January Fills the Chapel, August Empties It: Forecasting Calls and Pricing Preneed Twelve Years Early

You tried this in 2019. Somebody built a spreadsheet that averaged the last three Januaries, told you to expect 21 calls in March, and March brought 34. You staffed to the spreadsheet, ran your two directors into the ground, paid a retired colleague day rates to cover two Saturdays, and never opened the file again. That objection is fair. Here is what is actually different now.

This trade runs on two demand curves, and both of them are lumpy

The first curve is mortality. In most of the country deaths cluster hard from December through February — influenza and pneumonia season, cold snaps, the holiday effect on frail patients — and thin out in the late summer. The swing between your heaviest month and your lightest is not a rounding error; in a 240-call firm it is the difference between 27 calls and 15, run by the same two licensed directors and the same one preparation room.

The second curve is the cemetery's, and it barely correlates with the first. North of the frost line, winter deaths do not produce winter burials: cremated remains and vaulted burials queue in the receiving vault or in storage and come out in April and May, which is also when monument setting resumes, when the Memorial Day rush hits your grounds crew, and when every family who put off a committal service calls in the same fortnight. A cemetery superintendent can be idle in January and turning away Saturday interments in May while the funeral home attached to it experienced the exact opposite year.

Demand forecasting in funeral service means estimating, from your own history, how many calls of each type you will take in each of the next thirteen weeks — burial, cremation with service, direct cremation, preneed maturities — so that staffing, retort hours, casket and vault orders and Saturday capacity are set before the week arrives rather than after it.

What your old forecast really was

Last year's number, plus a feeling. Occasionally last year's number plus a percentage somebody read in a trade journal. It failed for a specific and forgivable reason: monthly call counts in a small firm are sparse and noisy. Averaging three Januaries of 24, 31 and 19 gives you 24.7, which is a number that has never once occurred. Classical forecasting methods want long, smooth, regular history. A firm doing 240 calls a year has 12 numbers a year and every one of them is jittery.

Meanwhile the thing that quietly moved underneath you was mix. Your call count can be flat while your average revenue per call falls, because cremation without a service replaces burial with a viewing at a fraction of the receipt. That is the number that kills firms, and it is invisible in a call-count spreadsheet.

flowchart TD
  A["96 months of calls, split by type"] --> B["Model separates burial, cremation, preneed maturities"]
  B --> C["Forecast range for the next 13 weeks"]
  C --> D["Set rota, retort hours, casket and vault orders"]
  D --> E{"Did the week land inside the range?"}
  E -->|"Yes"| F["Hold the GPL and the preneed guarantee"]
  E -->|"No"| G["Log the miss and the reason: flu wave, frost, hospice census"]
  G --> B
  F --> C

Why 2026 models handle your eight years of jittery numbers

Forecasting demand is now one of the highest-adoption uses of AI in industry — around 48 percent in manufacturing — and price optimisation runs at roughly 72 percent in retail and e-commerce. Those numbers are not there because the maths got clever. They are there because current models cope with messy, sparse, seasonal history that older methods choked on, and because running them stopped being expensive: frontier AI costs fell roughly tenfold from 2025.

Practically, that means you can hand over the whole mess at once — eight years of case records exported from your case management system, your interment register, the county's death counts, the local hospice census if your liaison will share it, even the weather — and get back a range rather than a single number. A range is what a funeral director can actually staff to: 22 to 29 calls in February, with the upper end concentrated in the second and third weeks.

Claude Cowork, which launched on 12 January 2026, and ChatGPT Work, which arrived 9 July on GPT-5.6, both take a goal and return finished work — a real spreadsheet with the rota, the order quantities and the assumptions written out. That last part is the difference from 2019. You get to argue with the assumptions instead of squinting at a black box.

The four price lines that actually move your margin

Your GPL is repriced once a year, usually in a hurry, usually by adding a percentage across the board. Four lines deserve to be modelled individually. The basic services of funeral director and staff is non-declinable and carries your overhead; underprice it and every cremation case loses money. Direct cremation is the shopped line and the one an out-of-state family compares against a national provider; it needs a defensible floor built from your actual retort fuel, operator time, refrigeration days and copies. Opening and closing at the cemetery, with the Saturday and frozen-ground differentials, is where a grounds crew's overtime either gets recovered or does not. And cash advances stay exactly as the FTC Funeral Rule requires: you disclose any markup, or you pass them through at cost.

The line that deserves the most modelling attention, though, is not on your GPL at all. It is the guaranteed preneed contract.

What guessing wrong on a preneed guarantee costs over eleven years

This is the arithmetic that ought to keep an owner awake. A guaranteed preneed contract locks today's price for goods and services you will deliver a decade or more from now. Your protection is trust or insurance growth — in many states you must place a set share of the contract, often somewhere between 70 and 100 percent, into trust within about 30 days, or fund it through a carrier's policy with an increasing death benefit. If your assumed cost inflation is a point below reality, the shortfall is not felt for years and then arrives all at once, on cases you cannot reprice.

Trust or policy growth offsets part of that, and how much depends on your state's trusting percentage and your carrier's crediting rate — which is exactly why this should be modelled against your own contracts rather than a rule of thumb. One percentage point, unnoticed for a decade, on one year of sales. You write preneed every year.

The staffing side of the same problem is smaller but immediate: if February runs seven calls above what you staffed for, you are paying overtime to two licensed directors and a removal technician, buying trade embalming or a locum director at day rates, and pushing services to the following Saturday, which is the week a family remembers as the one where you seemed rushed.

Where a model must never set the price

A forecast can tell you February will be heavy. It cannot tell you that the heaviest week will include the death of a nine-year-old, and that your firm will handle that case at cost because that is what firms in your town do. Do not automate discretion. Every funeral home carries charity cases, hardship reductions and the quiet write-off for the family whose insurance lapsed, and those decisions belong to the owner in the room, not to a pricing model.

Two more limits. First, do not let a model set at-need prices dynamically. Beyond the fact that the Funeral Rule requires you to give the same General Price List to everyone who asks, charging a grieving family more because demand is high in February is the single fastest way to end a hundred-year name. Annual repricing is the right cadence, informed by the model, decided by you. Second, treat any preneed guarantee change as a legal question first: state preneed statutes govern trusting, cancellation and refunds, and your state board is unimpressed by spreadsheets.

Frequently asked questions

I only do 180 calls a year. Is that enough history?

For call counts, yes — eight years of 180 gives roughly 1,400 cases, and current models cope well with counts this small when you feed them daily or weekly dates rather than monthly totals. For rare categories, no. Do not attempt to forecast infant cases, shipouts or disinterments; there are too few and they are not seasonal.

Can this predict which specific families will die?

No, and any vendor implying otherwise should be shown the door. What it forecasts is volume by type over a period, driven by season, local population, hospice census and history. The unit of the forecast is the week, never the person.

What is the smallest useful version of this?

Export five to eight years of cases with date, disposition type, contract value and whether it was preneed. Hand the whole file over and ask for a thirteen-week forecast by type with a stated range and the assumptions listed. Then compare it week by week against reality for a quarter before you change a single staffing decision.

Should the forecast change my casket inventory?

Modestly. Most firms already run selection room merchandise on quick supplier replenishment from Batesville, Matthews Aurora or a regional distributor, so the forecast matters less for caskets than for vault scheduling, retort hours and Saturday grounds crew. Order the crew and the retort to the forecast; keep the caskets on the supplier's truck.

Start Monday by exporting the case list, nothing else. Almost every firm discovers within an hour that disposition type was recorded inconsistently for two of the eight years, and cleaning that up is worth more than any model you could buy.

One thing a forecast will show you plainly: your heavy weeks and your missed calls arrive together, because the phone rings hardest exactly when everyone is at a service. CallSphere builds AI voice and chat agents that answer funeral home and cemetery lines 24/7, give callers the price information they ask for, take first-call details and book arrangement conferences — so a heavy February shows up in your case count rather than in the firm across town's.