Introducing Horizon: Lifesight’s Open-Source Marketing Forecasting Engine

Horizon is Lifesight’s patent-backed forecasting engine, now publicly available for anyone to inspect, test and improve. It’s the first step in opening up the methodology behind our measurement platform.

Million-dollar decisions are being made by marketing teams relying on forecasts whose internal mechanics remain completely hidden. A model says next quarter’s budget should skew toward one particular channel and away from another, but doesn’t specifically call out why, and marketers are moving budget based on these predictions. But how the model arrived at that number and whether it’s actually accurate, for most forecasting tools,is not shared. You’re asked to trust the output on faith alone and move on.

We don’t think that’s good enough.

Today we’re open-sourcing Horizon, our patent-backed forecasting engine built specifically for budget forecasting and marketing measurement. Anyone from marketers, developers, to researchers, can now examine the methodology behind our prediction models, validate the results against public data, and improve on them.

Why we’re doing this

The stakes here aren’t abstract. Marketing teams increasingly lean on forecasting models to decide where to put millions of dollars in advertising, and the confidence isn’t always matching the reality. Recent research from Incubeta found that 41% of marketing leaders admit some portion of their investment is failing to deliver its full value.

Yet the forecasting engines guiding those decisions are almost always proprietary black boxes. That makes it hard for brands to understand how a prediction was generated, harder still to validate its accuracy, and nearly impossible to build real confidence in it. Forecasting has become one of the most consequential inputs into marketing strategy, and one of the least transparent.

Opening up Horizon is our answer to that. It’s the first step in a broader initiative to publicly share the methodologies behind our causal measurement platform, starting, deliberately, with one of the core technologies commercial measurement vendors have always kept locked away.

“Marketing teams are making significant investment decisions based on forecasting models they often can’t inspect,” said Tobin Thomas, CEO and co-founder of Lifesight. “We don’t believe customers should have to trust us blindly. By open-sourcing Horizon, we’re giving marketers the opportunity to examine the methodology, test it themselves and hold us accountable.”

How Horizon works

Horizon is built on an ensemble forecasting methodology: rather than betting everything on a single model, it combines multiple forecasting models to produce budget forecasts that are more accurate and more reliable than any one model could deliver on its own.

The important part is what ships alongside the code. We’ve published Horizon’s benchmarking results, its validation methodology, and full supporting documentation, everything you’d need to reproduce our results, evaluate the approach, and contribute improvements. Horizon has been benchmarked against public datasets and leading forecasting models, including Prophet and NeuralProphet, so its performance can be judged on evidence rather than assertion.

“Publishing the code is the easy part,” said Rajeev Nair, Chief Product Officer and co-founder at Lifesight. “Forecasting models should be evaluated on reproducible performance, not marketing claims. That’s why we published Horizon alongside its benchmarks, validation methodology and supporting documentation, allowing anyone to independently evaluate the model against public datasets.”

Validated by science, not by us

Transparency only means something if the methodology holds up to outside scrutiny. That’s why Horizon is reviewed by our newly established, independent Scientific Advisory Council, a growing panel of experts who evaluate our research and methodologies at arm’s length from the product teams.

The council’s inaugural member is Dr. Ron Berman, Associate Professor of Marketing at The Wharton School, who advises us on causal inference, online experimentation and marketing analytics. As we expand this open-source initiative, that independent layer of review is what keeps it honest.

Get Horizon

Horizon, its supporting documentation and all benchmarking materials are available today on GitHub at github.com/lifesight. If you’re a marketer, dig into the methodology and see how the forecasts guiding your budget are actually built. If you’re a developer or researcher, explore the repository and contribute in line with the project’s contribution guidelines, we’d genuinely welcome the scrutiny.

Publishing the code is the easy part. Being held to it is the point.

Explore Horizon on GitHub