We price our work as a share of measured improvement, which means we care enormously about how fast an improvement can be measured. Energy is the best vertical we have found on that axis: prices settle hourly, positions are marked continuously, and a forecast can be scored against reality in weeks. Compare that with agriculture, where a yield model pays off once per harvest and you find out whether it worked a year later.
Why this market, right now
Brazil's free energy market is in the middle of a legally mandated expansion. Lei 15.269/2025, sanctioned in November 2025, sets the timeline for opening the market to low-voltage consumers, enabling access for roughly six million small businesses and 400,000 small industries within 24 months, with a final deadline of 25 November 2028 affecting more than 70 million consumers.
The growth is already visible without the low-voltage tranche. ABRACEEL reported 82,958 consumer units in the free market as of late 2025, up 21,547 in twelve months for accumulated growth of 35%. Industrial consumption is 95% free-market and effectively saturated; commercial consumption moved from 41% to 47% in a year and is where the current growth lives. The free market now represents about 43% of national electricity consumption.
That expansion creates buyers who need forecasting capability they did not previously need, on a deadline set by law rather than by budget cycles.
Where we work
Price and load forecasting for energy retailers
Forecast error in a retail energy portfolio is not an accuracy statistic, it is a settlement outcome. We build short and medium-horizon load and price models and, importantly, quantile forecasts rather than point forecasts - because a trader hedging a position needs to know whether a week is a safe bet or a coin flip, and a single expected value hides exactly that.
Portfolio and counterparty risk
CCEE's prudential monitoring covers a defined, enumerable population of licensed retailers, and a meaningful minority of them carry leverage or disclosure flags at any given time. Firms in that position have both the sharpest commercial need for better volume and price forecasting and the strongest regulatory incentive to demonstrate control.
Curtailment prediction for generation owners
Curtailed generation is power you produced and never got paid for. ABSOLAR projected a 7% contraction in Brazilian solar expansion for 2026, citing curtailment and connection obstacles, against roughly 32 GW of installed photovoltaic capacity and more than 4.3 million distributed generation systems. Predicting constrained-off generation and its cash impact changes hedging and contracting decisions for asset owners, and the realized loss is measurable against settlement data.
Non-technical loss detection for distributors
ANEEL reported in July 2026 that non-technical losses reached 45.0 TWh in 2025, or 7.1% of energy injected, with technical losses valued at R$11.7 billion. Abradee logged R$11.3 billion of damage, R$7.8 billion of it passed through into tariffs. The problem is concentrated: the top ten distributors account for 76.2% of losses. Computer vision over aerial, satellite, and vehicle imagery can localize irregular connections and prioritize inspection crew routing - and the fee attaches to recovered billed energy per inspection, a figure booked as taxable revenue and therefore audited by someone other than us.
What we will tell you before you ask
CCEE already publishes a free baseline. Daily PLD from the DESSEM model, plus a public calendar of sessions covering methodology and scenario projections. Any vendor who does not mention this is selling you something you can partly get for nothing. A paid model has to beat the free public output on your portfolio - which is precisely what a parallel run is for, and why we would rather prove it than assert it.
Two more things worth saying plainly. First, some competing forecasting research in this sector was funded through the regulated R&D levy rather than by customer fees, which means incumbents can price below their true development cost - a real competitive factor you should weigh. Second, the incumbents are technically sophisticated; this is not a market where a generic model wins on novelty. Our argument is not that we are cleverer, it is that we will run against your current method in parallel and only get paid if we beat it.
How we would prove it
- Parallel run. Our forecast runs alongside your incumbent method for an agreed window. Nothing changes in your operation.
- Scored on realized settlement, not on backtest accuracy - measured in cost or margin per MWh against the same hours.
- Agreed in writing first: the metric, the window, the attribution method, our percentage, and the cap.
- You pay from the measured difference. If we do not beat your current method, there is no performance fee.
The full mechanics are on our pricing page.