Marketing science studio
Marscie builds the models behind better media decisions — mix models, lift experiments and lifetime value — and wires them into the bidders, deals and clean rooms where your budget is really spent.
Why we exist
Most companies have a measurement team that produces decks and a media team that pulls levers — and nothing connecting the two. The model says one thing; the bidder optimizes for another.
Marscie closes that gap. One team that measures what works, and builds the machinery that acts on it.
Services
Hire us for a single engagement or the whole loop. Either way, the models, code and documentation are yours when we leave.
Find out what is truly incremental, and what each customer is worth.
Bayesian mix models that estimate the contribution of every channel — online, offline, price, promo and seasonality — with honest uncertainty instead of false precision.
Controlled regional experiments that show what a channel causes — no cookies, no user-level tracking, no trusting a platform's own scorecard.
Predictive models that score what a customer will be worth early in their life, so acquisition can be judged on value rather than on first-order cost.
Put the measurement where the money moves: in the auction.
Bidding algorithms for your DSP that optimize toward your business outcome — margin, predicted lifetime value, incremental reach — instead of a generic click or view.
Curated supply packages built on the sell side, combining inventory quality, attention and audience signals into deals your traders can simply switch on.
Privacy-safe collaboration between your first-party data and your partners' — designed, implemented and turned into analyses people actually run.
Turn the method into a product your teams — or your customers — use daily.
We help companies design and ship their own AI products: from a first prototype to a production system with evaluation, guardrails and a team that knows how to run it.
Approach
Short phases, working outputs at the end of each, and no step that depends on a black box.
We review your data, platforms and current measurement, and agree on the one decision the work has to improve.
We build the model or experiment in your environment, with assumptions written down and results you can challenge.
Findings become budget plans, bidding logic, curated deals or product features — whatever changes the outcome.
Code, documentation and training go to your team. Stay with us for refreshes, or run it yourselves.
Principles
FAQ
They answer different questions. A mix model covers every channel over time and guides budget allocation; a geo experiment proves the causal effect of one channel or tactic. The strongest setups use experiments to calibrate the mix model, and we usually recommend starting with whichever unblocks your next budget decision.
For a mix model, roughly two years of weekly spend and outcome data is a comfortable starting point. Geo experiments need regional sales data and the ability to vary spend by market. If you're short of either, the audit phase will tell you what's feasible before you commit to a build.
We are platform-agnostic. Custom bidding work happens in the DSP you already use, curation in the SSPs and curation platforms that fit your supply, and clean room work in whichever environment your data and partners sit in.
No. We work alongside them. Your agency or traders keep running campaigns; we supply the models, algorithms and deals that make those campaigns smarter, and train your team to maintain them.
It usually starts with a short discovery to pick a use case worth building and confirm the data supports it, followed by a working prototype, then hardening for production with evaluation and monitoring in place. You can stop after any phase.
Contact
A budget split, a bidding strategy, a product idea. We'll reply with how we'd approach it and whether we're the right people for it.
hello@marscie.com