How to forecast actual event attendance from ticket bookings

You've sold the tickets. Now the caterer wants a final number, the venue wants your arrival plan, and someone on the team says, "We usually lose a few people on the day."

That's a reasonable observation. It isn't enough to order lunch against.

An event attendance forecast estimates how many people will actually arrive, using valid bookings and attendance at comparable events. Start with the tickets that still admit someone, apply a defensible attendance rate to each meaningful group, then add expected arrivals from future sales and walk-ups separately. Show a range, because a single neat number hides the assumptions your suppliers need to understand.

The useful part is what you do with that range. Some orders can follow likely demand. Other commitments have to work for everyone entitled to attend. A forecast can help you reduce waste without leaving a paid guest without their meal, selling somebody else's place, or treating a quiet entrance as permission to cut the safety plan.

Let's work through a fictional paid conference and turn the booking count into an estimate you can explain, update and use before the spending deadlines pass.

Decide which attendance number you need

Start by finishing this sentence: "We need to estimate how many people will..."

Arrive at any point during the event? Be inside at lunchtime? Use the shuttle before the opening session? Attend the second day? Those are different questions, and one attendance total can't answer them all.

For a single-day conference, the first forecast might count unique audience members who enter at least once. Each person counts once, even if they go outside and return. That gives you a useful event total. It doesn't tell you the busiest moment at reception or how many people need a particular service.

Keep staff, speakers, exhibitors and contractors visible too. They may need meals, transport and space even though they aren't part of the ticket-holder forecast. Put them in separate rows so the audience number doesn't quietly become the headcount for every supplier.

Then distinguish actual attendance from valid admissions. A valid ticket represents somebody's right to enter under the event's terms. Your estimate that they might stay home doesn't remove that right. When you plan the venue and admission inventory, retain the ability to honour those tickets within the agreed limits.

Great Britain's Health and Safety Executive includes previous attendance, advance sales, similar events and changing conditions among the inputs for forecasting expected turnout. It also treats safe capacity as a separate part of assessing the venue. Check the requirements that apply where your event takes place; an attendance calculation doesn't establish a safe capacity.

For the rest of this guide, our main forecast means unique audience arrivals over one conference day. We'll add staff meals separately and examine arrival timing later. Writing that definition down prevents several people from using the same figure to mean different things.

Count admissions that remain valid

The first thing you want to do is take a dated booking snapshot. Record the event date, the report time and the cutoff used across every sales channel. If one channel is missing yesterday's transactions, resolve that before interpreting the total.

Count people admitted by the bookings, rather than orders. One employer buying ten individual admissions contributes ten places. A group product that admits several people needs to be converted into the number actually entitled to enter. Check the product's terms before assuming that one ticket always means one person.

Remove admissions that have genuinely been cancelled. A refund that cancels a ticket reduces the valid count. A partial refund for a changed service may leave admission intact, so a refunded amount alone isn't enough to tell you what happened.

Transfers and resale need the same attention. If one admission changes hands, you still have one person entitled to use it. Counting the original sale and its replacement as two audience members overstates both expected attendance and the places you've committed.

Separate confirmed admissions from reservations that haven't become valid tickets. A sponsor saying it might bring a team isn't the same as an issued guest allocation. An employer holding places while it seeks approval isn't necessarily a paid booking. Write the status clearly and treat unresolved allocations as an assumption, with an owner who can confirm them.

Our guide to reconciling tickets and attendance covers the differences between sales, admission records and check-ins. Here, you need a clean starting count before estimating how many of those people will arrive.

Preserve the snapshot you used. Next week's live dashboard will show a different position, and you'll want to explain whether a forecast changed because bookings moved or because you changed an assumption. You can't reconstruct that reliably from a final total alone.

Choose history that resembles this event

Your own completed events are the best starting place when they reflect the purchase and journey your current audience is making. Start with the same series, then consider the audience, location, event length, price and booking conditions.

A local evening talk and a destination conference can attract equally committed buyers while requiring very different travel arrangements. Moving a recurring event to a weekday can change who can attend. Adding an online option may change physical turnout even if the headline booking count looks familiar.

Check whether the previous editions offered the same refund and transfer options. Also check when the attendance rate was measured. If your current supplier decision happens a week before doors, a historical rate based only on tickets still valid at opening time leaves out cancellations that occurred during that final week.

You have two workable choices. Use historical snapshots from the same lead time and measure the eventual attendance of those admissions, including any that later cancelled. Or forecast future cancellations separately, then apply an opening-time attendance rate to the admissions expected to remain valid. Explain which method you're using and avoid deducting the same cancellations twice.

For our worked example, we'll use the first method. Rates describe the eventual arrivals from admissions valid seven days before each event. Later cancellations are already reflected in those rates. We won't subtract another cancellation allowance from the same group.

Read the records before selecting the editions. Keep genuinely comparable disappointing events as well as the strong ones. If a severe disruption makes one edition unrepresentative of ordinary conditions, retain it as a disruption scenario and explain why it isn't driving the main estimate.

Don't silently discard it because it produces an inconvenient number. Someone reviewing the forecast should be able to see which editions you chose, which you excluded and why. A short explanation beside each event is more useful than a large average assembled from everything your business has ever hosted.

Calculate attendance rates from matching counts

An attendance rate is the number who arrived from a defined booking group, divided by that group's valid admissions at the chosen snapshot. Match each counted arrival to an admission in that snapshot group, following a valid transfer or replacement once rather than counting both versions.

Suppose an edition had 200 public admissions valid seven days before doors. Of those admissions, 180 eventually produced an arrival. The rate for that snapshot group was 180 divided by 200, or 90%. Somebody buying later doesn't belong in that numerator, even if they attended the same event.

Here's a fictional history for Northbank Forum, our one-day paid conference. These are invented teaching figures, not industry benchmarks or Loopyah customer results. Three comparable editions recorded public-ticket arrivals of 180 from 200 admissions, 162 from 180, and 108 from 120. All use the seven-day snapshot definition.

Across those editions, 450 of 500 public admissions produced arrivals. The pooled attendance rate is therefore 90%. Pooling means adding the arrivals and admissions first, then dividing. It gives larger groups more influence because they contain more admissions.

You can also inspect each edition separately. In this example, each happens to have the same rate. Real histories will vary, and that variation is part of the evidence you need. Don't conceal an unusually weak edition inside a reassuring pooled figure.

Other Northbank records show 160 arrivals from 200 partner admissions and 30 from 60 complimentary admissions, again using comparable seven-day snapshots. That gives provisional rates of 80% and 50%. The smaller complimentary history deserves more caution, particularly if its guest allocation process has changed.

These rates describe these fictional groups. Paying for a ticket doesn't create a universal attendance percentage, and a complimentary guest isn't automatically less interested. Use a group split when your records show a relevant difference or when its booking arrangements require separate handling.

I'd start with a few groups you can explain. Dividing a small history by ticket type, purchase week, distance and acquisition channel can leave each calculation dependent on a handful of people. More detail can make the forecast look sophisticated while making its assumptions harder to defend.

Build the estimate without counting anyone twice

Seven days before Northbank Forum, there are 300 public admissions, 100 partner admissions and 40 complimentary admissions. That's 440 valid audience admissions in total. The agreed event plan allows at most 500 audience admissions, with staff and other working roles accounted for separately in the venue plan.

For the admissions already valid, the central estimate is straightforward:

• Public admissions: 300 multiplied by 90% gives 270 expected arrivals.

• Partner admissions: 100 multiplied by 80% gives 80 expected arrivals.

• Complimentary admissions: 40 multiplied by 50% gives 20 expected arrivals.

• Existing admissions together: 270 plus 80 plus 20 gives 370 expected arrivals.

That's the forecast for the current booking groups. Sales remain open, so it isn't the final event forecast yet.

The organiser separately expects another 20 advance public admissions and 12 walk-up admissions. For this illustration, the future advance buyers are assumed to attend at 90%, giving another 18 arrivals. The walk-up estimate counts people who purchase and enter at the door, giving another 12 arrivals directly. Don't apply the attendance rate again to a figure already defined as arrivals.

The central event forecast is therefore 370 plus 18 plus 12, or 400 unique audience arrivals. Before allowing for any later cancellations, the planned audience admissions total 472: the existing 440, another 20 in advance and 12 at the door. That remains within this fictional event's audience limit.

The 20 future advance sales are an assumption, not part of today's booking record. Our ticket sales curve guide explains how to estimate the sales still to come. Keep that forecast connected to the attendance work without treating predicted sales as money already collected or guests already confirmed.

Use a separate attendance assumption for late buyers if your evidence supports one. We've held it at 90% to keep the example readable, not because people booking tomorrow must behave like people who booked weeks ago. Record that limitation beside the calculation.

Show a range with reasons behind it

The central estimate gives the team somewhere to start. Now ask which assumptions could change enough to affect the decisions you still have to make.

For Northbank's lower-attendance scenario, suppose public attendance falls to 85%, partner attendance to 60% and complimentary attendance to 25%. Assume no further advance sales or walk-ups. Those assumptions produce 255 plus 60 plus 10, or 325 arrivals.

For the higher-attendance scenario, use 95% for existing public admissions, 90% for partners and 75% for complimentary guests. Assume another 40 advance public admissions at 95% attendance and 20 walk-ups. That gives 285 plus 90 plus 30 plus 38 plus 20, or 463 arrivals.

The higher scenario uses all 500 audience admissions: 440 already valid, 40 future advance admissions and 20 at the door. Further sales couldn't simply be added. The inventory plan would have to keep respecting the event's limits and every admission already issued.

So the fictional scenarios are 325, 400 and 463 arrivals. They are planning cases chosen to expose different outcomes. They aren't a statistical promise that attendance must finish between 325 and 463. All 500 eligible audience members might arrive, and a major disruption could produce fewer than the lower case.

Hyndman and Athanasopoulos explain in Forecasting: Principles and Practice that a prediction interval has a specified probability under a forecasting model. Choosing three plausible scenarios doesn't establish that probability. Unless you've built and checked such a model, call these scenarios and state their assumptions.

Challenge the scenarios as stories about the same event. A transport disruption might reduce arrivals and concentrate those who do attend into a shorter period. Stronger late sales might accompany a programme announcement that also brings more existing buyers. Avoid combining convenient assumptions that couldn't reasonably happen together.

Keep the case where everyone entitled to attend arrives visible even when it seems unlikely. That's the case your admission commitments and promised services may still need to accommodate. The central forecast helps you judge likely demand; the right preparation also depends on the consequence of getting that demand wrong.

Treat a large group as a shared decision

A single employer can account for a meaningful share of your audience. Its staff might travel together, depend on the same manager's approval or share a competing deadline. Their decisions aren't necessarily independent.

That matters even when the average attendance rate is sound. Ten unrelated people changing plans may have a modest effect on the total. One coach cancellation can remove a whole group at once. Multiplying every admission by a percentage gives an expected count, but doesn't explain how concentrated the uncertainty is.

For Northbank, the 100 partner admissions might include one employer with 40 places. Ask the partner to confirm the allocation, travel arrangements and contact responsible for changes. Keep the 40 valid admissions in the count while recording what remains uncertain.

Then test the operational effect if that group arrives together or can't attend. Don't subtract it from the lower scenario again if the lower partner rate was already chosen to represent that same possibility. Write the alternative clearly so somebody reading the total knows whether it includes the group.

A named attendee list can help with a specific delivery need, such as arranging badges or checking meal choices. It doesn't make attendance certain. Equally, an unnamed but valid group allocation can still produce people at the door under the terms you've agreed.

Chase the information that changes a decision. If an employer can confirm its coach arrival time, that may be more useful to reception than another general promise that everyone is excited. If a partner can release genuinely unused places through the agreed process, update admission inventory only after that release is effective.

You don't need to turn this into a personal reliability score for every attendee. Group-level planning can identify a shared transport problem without labelling individuals as likely absentees. Keep the evidence relevant to the event and the action you're deciding.

Forecast when people arrive as well as how many

Imagine Northbank's 400 central-case arrivals coming through the door. If most arrive during a short period before the keynote, reception faces a different job than it would with the same total spread across the morning.

Build an arrival profile from comparable events where entry times are reliable. Count first arrivals in useful time periods and record what was happening then: doors opening, breakfast service, a coach arriving or the main programme beginning. Keep repeat scans and re-entry separate.

For an invented timing illustration, suppose 160 of Northbank's 400 arrivals reach reception between 8:30 and 9:00. That's 40% of the day's audience in half an hour, averaging about 5.3 arrivals per minute. Actual arrivals can bunch within that period, so the average isn't the peak rate reception must handle.

The remaining 240 people arrive at other times. They aren't all inside by 9:00, and the 160 figure doesn't tell you how many are still waiting outside. Queue length depends on how quickly people arrive and how quickly staff can serve them, including searches and ticket problems.

HSE's guidance on crowd controls around entry and departure recognises that traffic can affect arrival timing and that queue design, information and stewarding affect crowd flow. It's Great Britain guidance; use the relevant local advice and the venue team's assessment for your event.

Our check-in staffing guide takes the next step, connecting peak demand with measured service times and a workable team. Send that team the timing assumptions, expected group arrivals and unresolved travel issues alongside the event total.

If the venue needs occupancy, count the people present at the same time. Day-long unique arrivals don't measure this. You need reliable entries and exits, the people already present, and an agreed way to handle missing records. Ask the responsible venue team how occupancy is monitored rather than subtracting an assumed departure percentage. A person who leaves and returns still uses space when they're back inside.

For a multi-day event, repeat the exercise by day. A person with a weekend pass counts once in the weekend's unique audience total and potentially once on each day they attend. Adding daily attendance gives person-days of attendance, not the number of distinct people you reached. Meal and transport orders may need the daily counts, while your audience report needs the distinction preserved.

Choose supplier quantities around the actual promise

Here's where the forecast becomes commercially useful. You can estimate 400 arrivals and still choose different quantities for different services, because the cost of a shortage and the options for correcting it differ.

Start with the promise made to attendees. If every valid ticket includes lunch, everyone who attends with that ticket must receive the promised lunch. An average attendance rate doesn't let you quietly replace that promise with "available for the first people through the door."

Ask the supplier what can change, until when, and at what price. A quantity that can be increased reliably on the morning is different from one that becomes impossible to alter days before the event. Verify the whole delivery arrangement, including dietary meals, staffing, storage and service timing.

For Northbank, assume all audience admissions include lunch and 24 working staff need lunch too. Staff meals are separate from the audience forecast, so the central meal estimate is 424. If all 500 audience places are eventually issued and used, the requirement becomes 524 meals.

Consider two fictional catering offers. Every price below is an invented USD charge complete for the stated catering scope, including applicable tax, with no assumed tax recovery. These are examples of contract choices, not market prices or a recommendation to use these quantities.

The first offer requires ordering all 524 meals in advance at USD 24 each. That costs USD 12,576. The price buys certainty of supply under the agreed service arrangements, even if fewer meals are eaten.

The second offer allows an initial 424 meals at USD 24, a fixed USD 200 availability fee, and up to 100 additional meals at USD 28 each. The supplier must explicitly guarantee that the additional meals can reach service in time, including the required dietary provision. Without that confirmed ability, the offer doesn't solve the shortage problem.

The initial meals cost USD 10,176. Adding the availability fee makes the starting commitment USD 10,376. If none of the additional meals are needed, that's USD 2,200 less than ordering 524 upfront. If all 100 are needed, the bill reaches USD 13,176, which is USD 600 more than the first offer.

Neither offer is automatically better. The flexible offer lets you spend less when attendance is lower, while paying more if everyone comes. You need enough budget and available cash for its maximum bill. You also need a named person who can request the extra meals before the supplier's deadline.

If additional meals can't be delivered during service, agree a sufficient quantity before the final order deadline. Don't use the central forecast as an excuse to accept a shortage you have no way to fix. Confirm access arrangements and specific dietary commitments directly; they shouldn't disappear inside an average attendance allowance.

Apply the same reasoning to other suppliers. Reusable stock may be easier to hold in reserve than perishable food. Security and medical provision depend on the responsible assessment and agreed plan, rather than a proportional cut whenever expected attendance falls. Keep safety requirements and contractual service promises explicit when deciding what is genuinely flexible.

Keep attendance, revenue and cash separate

A lower attendance forecast doesn't necessarily mean you lose the face value of every unused ticket. If a buyer paid and the money remains yours under the applicable terms, their absence hasn't reversed that sale. Check the actual refunds, replacement admissions and payment deductions before changing the revenue forecast.

What may change is spending at the event. You might expect fewer food purchases or fewer paid workshop upgrades. Those are separate assumptions about what attendees buy, and they need their own costs. Revenue from an extra sale isn't the same as the amount left after providing it.

For a small fictional example, suppose ten additional attendees each buy USD 15 of refreshments that cost USD 9 per person to supply, including all variable charges for those sales. The extra revenue is USD 150 and the extra variable cost is USD 90. That leaves USD 60 before any other event costs. These invented figures don't establish that the ten people will arrive or make those purchases.

If bringing them requires an additional USD 100 transport service, refreshments alone wouldn't cover it under those assumptions. You might still provide transport because it's a confirmed commitment or solves an access problem. State that reason and fund it explicitly, rather than counting the original ticket revenue a second time to make the decision look profitable.

Attendance also doesn't release you from a supplier minimum. If the catering agreement bills a minimum quantity, a lower forecast creates no saving below that quantity. Likewise, stock already bought and wages already committed don't disappear because the expected audience falls. Separate costs you can still change from those already fixed.

For Northbank's flexible catering offer, the USD 2,200 difference at the starting quantity is conditional. Additional meals reduce it, and the maximum bill exceeds the upfront offer. Keep the maximum available in the spending plan until the uncertainty resolves. Saving against one possible outcome doesn't provide cash to pay the bill in another.

Use the same care with sponsor reporting. A booking forecast can help a sponsor prepare stock or staff a stand, provided you label it as an estimate and explain the count. After the event, report the attendance you can substantiate. Don't turn expected arrivals into verified attendance or imply that everyone entering the venue visited the sponsor.

This separation makes the forecast easier to use in a budget meeting. The attendance estimate informs specific purchases and sales assumptions. The revenue forecast records what you expect to earn under the relevant terms. The cash plan records when money can actually pay the bills. Connect them through the transactions that change, keeping each number's meaning clear.

Use current signals without inventing certainty

You might have more information than bookings alone. Attendees have replied to a travel question, selected a session or downloaded joining information. Those actions can help you understand what people are planning, but you need evidence before converting them into attendance probabilities.

Somebody who doesn't open an email can still attend. Somebody who confirms enthusiastically can still become ill. A confirmation request is useful for resolving arrangements, and its value grows when you can compare responses with completed attendance across similar events.

Until then, describe what the signal actually says. "This employer confirmed its coach" is more informative than raising every partner ticket's attendance assumption because a contact sounded confident. Record the new information, the group affected and the reason it changes your estimate.

Keep planned improvements separate from delivered ones. A reminder campaign you haven't sent cannot yet explain a higher forecast. Even after sending it, a stronger turnout doesn't by itself prove that the campaign caused the difference. Our guide to reducing no-shows covers practical interventions; evaluate their results without assuming the uplift in advance.

Use only the personal information needed for the planning task. The UK's Information Commissioner's Office describes data minimisation as keeping personal data sufficient, relevant and limited to the purpose. That guidance is under review, so check the current rules that apply to your organisation.

For an attendance estimate, aggregate ticket-group counts may be enough. The caterer needs meal quantities and appropriate dietary information through the agreed process, not a list of people you've guessed are unlikely to attend. Separate that operational information from speculative notes and give access only to people who need it.

When you don't have a comparable attendance history

A first edition still needs a plan. You just have less evidence for the attendance assumptions, and the forecast should say so.

Start with valid admissions and the delivery commitments they create. Look for comparable formats and journeys, and ask experienced operators what changed their turnout. Treat any borrowed rate as provisional unless you can understand how it was measured and why the audience resembles yours.

A published percentage without a clear population, event type, booking definition and measurement date won't resolve your uncertainty. Free registration at a broad online event doesn't establish the attendance rate for a paid local course. A precise-looking number can be a poor fit for the decision.

In their chapter on judgmental forecasting, Hyndman and Athanasopoulos recommend clear tasks, documented assumptions and systematic evaluation. Apply that approach by explaining what you know, what you're assuming and what would make you revise it.

Keep the scenarios broad enough to expose a difficult outcome, and negotiate options before you need them. If catering can only be changed before a deadline, make that deadline part of the forecast review. If you can't obtain a credible attendance assumption at all, prepare against the valid commitments and explain the cost of that uncertainty.

Start collecting useful history now. Save booking snapshots at the decisions you actually make, then record attendance consistently after the event. Your second edition can have better evidence without requiring a complicated prediction system or an individual score for every buyer.

Update before the next decision becomes expensive

Choose a review rhythm around the event's deadlines. A weekly update may be enough early on. In the final period, a supplier cutoff, a large group change or a transport issue can justify another review sooner.

At each review, save a new booking snapshot and reconcile it to the previous version. Explain new sales, effective cancellations, transferred admissions and confirmed allocations. Then revisit the assumptions that could change the pending decision.

Use rates matched to the new lead time where your history allows it. A rate measured seven days before doors includes a full week in which plans can change; applying it unchanged the night before may understate arrivals. If matching records aren't available, explain the limitation and review the remaining cancellations and travel uncertainty directly.

Keep the estimate of future sales separate. Yesterday's forecast of another 20 advance buyers shouldn't remain untouched after those 20 people buy. They now belong in the current booking count, and the estimate of sales still to come needs reviewing. Otherwise the forecast can count them twice.

Changed conditions may affect both attendance and timing. A train disruption can leave some ticket holders unable to travel while causing others to arrive later together. A low morning check-in count therefore needs investigation before you reduce lunch or release support staff.

Use the ticketing records you already have. Loopyah's ticketing and check-in tools let door staff scan QR tickets on their phones with check-in access. That supports recording admission at the door. You still need to define the attendance count, reconcile exceptions and choose the forecast assumptions described here.

Don't assume a scan total is complete while another entrance is using a manual fallback or a device hasn't finished reporting. Confirm the actual state with the door team. Record the agreed correction once those records are available, preserving enough information to explain the change.

Send a short update to everyone acting on the forecast. State what changed, which quantity is now agreed, who can authorise an increase and the final time for that decision. A revised estimate helps only when the people placing orders and running service actually receive it.

A forecast brief you can reuse

Your working calculations can be detailed. The people making decisions need a short brief that makes the estimate and its limits easy to understand. For Northbank, that brief could read as follows.

• Event and count: Northbank Forum, one day, unique audience members entering at least once. Staff and working roles are counted separately.

• Snapshot: seven days before doors, with a recorded report time and all booking channels checked.

• Current admissions: 300 public, 100 partner and 40 complimentary, totalling 440 valid audience admissions.

• Evidence: comparable seven-day snapshot groups, using pooled attendance rates of 90%, 80% and 50%. The complimentary history is small.

• Central forecast: 370 arrivals from current admissions, 18 from 20 assumed further advance admissions and 12 walk-ups, totalling 400 audience arrivals.

• Scenarios: 325 lower, 400 central and 463 higher, with the separate case of all 500 possible audience ticket holders arriving retained for delivery planning.

• Admission limit: at most 500 audience admissions under this event's agreed plan. Working roles remain separately accounted for.

• Catering decision: 424 initial meals including 24 staff meals, only if the guaranteed option for another 100 meals can meet service timing and dietary requirements. Maximum catering bill USD 13,176.

• Next action: name the person confirming supplier terms, the person monitoring arrivals and the last time additional meals can be requested. Record the next review before that deadline.

Replace these fictional details with your own. Include unresolved information beside the decision it affects, rather than writing a general warning at the bottom. If the partner coach remains unconfirmed, reception and catering should know which group and time period are affected.

Keep the latest agreed brief easy to identify and retain earlier versions for review. Suppliers shouldn't have to choose between several numbers circulating in messages. Someone covering for the organiser should be able to find the current quantity and authority without rebuilding the forecast.

Review the forecast against what actually happened

Once the event is complete, reconcile attendance before judging the forecast. Combine the agreed door records, resolve duplicate entries and distinguish first admissions from re-entry. Record any remaining gaps instead of treating an incomplete count as the final truth.

Then compare actual attendance with the forecast saved before the relevant decision. If Northbank forecast 400 audience arrivals and ultimately recorded 418, it underestimated by 18. The catering review also needs the actual staff meal count and quantities served; audience arrivals alone won't explain the whole bill.

Look at the groups as well as the total. Public attendance could be above expectation while partner attendance is below it, leaving the total close by coincidence. Repeating the same assumptions would preserve the underlying mistakes even though the headline forecast looked good.

Check the range honestly. An outcome within your scenarios doesn't prove they had a particular probability of success. An outcome outside them asks you to investigate the assumptions, the records and any new event. It doesn't automatically mean that a more complicated method would have predicted it.

Test future changes against information you could actually have known at the time. The same forecasting textbook stresses evaluating forecasts using new observations, rather than judging a method only on the data used to build it. For an event series, try predicting a later edition using only earlier editions and the booking snapshot then available.

Keep errors separate by lead time. A forecast made the night before doors has more information than the one used for a catering commitment a week earlier. Compare like with like before concluding that the method has improved.

Finally, record the practical result. Did the extra-meal option work? Did the door team receive the coach update? Did a decision arrive too late to change a purchase? Use those answers to improve the next supplier agreement and review schedule. Before the next supplier cutoff, check your current admissions against that record and confirm which quantities the team can still change.