Banking & Airlines β where you can't stockpile inventory, every seat and teller hour is perishable, and getting demand wrong costs real money.
In manufacturing, if demand exceeds capacity today, you can build inventory yesterday. Services don't get that luxury. A bank teller hour at 2 PM on Tuesday or an empty airline seat on Flight 447 that goes unused is gone forever β that's what "perishable capacity" means.
Interview question you should nail: "In a service setting, what general operations variable is NOT available compared to production?" Answer: Finished-goods inventory. Services cannot stockpile output to smooth demand.
Build-to-stock. Produce in slow months, sell in peak months. Inventory acts as a shock absorber between supply and demand.
You can't "pre-process" 500 loan applications and put them on a shelf. Teller time, call center minutes, and loan officer capacity are consumed the moment they're delivered β or lost.
An empty seat on today's LAXβJFK flight generates $0 in revenue. Once the plane departs, that capacity is permanently gone. That's why airlines overbook.
| Dimension | Manufacturing | Banking | Airlines |
|---|---|---|---|
| "Product" | Physical goods | Transactions, loans, advice | Seat-miles (a seat flying a route) |
| Can you inventory it? | Yes β warehouses | No β service delivered in real time | No β seat is perishable |
| Primary capacity lever | Machines + workers + overtime | Staffing levels + cross-training | Aircraft + flight frequency |
| S&OP strategy emphasis | Level (absorb via inventory) | Chase (match staff to demand) | Yield management (price to fill) |
| Key forecast driver | Orders / shipments | Transaction volume by hour/day | Bookings by route & fare class |
The same four time-series methods you study for manufacturing β simple moving average, weighted moving average, exponential smoothing, and linear regression β all apply to services. The difference is what you're forecasting and why the stakes are different.
Lincoln Savings has 12 branches in Denver. They need to forecast walk-in customer transactions per hour to schedule tellers. Mondays and the 1st/15th of each month (paydays) spike 40% above average. Summers are slower; December surges for holiday lending.
The bank uses exponential smoothing (Ξ± = 0.3) for weekly totals and applies seasonal indices to adjust for known patterns (payday spikes, holiday peaks). They also run a 3-week moving average as a crosscheck.
Why it matters: Too few tellers β long wait times β customers leave for a competitor. Too many tellers β labor cost waste. There's no "teller inventory" to draw on β it's pure capacity matching.
Frontier needs to forecast bookings per fare class on its DenverβMiami route for the next 6 months. Historical data shows strong seasonality: winter demand spikes (snowbirds), summer dips. They also see a clear upward trend as Denver's population grows.
Because there's both trend and seasonality, simple exponential smoothing won't cut it. Frontier uses trend-adjusted (Holt's) exponential smoothing and applies seasonal decomposition. For new routes with no history, they fall back on qualitative methods β expert judgment, analogies from similar routes (historical analogy), and the Delphi method.
Why it matters: Under-forecast β turn away passengers, lose revenue. Over-forecast β fly with empty seats (revenue = $0 on those seats).
This is a favorite quiz question β and a real interview question. The answer always depends on the demand pattern:
Simple Moving Average or basic Exponential Smoothing. Bank ATM transactions in a stable neighborhood. No trend, no season β just smooth the noise.
Holt's (Trend-Adjusted) Smoothing or Linear Regression. Airline route growing 5% per year. SMA would always lag behind the trend.
Decomposition β deseasonalize the data, fit a trend line, then reseasonalize. Holiday lending at a bank. Snowbird routes for an airline.
Qualitative methods: Delphi, market research, historical analogy. No historical data exists, so time-series won't work.
Measuring forecast quality: MAD = average absolute error. MAPE = average absolute error as a % of demand (better for comparing across different-sized items). Tracking Signal = RSFE Γ· MAD β if it drifts beyond Β±3β4 MADs, your forecast is biased and needs fixing.
In manufacturing, the three pure S&OP strategies are Chase (vary workforce), Level (constant workforce, absorb via inventory), and Stable WorkforceβVariable Hours (use overtime/slack). In services, there's a critical twist:
The Level strategy has a problem in services: You can't build up inventory during slow periods to sell in busy ones. A bank can't "pre-serve" customers. So services naturally lean toward chase (adjust staffing) or demand management (shift customer timing through pricing, appointments, etc.).
A regional bank forecasts that Q4 loan demand will spike 30% above Q3 due to holiday personal loans and year-end business lending. Their S&OP response:
Cost trade-off: Hiring + training cost vs. the revenue lost from turning away loan applicants.
An airline can't easily "hire more planes" for the holidays β aircraft are fixed capacity with long lead times. Instead, airlines combine:
This is why the textbook says yield management is the dominant S&OP tool for airlines β the pure strategies don't fully apply when capacity is truly fixed and inventory is perishable.
| Strategy | How It Works | Bank Example | Airline Example |
|---|---|---|---|
| Chase | Vary workforce to match demand | Hire temp tellers for Q4 | Add seasonal routes / frequencies |
| Level | Keep workforce constant; absorb via inventory | β Can't pre-process transactions | β Can't stockpile empty seats |
| Stable + Variable Hours | Keep headcount, flex overtime | Extend branch hours on paydays | Crew overtime for schedule surges |
| Yield Management | Price dynamically to fill capacity | Lower loan rates in slow months | SABRE-style fare optimization |
Yield management is the process of allocating the right type of capacity to the right customer at the right price and time to maximize revenue. The textbook traces its modern origin to American Airlines' SABRE system.
Four conditions (memorize these β they come up on exams and in interviews):
Adding a hotel room or aircraft seat is enormously expensive. You're stuck with what you have.
The marginal cost of one more passenger in an existing seat is almost nothing (a bag of pretzels). So any revenue > $0 is better than an empty seat.
An unsold seat on today's flight, an empty hotel room tonight β gone forever. Unlike a widget on a shelf.
Reservations let you segment customers by booking time and willingness to pay (business vs. leisure travelers).
The textbook lists four operational requirements β another great exam question:
Pricing must appear logical to customers. Saturday-night-stay requirements, advance purchase discounts, and fare classes all create "fences" that let you charge different prices for the same seat.
Must manage variability in arrival times, service duration, and gaps between customers. A hotel guest might check out early; a flight might have 12 no-shows.
The system must handle the actual service delivery β routing, scheduling, and real-time adjustments.
Staff must work in an environment where overbooking and price changes are standard occurrences that directly impact the customer.
Interview-ready answer: "Why do airlines overbook?" β Because seats are perishable, variable cost per passenger is near zero, and historically 5-15% of passengers no-show. Overbooking captures revenue that would otherwise be lost forever. The critical ratio (Cu / (Cu + Co)) determines optimal overbooking level β Cu is the lost revenue from an empty seat; Co is the cost of bumping a passenger.
Work through these to build instant recall for your quiz and job interviews. Your score is tracked at the bottom right.
Compared to a manufacturing setting, which general operations variable is not available in a pure service setting like banking or airlines?
Frontier Airlines sees that bookings on DenverβMiami have been growing 8% per year with strong winter peaks. They want to forecast next winter's demand. A simple moving average keeps under-forecasting. Which technique should they switch to?
First National Bank forecasts Q1 demand at 10,000 transactions/week and Q3 demand at 6,500 transactions/week. Each teller handles 250 transactions/week. If they use a level strategy, how many tellers should they staff year-round?
A 180-seat flight has a no-show rate of about 10%. The ticket price is $300 and the cost of bumping an overbooked passenger (hotel + rebooking + voucher) is $800. Using the critical ratio concept, which statement is most accurate?
A bank's loan demand forecast has accumulated these errors over 6 months: RSFE = +450, Sum of Absolute Deviations = 600, giving MAD = 100. What does the tracking signal tell you?
For each business, decide whether yield management is a strong fit or a poor fit. Think about the four conditions: fixed capacity, low variable cost, perishable inventory, and advance selling.
Drag each real-world service action into the correct S&OP strategy bucket.
Plug in your own numbers to see how the critical ratio works. This is the newsvendor logic applied to perishable service capacity.
The textbook says yield management pricing must appear "logical to the customer" and uses rate fences. Which of the following is the BEST example of a rate fence?
A brand-new bank branch opens in a suburb with no transaction history. The operations manager needs a demand forecast for staffing. Which forecasting approach makes the most sense?
The textbook credits the widespread scientific application of yield management to which company's computerized reservation system?