Our Technology: Revolutionary Forecast Modeling

Our experience is solving complex systems problems. We found that our Consumer Goods colleagues were dealing with similar problems. They were using data to drive marketing plans and forecasts but remained unsatisfied with outcomes. Clearly something was wrong.

Most teams use one set of tools to make plans, and are then burdened to somehow turn those plans into forecasts. This is backwards. We realized this was due to two specific limitations of common statistical and "black box" A.I. models. We built our technology to empower teams to use the forecast to know what to plan.


Decision Granularity

Most models don't operate at the level brands actually spend their money, like treating $1 of Facebook like $1 of TV.


Concurrent Causes

Most models aren't able to partition causality amongst concurrent drivers, like every Black Friday season with low prices and high ads.

A Catalog of Causality

Our team of MBAs, PhDs, computational scientists, and applied mathematicians realized that we needed a different modeling approach to unlock true causality of demand and make better business decisions.

"Black box" and statistical models rely solely on historical data - they may have millions of data points, but they never see the variation you'd need to solve the concurrent causes problem.

Those models lack the causal structure to understand what has happened and what could potentially happen if they make a change, so we went to the literature to build a causal model of consumer purchasing.

What really drives sales up or down? We use Bayesian A.I. techniques to get to the best solution given all possibilities


Price effect on sales

We capture how likely consumers are to purchase at each retail channel as price varies

Susceptibility to Ads

Total impact of advertising

We capture how likely consumers are to purchase at each retail channel as media exposure varies

Impact of Seasonality

Varying Purchase Likelihood

We model true seasonality of a product, excluding all other input factors.

Actionable Forecast

Causal Model with All Effects

We make this model totally accessible to our users, allowing them to explore the results and explore the results and build plans


Our solutions to common implementation obstacles

The data isn't assembled

We integrate your data–even agency APIs & those spreadsheets. We leverage point of sale data, and merge internal data with external data to see what's really true.

You don't know which data is bad

Our “catalog of causation” auto-flags unexplainable data (unexplained high prices charged at a certain retailer, for example). That data can then be discarded or corrected as appropriate.

"Actionable insights" aren't

We serve what-if scenario planning apps, not dashboards. We provide the right dashboards, but those are also integrated into the business process, so user can directly submit plans for approval.

You need answers, not tracking tools

Our optimizers tell you the best answer and allow granular overrides. Users can simulate and optimize each individual element of the plan at the same time.