Each application answers one question that matters in people's lives: what your savings will earn, how fast an outbreak is growing, whether a new law changed anything. You get the answer with a range, the main factors behind it and a record of how close our past answers came. Below are the questions we work on in each field.
Where we start. Our first live work is in financial markets, where every forecast gets fast, public feedback. In the other fields we start with a small, time-limited pilot alongside a partner who owns the problem. If that could be you, we would like to hear from you.
Get in touchMarkets trade around the clock and every price is public, so a forecast is scored within hours. That makes them the toughest test for our methods, and what we learn there carries into every other field.
We build quantile bands and check their coverage on a sealed year before making any claim, so the bands catch what they say they catch.
We measure volatility and trend from price alone and sort the market into calm, trending or stressed. A model can then adjust to the regime, or stand down in it.
We check every input and output for drift, and set position limits and kill switches before the first trade. When something drifts, position size comes down before losses compound.
Official economic figures arrive weeks or months after the fact and are often revised. Timely data can give a first estimate sooner, as long as it is honest about how uncertain it is.
We nowcast from timely indicators such as card payments, job postings and energy use, and score each estimate against the first official release. You get an early estimate with a range, and a record of how close past estimates came.
We build repeat-sales and hedonic price indices that control for what is on offer. The result is a regional index that separates real market moves from changes in the mix of homes.
We combine employment and input-output data to trace how a shock in one sector spreads to the others. Planners get a ranked view of which regions are most exposed, early enough to act.
Food, health and public services run on decisions made with incomplete data. Better estimates can mean less waste, earlier warnings and fewer surprises for the people involved.
We forecast demand from past distribution, the calendar and the weather, and check each forecast against what was actually collected. That means less food thrown away and fewer days when supplies run short.
We estimate current cases with a correction for reporting delays and measure growth over short windows. You get a current estimate with a range, in place of last week's number.
We project the population from births, deaths and migration as a set of scenarios, not a single number. A council gets a range it can budget against, updated as new data arrives.
Laws and policies are introduced to change something, but their effect is rarely measured with care. A good evaluation separates what a rule changed from what would have happened anyway.
We compare against regions or periods where nothing changed, using difference-in-differences or synthetic controls, and run placebo tests to see whether the result holds. You get an effect size and a plain statement of how well it holds up.
We use interrupted time series that test for earlier trends and allow for seasonal patterns. You get a clear answer on whether the timing points to the rule or to something else.
We write the evaluation plan in advance: the outcome, the data, the comparison group and the smallest effect the data could detect. You see what an evaluation can and cannot show before any money is spent.
Warming is a slow trend under fast, noisy weather, and its effects show up in ecosystems as well as in temperature records. The work is telling the two apart, and helping the people who have to adapt decide what to do first.
We compare each year with a fixed reference period and test the trend in a way that allows for one year depending on the last. The result is a trend estimate with an interval that does not overstate certainty.
We model survey counts and correct for observer effort, site changes and missed detections. Conservation groups get a trend they can rely on, with its uncertainty stated.
We turn local heat, rainfall and runoff data into one readable picture and compare each measure on the same baseline. A council gets a comparison it can discuss and act on.
Power has to be produced the moment it is used, and more of it now depends on the weather. Better forecasts mean less backup generation, fewer price spikes and a grid that can take more wind and solar.
We map ensemble weather forecasts to demand and generation, and score them against what was then measured. A grid operator or energy desk gets probabilistic ranges to plan around.
We forecast prices as probabilities and check the extremes separately, since the tails matter most. You get warning of tight hours early enough to shift load or hedge.
We build scenario models from metered consumption and test them on areas where adoption is already high. The result is an estimate of when and where the grid will need reinforcing.