Modeling lifetime value when contracts, churn and tariffs all move independently

Three households sign on the same Tuesday: one takes a two-year fixed offer, one sits on a rolling variable rate, one is on a prepayment meter. Run all three through a single blended churn ratechurn rateChurn rate is the percentage of customers or revenue lost over a period. It measures how fast a business loses its existing customer base.View full definition → and a single margin assumption and you will be wrong about each of them, in different directions and by different multiples. The fixed customer generates almost no risk for 23 months and then all of it at once. The variable customer decays smoothly. The prepayment customer can stop producing revenue without ever showing up in a churn report.

That is the modeling problem this lesson solves. Contract length, tariff type, payment method and usage move independently, and the "average revenue times average tenure" shortcut quietly assumes they move together.

Why energy LTVLTVLifetime Value: the total revenue (or profit) a customer generates throughout their entire relationship with your business.View full definition → is harder than subscription LTV

In SaaS (Software as a Service), lifetime value usually hangs on one dominant lever: the churn rate. Energy value per customer is a moving target, and four things move it:

• Contract length: fixed terms and benefit periods (typically 12 to 24 months in the UK, Europe and Australia) versus rolling variable arrangements.
• Tariff type: price locked for the term versus a standing or variable rate that tracks wholesale, with a regulated ceiling on the standing cohort in many markets (Ofgem's cap in Great Britain, the AER's Default Market Offer in Australia).
• Payment method: direct debit, pay on receipt, prepayment. This changes bad debt, cost to serve and the shape of the survival curve.
• Usage band: annual kWh (kilowatt-hour), and whether the meter also exports.

The structure you want is:

LTV = Σ (expected margin in period t × probability of survival through t) − cost to serve

Retention probability is not a constant you can borrow from a company average. Take the renewal-point and save-desk figures from the retention benchmarks lesson as inputs; the work here is the structure they plug into.

Building the model: levers that move independently

1. Contract length sets the churn clock

Fixed-term customers have near-zero in-term churn (exit fees see to that) and a sharp hazard spike at expiry. Rolling customers have a roughly constant monthly hazard driven by comparison sites and outboundoutboundProactive outreach that pushes your message to targeted audiences through advertising, email, or direct prospecting, initiated by the seller rather than the buyer.View full definition → campaigns. Two different survival curves, not one rate applied twice.

There is a third hazard sitting underneath both: move-out. When a tenant leaves, the account closes regardless of how good your offer was. In rental-heavy postcodes this can be a large share of gross churn, and it is invisible if you only count "lost to competitor". Fold it into competitive churn and you will fund save offers against a population that cannot be saved.

2. Tariff type changes both margin and stickiness

Fixed tariffs lock in a margin and hand the wholesale risk to the retailer, which is how a wave of European suppliers failed through 2021 and 2022. Variable tariffs pass that risk back to the customer, which protects margin but correlates with more shopping around. A workable planning assumption, not a constant: variable-tariff churn runs somewhere around 1.3 to 1.8 times fixed-tariff churn in competitive markets.

Amber Electric in Australia inverts the whole structure. Members pay a flat monthly subscription (roughly A$20) and get wholesale spot prices passed straight through. Margin per customer is then independent of usage, of wholesale price and of tariff mix, which makes the revenue side of the model almost trivial. The churn side gets harder: exposure to spot prices means attrition correlates with market events rather than with the calendar. When AEMO suspended the National Electricity Market spot market in June 2022, the cohorts with a reason to leave that month were the ones watching prices in an app. Model that cohort with a hazard conditioned on wholesale volatility, and hold a reserve for the month it spikes.

3. Usage band, and the solar exception

High usage lifts absolute margin at the same percentage, and high-usage households bundle more, which is itself a retention effect. The exception matters in Australia, where roughly a third of houses have rooftop solar. An exporting household can run small or negative net consumption through spring and autumn while still absorbing full cost to serve and a feed-in payment. Priced on kWh alone, that cohort looks like the worst on the book.

AGL Energy and Origin Energy, each with several million customer accounts, sell solar, batteries and virtual power plant participation into exactly that base, which turns the margin line into a mix of retail spread, equipment and flexibility value. If your LTV model only knows consumption, it will tell you to stop acquiring the customers with the longest product ladder.

4. Prepayment: the cohort that leaves without switching

Several million British households pay in advance, and the economics differ on every axis. No bad debt, because the money arrives first. Higher cost to serve, because of top-up infrastructure and support volume. And a failure mode with no equivalent elsewhere: self-disconnection. A household that stops topping up remains an active account, keeps counting in the customer base reported to the board, and contributes nothing. Build the prepayment survival curve on top-up frequency rather than account status, or the model will carry dead weight for years. The "they cannot leave anyway" assumption is also only half true, since debt assignment rules let a prepayment customer switch while carrying debt up to a capped amount.

Worked example: three cohorts from the same book

Illustrative figures, for teaching, not published data.

A: dual-fuel, 24-month fixed, medium usage. A$180 annual gross margingross marginGross margin is the share of revenue left after subtracting the direct cost of producing goods or services, expressed as a percentage of revenue.View full definition → per fuel, so A$360 combined. Near-zero in-term churn, then a renewal cliff; expected tenure around 2.5 years.

B: single-fuel electricity, rolling variable, medium usage. A$150 annual margin (thinner, because variable rates are priced to survive a comparison table), 3% monthly churn, so roughly 30% a year.

C: subscription with wholesale pass-through, Amber-style. A$240 annual subscription margin, no wholesale spread either way, assume 25% annual attrition.

LTV_A ≈ 360 × 2.5 years              = A$900
LTV_B ≈ 150 × (1 / 0.30)             = A$500
LTV_C ≈ 240 × (1 / 0.25)             = A$960

Two things to notice. Cohort A beats B by 1.8x on retention structure alone, not on price or usage. And cohort C, the one with the smallest bill exposure, models highest, because flat margin plus moderate churn beats thin margin plus fast churn. The variance is the catch: C's attrition arrives in lumps tied to price events, so the mean is right and the quarter can still be badly wrong.

Undiscounted five-year streams flatter everything. At around 8%, the annuity factor is closer to 4.0 than 5.0, so trim roughly a fifth before comparing to the acquisition costacquisition costCustomer Acquisition Cost (CAC) is the total sales and marketing spend divided by the number of new customers gained in a period. It measures how efficiently you grow.View full definition → the CAC lesson calculates for your channel mix. The 3:1 LTV to acquisition-cost heuristic that circulates from Bessemer's memo on SaaS metrics is a useful guardrailguardrailRules and controls that keep an AI system inside safe, legal and on-brand boundaries, blocking outputs and actions that cross the line.View full definition →, though it was never built for a regulated commodity.

Putting it into a simple cohort model

Do not collapse this into one blended number. Build a matrix of contract length x tariff type x payment method x usage band and calculate per cell.

Segment: [Dual-fuel, Fixed-24mo, Direct debit, High-usage]
  Monthly margin = usage-based revenue × margin %
  Survival curve = 1.0 during term, then renewal hazard
  Move-out hazard applied to all months
  LTV = Σ (month 1..60) margin(t) × survival(t) × (1+r)^-t

Twelve to twenty-four cells is realistic for a mid-size retailer. Two counting errors ruin the output. First, modeling per meter rather than per household double-counts dual-fuel customers and inflates the book. Second, holding today's margin flat for five years ignores back-book decay: where regulators require a better-offer prompt on the bill, as Victoria does, tenured margin drifts down by design. Regulator monitoring reports segment households by fuel and contract status for the same reason, as the European Commission's energy retail market work shows.

Where this changes marketing decisions

• Reallocate acquisition budget by cell, not by headline cost per sign-up. A cheap single-fuel variable switcher can be the worse buy.
• Set willingness to pay for a save or a win-back against that segment's modeled LTV, using the intervention costs the benchmarks lesson establishes rather than a company average.
• Watch for the mix-shift illusion in board reporting. Blended LTV rises when you acquire fewer variable customers, even if nothing about retention improved. Report the matrix, not the average.
• Price the prepayment and solar cohorts on full cost to serve before deciding they are unprofitable; the fix is often a product change rather than a spend cut.

Key Takeaways

• Fixed-term, variable and prepayment cohorts need three different survival curves, plus a move-out hazard that applies to all of them.
• Amber Electric's subscription plus pass-through model makes margin independent of usage and wholesale price, and makes churn dependent on market events instead of the renewal calendar.
• Prepayment revenue can stop without any churn event; model it on top-up behaviour, not account status.
• Solar exporters look worthless on a kWh-only model, which is why AGL and Origin measure that base on a wider product mix.
• Discount the stream (an 8% rate turns five years of margin into about four) and build the matrix at twelve to twenty-four cells before comparing anything to acquisition cost.