The 4 Models Behind CTC's 3% Accurate Revenue Forecast

Most brands plan their year around gut feel, a spreadsheet, and last year's numbers. They pick a revenue target, reverse-engineer a spend number, and hope the math holds. When it does not, they find out 30 days too late. CTC runs a different system. Four proprietary models, connected in sequence, produce a daily profit and loss forecast with 3.1% accuracy across more than $3 billion in managed gross merchandise value. That accuracy is not luck. It is architecture.

The System Behind the Forecast

The four models form a chain, not a list. Each one feeds the next. The Spending Power Model establishes how efficiently any level of ad spend converts into new customer revenue. Those new customers pass into the Retention Model, which forecasts how much revenue each acquisition cohort returns in every subsequent month. The Event Effect Model then distributes that revenue across individual days with precision, accounting for sales, product drops, and VIP campaigns without ever breaking the monthly budget target. Finally, the Creative Demand Model closes the loop by calculating how many new ads must be produced to sustain the plan. Every dollar, every day, accounted for.

Forecasting is as much an exercise in execution as it is in planning. The model creates the map. The profit engineer navigates it daily.

Model 1: The Spending Power Model

The Spending Power Model is built on ACoNS, defined as total ad spend divided by new customer revenue. It measures how many cents of ad spend it takes to generate one dollar from a first-time buyer. Lower ACoNS means higher efficiency. As spend increases, ACoNS rises because each incremental dollar works harder to reach new audiences. The model captures this relationship as a linear equation, a slope and an intercept, for every brand and every month.

The slope is what CTC calls spending power. Specifically, spending power is defined as one divided by the slope: how many additional dollars of monthly spend it takes to degrade ACoNS by one point. A brand with high spending power, say $1 million, can absorb significant budget increases before efficiency drops. A brand with low spending power, say $200,000, will see acquisition costs spike quickly with any scale. This degradation curve is unique to each brand and is the first building block of the monthly revenue forecast.

The model surfaces three distinct optimization points on that curve, each representing a fundamentally different business strategy. The first is max contribution margin: the spend level that produces the most new customer margin dollars in month one. The second is max revenue at break-even: the level where a brand maximizes top-line new customer revenue without going negative on contribution margin. The third is max lifetime contribution margin: a brand accepts lower month-one margin in exchange for higher acquisition volume, then captures the long-tail LTV of those customers over a defined time horizon of 60, 90, or 120 days. The model makes all three scenarios visible and computable inside Statlas. No guessing which trade-off to make.

Model 2: The Retention Model

New customers are not one-time transactions. They are the seed of a compounding revenue stream. The Retention Model quantifies exactly how that stream unfolds. The core metric is LTV lift: the percentage of a cohort's first-order revenue that returns as repeat revenue in each subsequent month. If a January cohort spent $100,000 on their first orders, and month-three LTV lift for that cohort is 2%, the model expects $2,000 in returning revenue from those customers in March.

LTV lift follows an exponential decay curve. It peaks in the first month after acquisition, drops sharply through the next 60 days, and then flattens to a low, stable terminal rate. For any future month, the total returning customer revenue is the sum of all active cohort contributions: every past cohort's original revenue multiplied by its predicted LTV lift at that cohort's age. December's returning revenue comes from November at month-one lift, October at month-two lift, September at month-three lift, and every earlier cohort at its corresponding rate.

A growing active customer file predicts revenue growth. A shrinking one signals trouble, no matter how large the total email list looks. Many of those customers are lapsed and not coming back.

Because the Retention Model draws directly from the Spending Power Model's new customer output, the two models must operate as a connected system. An increase in acquisition volume ripples forward into every future month's returning revenue projections. The models do not run in isolation. They compose.

Model 3: The Event Effect Model

Monthly totals are necessary but not sufficient. Brands need to know what each individual day should look like, because deviation from that daily plan is an early warning signal. A sale day that underperforms its modeled spike tells you something is wrong before the month-end result confirms it. The Event Effect Model creates that daily precision.

For every historical marketing event, the model measures two signals. The first is return ratio: how much that event shifts revenue toward returning customers relative to a normal day. Loyalty campaigns and VIP drops consistently pull existing customers back into purchase. The second signal is AMER lift: how much more or less efficiently paid acquisition performed during the event. High-intent sale periods often spike conversion rate, making acquisition dollars go further than on a baseline day.

The model ingests more than two years of marketing calendar data, email behavior, and SMS behavior. Every past event is tagged by type: promotion, product launch, VIP drop, loyalty campaign. That historical database forms the baseline for predicting future events of the same type. Day-of-week effects layer on top: every brand has predictable within-week revenue patterns tied to its category and buyer habits. Together, these two signals produce a daily distribution that tells the profit engineer exactly where in the month revenue should land, and how hard each day should run.

The critical constraint: event effects are applied as relative lifts, not absolute overrides. If a month is planned at $30,000 spend and a 3.0 AMER, the daily outcomes always reconcile to those monthly targets. Event days run hotter. Surrounding days compress proportionally. The budget constraint is never violated. The model does not change the total. It reshapes when that total lands.

Models Are Tools, Not Oracles

The four models are not set-it-and-forget-it systems. CTC's data team tunes each one per brand on a defined cadence. Spending Power and Retention Models update quarterly. The Event Effect Model updates continuously as new events are added and actualized. Each model has configuration controls that allow the team to adjust for brand-specific behavior as it evolves.

When predictions deviate meaningfully from reality, the question is not whether the model is broken. The question is what changed in the business that the model has not yet captured. Sometimes a retune is needed. Sometimes the brand has entered a genuinely new operating environment and the model needs more data before it can be trusted again. The operating principle is direct: when the map no longer matches the territory, update the map.

The value of the model is not in producing a perfect point estimate. It is in surfacing deviations quickly enough that the profit engineer can course-correct before small problems compound into large ones. Precision forecasting only creates value if the system reacts to what it finds.

Frequently Asked Questions

See the Modeling System in Action

The Prophit Engine puts all four models to work for your brand, forecasting every dollar, every day, with the precision your growth plan requires.

Talk to Us