Vectorial Analytics -
Narrow Down the Unknown

strategic challenges

How to achieve incremental media and finance planning that's not just in marketing dashboards? What is the balance between growth and science? How to move away from the last click trap? Why are the vendor results not actionable enough? How do I reconcile different numbers from MMP, platforms, in-house attribution, experiments and MMM?

technical challenges

How to model a complex business with multiple sales channels? How does triangulation really work? Can we still approach incrementality even when experimental interventions are not possible? Can I trust the saturation curves? Does time-varying models account for CPM pricing change? How to combine pLTV and MMM?

Coming soon:
Get excited about Robyn 2.0 and her little sister Yva

Yes, Robyn was named after my first born daughter. Despite the unexpected halt of development, Robyn remained a popular choice of thousands of advertisers and agencies.Now we're excited to announce that Robyn 2.0 will get a major upgrade to combine the strength of all mainstream packages and rejoin the open source community, and her little sister called Yva will offer advanced commercial features!

Evolutionary Bayesian & Geo-model in Python

Robyn 2.0 is aiming to unify the strength of all mainstream open source MMM packages into one, matching the major features. Of course it'll be Bayesian, supports Geo-hierarchical model and in Python.

Native handling of baseline

The larger the business, the more important the baseline. It's the part that's not directly influenced by your steerable levers like media spend, and it's the part most models hide. Robyn 2.0 provides more control over baseline estimation.

One budget allocator that accepts all tools

BYOC - Bring your own curve! The new Robyn budget allocator 2.0 doesn't just work on Robyn. It also accepts the outcome from other open source packages, from your vendors or even from you spreadshee. Imagine allocate budget across two different country from two different tools.

DECOMP.RSSD, the upgraded Robyn DNA

One of the most powerful concept from Robyn translated into a Bayesian framework, eliminating extreme decomposition from the start. It's enhanced through an explicit baseline penalty term as well as calibration nuances.

Best model candidates, not just one

The pareto-optimal model candidate is another beloved feature from Robyn, providing multiple outcome that share the same statistical strength. We don't pretend to know better and leave the final choice to you.

Advanced commercial features

On- and offline sales joint modeling, non-experimental calibration, adstock & saturation calibration, campaign path modeling, advanced pLTV integration, and more to come soon in Yva!

Hi, nice to meet you, this is Gufeng!

I helped hundreds of advertisers and agencies from eComm, fintech, Retail, FMCG, gaming etc. across EMEA, APAC, LATAM, and NA over 15 years. Now let's chat!

• add me
• email me

Consult

I help your team and marketing analytics stack to grow into world class. My focus includes measurement stack and signal audit, triangulation strategy, media & budget planning, multi-layer test & learn strategy, metric context and AI readiness, open source measurement intro.

• Details

Build

Complex advertisers with multiple product categories and sales channels have specific challenges in modeling. I develop advanced features to tackle these, e.g. on- & offline sales model, non-experimental R&F calibration, adstock calibration, curve calibration, custom budget allocation, pLTV MMM integration.

• Details