On September 25, Lifesight -- a marketing measurement platform that sells tiered enterprise plans for combined MMM, incrementality testing, and attribution -- published the code, methodology, documentation, and benchmarking materials for Horizon, its demand forecasting engine, on GitHub. Anyone can clone it. Anyone can audit it. Anyone can run it on their own data. Lifesight co-founder and chief product officer Rajeev Nair told AdExchanger the company did this to build trust and establish that forecasting should be part of every measurement system. That explanation is charitable. The honest explanation is more uncomfortable for the rest of the measurement category.

What Lifesight just put on the table

The tool Lifesight open-sourced is not the MMM itself. It is the demand forecasting engine that runs underneath it -- the piece that predicts what baseline revenue would have been if the brand had not touched media, pricing, or promotions at all. Modelers call this the baseline, and Dr. Ron Berman, an associate professor of marketing at Wharton and the first member of Lifesight's newly created Scientific Advisory Council, framed the problem precisely in the same AdExchanger piece. Extrapolating from measurement models is usually accurate only for very small changes. The baseline forecast is where MMM runs out of road. That piece -- the part that decides what would have happened anyway -- is what Lifesight just made free.

This matters more than it sounds. For an established brand, somewhere between 50% and 60% of revenue comes from factors that have nothing to do with marketing -- seasonality, brand loyalty, word of mouth, repeat purchases, distribution, shelf placement. That revenue would arrive whether or not a single ad ran. The marketing-attributable lift is the difference between the forecast and the baseline, which means the baseline forecast is what every marketing measurement vendor is actually selling. The marketing model is the wrapper. The baseline is the engine. Lifesight just gave the engine away for free.

Why the rest of the measurement category should be nervous

Google Meridian, Meta Robyn, and PyMC-Marketing -- the three open-source MMM libraries most often cited as benchmarks -- all attack the same problem from the same angle. They make the historical decomposition more transparent. What they do not do is what Horizon does. They cannot answer the second question. They cannot forecast what happens tomorrow. They are dashboards looking backward. They are not decision systems looking forward.

Horizon's method is the one piece of engineering worth stealing. It runs multiple forecasting models at the same time and combines their outputs instead of betting everything on one, a technique called ensemble forecasting. No single model catches every signal, but averaged together they land closer than any one of them alone. Berman's point is that this is settled in academic forecasting research and barely known in marketing practice, where teams still pick a winner, write it down, and get burned when the winner changes next quarter. Horizon also runs alongside whatever measurement stack you already have rather than replacing it, so it layers over Lifesight MMM, Google Meridian, or something your team built in-house. Nair is candid that Horizon is a layer over existing open-source libraries rather than a proprietary breakthrough. The differentiation is in the execution: making it easy for brands to orchestrate, calibrate, and then make a decision with it.

Lifesight's bet is that the next three years of measurement competition will run through forecasting, not historical attribution, and that being the vendor whose forecasting engine is publicly audited, peer-reviewed by a Wharton-led Scientific Advisory Council, and reproducible on customer data is a defensible position. The bet is that marketing measurement buyers in 2027 will choose vendors based on what they can defend about the future, not based on what they can show about the past.

The measurement vendors that have nothing comparable to open-source are now in an uncomfortable position. Their customers will start asking whether their forecasting methodology is auditable. Their customers will start asking what their baseline model does that Horizon does not. Their customers will start asking whether the seven-figure annual contract they signed is paying for measurement or paying for a black box. Lifesight just gave every measurement buyer in the market a free benchmark to evaluate them against.

What marketers should do with this

First, ask your current measurement vendor for the same artifacts Lifesight published. The code is not strictly necessary -- most marketing teams do not have the data engineering to evaluate it. The methodology, the documentation, the benchmarking against independent data sets, and the peer review are. If your vendor cannot produce those artifacts, your measurement is not auditable, and you cannot defend the number your CMO is presenting to the board.

Second, run Horizon on a slice of your own data before you re-sign any measurement contract in 2026 and 2027. You do not need to replace your vendor. You need an independent baseline to evaluate whether your vendor's baseline is reasonable. Lifesight just gave every marketer in the market the tool to build that independent baseline.

Third, decide what you actually want from a measurement partner in the forecasting era. Historical MMM answers what happened. Forecasting answers what is about to happen. The vendors that survive the next eighteen months will be the ones that answer both. The vendors that only answer the first question will be the ones marketing teams quietly stop renewing.

The measurement category has spent five years becoming more open. Open-source MMM arrived. Open-source incrementality arrived. Now open-source forecasting has arrived. The vendors that have been charging premium prices for black-box baselines are the ones who should be reading the Lifesight announcement most carefully. The future of marketing measurement is not a more accurate historical view. It is an auditable forecast of what happens next. Lifesight just gave that future away for free.

Sources