stressed teams and abandoned orders. For healthcare and education operations, service risks carry greater weight than a small efficiency gain. For recreation venues, a sudden surge can affect safety as well as service. Each segment needs its own definition of an acceptable balance.
Internal data helps because most businesses already generate it. The first task is to describe a busy period precisely: covers per fifteen minutes, average order time, queued orders, table turns, meal trays or litres of coffee sold. Translating activity into labour demand then requires an understanding of the work. A new team member may need more supervision, a busier menu may require different preparation, and a full dining room may not generate more till transactions per server. A sound forecast connects those operational details.
Forecast accuracy and roster utility are different measures. An hourly forecast might be excellent but still produce an impractical roster if employees cannot work the predicted times. Conversely, a less granular forecast can be useful when it produces stable shifts. Set service targets around queue time, table availability or tray accuracy rather than relying on a single coverage percentage. Those measures connect staffing decisions to outcomes the kitchen, ward, classroom or store can actually see.
The signals already inside your business
Point-of-sale data is the strongest starting point, provided teams can separate timestamps from daily totals. Useful fields include transaction time, order channel, basket size, menu category, modifiers, discounts, refunds and completion status. Reservations, walk-in counts, online orders and drive-through arrivals add context. Loyalty data can reveal regular visit patterns, but it should not become a substitute for measuring every customer. The aim is a clean view of demand, not a portrait of individual guests.
Operational records can strengthen the signal. Prep volumes, stock movements, waste and production times show whether product demand translates into a heavier workload. Staffing systems show contracted hours and declared availability, while historical rosters expose recurring gaps between scheduled and actual activity. Management should compare planned roles with work actually performed, not treat clock-in time as a measure of individual productivity. A person cannot be labelled inefficient because a rush compressed tasks that the model failed to recognise.
Local context can then adjust the baseline. School term calendars, public holidays, severe weather, transport shutdowns and major events can change trading patterns, but the core forecast should still come from the business’s own history. Start with simple comparisons by daypart, site, channel and event type. Regression or time-series methods can capture trends and recurring patterns; machine learning becomes useful where many variables interact. Back-test against prior weeks, measure errors and retrain when the business changes. Clean, consistent data beats a sophisticated model trained on blind assumptions.
Data preparation is where many projects lose momentum. Agree on one clock and timezone, reconcile refunds and voids, distinguish dine-in from takeaway and flag incomplete shifts. Missing values should be recorded rather than silently replaced with zero, because no cover recorded at 7am may mean no demand or no data. Assign ownership for corrections and test the pipeline before discussing algorithms. A transparent weekly file that the manager trusts can outperform a real-time platform with inconsistent inputs.
From forecast to staffing decisions
Demand becomes a staffing decision through role-specific work standards. In simple terms, required labour hours equal forecast activity multiplied by the minutes needed to complete that activity, divided by each worker’s productive time. A bakery, for example, may need different production and counter minutes from a bistro. Add time for handover, cleaning, breaks and a modest uncertainty buffer rather than relying on perfect attendance. The output should be a range, with a recommended roster and a clear explanation of the busiest intervals.
An optimisation step can then place people where they add the most value while respecting constraints such as certifications, role capability, contract hours, availability, maximum consecutive shifts and agreed rest. This is workforce planning, not an automatic employee instruction. Managers need to approve the result, consider skill cover and protect stable hours where possible. The Fair Work National Employment Standards and relevant modern awards or enterprise agreements determine ordinary hours, overtime, breaks and penalty rates. A model that recommends unlawful hours has failed, regardless of its forecast accuracy.
Fine-grained forecasts still need sensible shift shapes. Creating a new fifteen-minute shift for each predicted spike fragments the day and can increase travel and fatigue. Combine adjacent demand into practical blocks, preserve meal breaks and give team members notice of changes. A forecast-led roster should improve the working day, not merely fill a grid. Track roster stability alongside cost: frequent swaps may erase efficiency gains, while a stable team can handle demand variation that the forecast did not predict.
Scenario planning turns a forecast into an operational tool. When a venue expects a larger crowd, generate additional role requirements and compare them with staff availability, overtime and service risk. If a promotion will drive more takeaway orders, model the effect on kitchen workload separately from front-of-house transactions. Store the chosen assumptions and revisit them after the event. This creates a repeatable explanation of why the roster changed, which helps managers, finance teams and employees understand the decision.
Rostering within the Australian workplace
The local employment framework makes careful rules essential. A full-time employee’s ordinary hours are generally tied to a 38-hour week, but awards and agreements may specify different patterns, including weekend work. Saturday, Sunday and public-holiday penalties must be applied when they apply, and public-holiday substitution rules can affect the actual rate. Meal and rest breaks, rest between shifts and limits on consecutive hours also vary by agreement. These requirements belong in roster constraints from the beginning, not in a review after the schedule is published.
Other Australian realities shape who can work when. Care and education rosters may need to cover fixed meal times and statutory entitlements, while casual teams can have limited availability around day jobs, study and family commitments. Temporary visa holders may also have work conditions attached to their visa, so teams should never assume every worker has the same flexibility. Store the availability that workers agree to share, keep sensitive details secure and comply with the Privacy Act. Ask for skill information voluntarily and restrict access.
Communication matters as much as the mathematics. In Australia, a roster is the schedule, a swap is a change of shift and the arvo rush may be the lunch peak; using shared language can make discussions quicker without making the process casual. Ask staff why a proposed shift may fail, especially when the model sees only transactions rather than travel, childcare or physical workload. Resources focused on restaurant labour planning can complement this exchange by connecting workforce decisions with real service conditions on the floor.
Make the forecast operational
Begin with a contained pilot rather than an enterprise rollout. Select a few sites with reliable data and enough schedule variation to learn from. Establish a simple baseline, clean timestamps and role assumptions, then run the forecast beside the existing roster for several weeks. Refresh ordinary forecasts weekly and daily when major events are likely. Review forecast error, labour cost as a share of sales, transactions per labour hour, overtime, absence coverage, customer wait times and roster changes. A pilot is working when it improves both service and the working week, not just one financial metric.
Create a monthly governance routine to review errors by site, daypart and event type. Seasonal campaigns, menu changes, store refurbishments and new opening hours can make yesterday’s pattern unreliable. Record manager overrides and the reason for each one; repeated overrides often reveal missing variables or unrealistic work standards. Keep personal data out of individual performance scoring, explain the logic behind recommendations and give staff a route to correct availability information. Predictive planning should strengthen trust between the roster, the team and the customer.
| Planning method | Main input | Advantage | Limitation |
|---|---|---|---|
| Fixed staffing | Budget and broad weekly patterns | Simple and familiar | Blurs short peaks and creates hidden queues |
| Forecast-led staffing | Timed transactions, covers and local events | Matches capacity more closely to demand | Needs clean data, clear assumptions and regular review |
| Forecast with operator review | Model recommendations checked against skills and service needs | Balances numerical evidence with local knowledge | Requires documented judgment and accountable approvals |
For operators ready to put the idea into practice, engage with the full FARE conference experience: choose sessions that match your operating segment and use the supplier exhibits to test tools rather than accept generic promises. Bring one recurring scheduling problem, compare notes with peers and take a small pilot back to your business. Register for FARE, build your network, and turn your next busy service into a roster your team can trust.