CAFE replays your own sales history with the prices it would have set and tells you how much more you would have earned. Free, on your computer, before you pay anything.
14 days · 5 backtests · No card, no call
Reliable estimate of extra profit
Over 65 periods, against the prices that were actually charged.
Cumulative profit difference, CAFE minus actual
The range includes zero, and the report says so in the first paragraph.
Computed on this computer
What it does for a buyer, what runs under the hood for the people who sign off on it, and why the number it gives you holds up.
The CAFE mark is the product in a single gesture: your sales history closes around a line that climbs to one decision, the price, in signal orange. Turn it however you like; it always comes back facing you.
Load your price and sales history and get an executive report: how much more CAFE's prices would have earned over your past periods, against what you actually charged. It runs the real engine, on your computer.
Write to us and we send you the installer for Windows, macOS or Linux. It runs on your computer: your data never leaves it.
Name and email under “Calibration key”. The key activates right away: there is no call and no card.
A CSV with one row per period: date (YYYY-MM-DD), price and units sold, plus cost if you want the result in money. At least 52 periods; two years of weeks is better. The trial evaluates your history up to 30 days before the key was issued.
What the report tells you
Your data never leaves your computer: the backtest and the report open no network connection. The trial report gives the gain, its range and the price response; the prices CAFE would have recommended period by period, the price for your next period, the price test and pricing a whole catalog come with a license. The trial uses your history up to 30 days before it was issued (more recent periods are left out) and runs 5 backtests. One trial per computer and per email.
Limited on purpose
These limits are signed into the trial key. Choose the products that matter most before you start.
14 days · 5 backtests · No card, no call
The sample is the full-license report of a simulated product, so the true answer is known and printed next to the estimate. It includes the recommended prices, which a trial report does not.
Every number below was measured, most of them on products the tuning never saw. Version 4.7 brings together everything built since 4.3.2.
Import from Shopify, WooCommerce, Odoo, SAP Business One or Excel, price every product, approve or adjust each one with your rules checked, and export a bulk-upload file. On simulated catalogs: profit left on the table from 5.96% to 3.29%.
On a real retailer's sales, the 80% demand range now holds 83% of what sold (4.7.0: 44%), and the forecast after a price move beats the recent average (84% against 94% error).
Products that sell zero units in many weeks are forecast with their own probability of selling. All 1,393 real products: 0 errors (4.7.0 stopped).
Professional and Enterprise: a random split of your products, CAFE's prices on one part, yours on the other, and the extra profit with a 90% range. In simulations the range held the truth 89–92% of the time.
On the 14 synthetic markets, 13 price exactly as in 4.7.0; the catalog benchmark improved to 3.33% of profit lost.
A paid license now works as soon as you paste the key: no files to create. Every download comes with a step-by-step user guide in Spanish and English.
Switch to compare the same measurement in both versions.
Target 80%: closer is more honest
Lower is better
1,393 products of a real retailer
The starting point: one product at a time.
Catalogs: short histories borrow from their category.
Impact tracker, and categories whose prices followed demand.
Products that move each other's sales, priced together.
Fixes to what 4.6.0 left measured.
Catalog console: import, review and export.
Run on real sales: no product stops a catalog.
Honest ranges when the price moves; clean data as before.
One retailer, no costs, no inventory: real-data numbers are evidence, not a promise. Your own backtest is the number that counts.
14 synthetic markets whose best price is known exactly, each simulated 20 times. CAFE is scored on how much of the best achievable profit its recommended price gives up. Lower is better.
of runs: the 90% band held the true best price
of runs: the demand range held the realized demand (stated: 80%)
average gap between the backtest's reliable estimate and the true gain
runs: 14 markets × 20 replicas, including the ones CAFE loses
Share of the best achievable profit lost
7 markets: noise, seasonality, events, competitor, cost drift, stockouts
Where extrapolating a fitted curve fails
Confounded history with a valid instrument
No method can recover the price response; CAFE flags it and proposes a test
The hardest market: CAFE warns, and still does poorly
The band for the best price contained the true best price in 91.4% of 280 runs, and the demand range contained the realized demand in 79.7%. These are synthetic markets: they show the method recovers a known truth, not what it will earn in your business.
The flow stays readable: estimate how demand reacts to price, check whether the data support that estimate, apply your business rules, and choose the price that maximizes one declared objective.
CAFE reads price, demand, inventory, cost, seasonality, and competitor signals and estimates the price elasticity causally, using cost changes as an instrument when they exist.
If prices barely moved or the estimate is too imprecise, CAFE reports that the elasticity is not identified instead of presenting a guess as an optimum.
Price bounds, minimum margin, minimum advertised price, maximum change per period, inventory coverage and price endings are hard limits. By default the price stays within your observed price range plus 5%.
The optimizer maximizes expected profit minus a risk term on the worst 10% of outcomes, and returns the price with its uncertainty band, the cost of each binding rule, and an explanation.
CAFE is designed to feel credible to technical stakeholders while staying practical for operators who need clear actions rather than abstract analytics.
Each recommendation states the elasticity and how it was identified, the cost used, the rules that hold the price back and what each one costs, in plain language. Enterprise adds an audit record per decision.
One declared objective: expected profit minus a risk term, with a single risk-tolerance dial. Business rules are constraints, never traded off against profit, and CAFE does not recommend prices far outside what you have charged unless you allow it.
The engine runs offline, without external APIs, and can be delivered as a protected internal package with hardware-bound licensing.
One-time licensing, compiled engine modules verified to price exactly like the source, optional obfuscation, and hardware-bound license enforcement support tighter delivery control.
CAFE is best suited for pricing-heavy retail and e-commerce environments where pricing is reviewed frequently, data is available internally, and decisions need a clearer analytical basis.
For environments balancing inventory, margin discipline, and regular competitor movement.
For teams that already monitor demand shifts, campaigns, and competitive pricing but need better decision structure.
For organizations that want an internal pricing capability with explainable outputs and controlled delivery.
A one-time license for internal use, comparable to one year of a mid-tier SaaS pricing tool. Core already includes the full pricing decision and the catalog console; the higher tiers add the impact tracker, reports, versioning, audit and feedback.
Start at zero
Run the free backtest first. Buy only if your own number justifies it.
Swipe to compare plans
One-time payment
The full pricing decision: causal elasticity, a warning when the data are not enough, a proposed price test, business rules and calibrated uncertainty bands. For a catalog: products with little history priced from their category, and products that compete priced together. Catalog console with importers. Hardware-bound license and guided installation.
One-time payment
Adds the impact tracker (what CAFE generated, measured against a random split of your products), simulated scenario reports, model versioning, and an optional causal-graph diagnostic. The recommended price is the same as in Core. One-time cost comparable to one year of a mid-tier SaaS pricing tool.
One-time payment
Adds a full audit trail for every pricing decision and a feedback loop that adjusts the demand forecast with real outcomes. For organizations requiring compliance logging.
Prices in USD. One-time payment: no subscription, no automatic renewal, no recurring charges. Applicable tax is calculated at checkout by Paddle, the merchant of record.
Buyers usually want the same things clarified: what the engine does, what gets delivered, who owns operations, and how risk is handled.
Start with the free trial. If you would rather talk scope, deployment or licensing first, write to us.
14 days · 5 backtests · No card, no call
