Top › Column

PESO as a Formula: Econometrics and AI for Explainable Marketing

October 25, 2026 | By Tomonori Hata, Representative Member, HATA LLC.

The PESO model is a way of designing communication by combining four types of media: paid, earned, shared and owned. As an idea, however, it has a weakness: it is hard to explain with numbers which medium contributed how much to the results. This article introduces a concept for addressing that weakness with econometrics and AI.

1. The problem: PESO is an idea, and results are hard to explain with numbers

Paid media makes it easy to follow cost and results. By contrast, the contributions of press coverage and word of mouth (earned), social media (shared) and your own site (owned) are harder to see, so attention tends to go to advertising spend. What executives want to know is where the next yen and the next hour should go.

2. An existing approach: marketing mix modeling (MMM)

MMM uses statistical models to estimate the relationship between marketing activities and sales. It builds in two ideas about time:

• Effects accumulate over time and appear with a delay (adstock)
• The more you spend, the less each additional unit achieves (saturation)

In recent years, Bayesian estimation, which includes uncertainty in the estimate, and calibration with experiments have spread. Well-known tools include Meta’s Robyn, Google’s Meridian and the open-source PyMC-Marketing.

3. The concept: treating PESO as a function of time

The core of the concept is to combine the four PESO media into one model as functions that change over time. In words, it is simple:

Results(t) = baseline demand(t) + effects of the four media(t) + seasonality and external factors + error

• Each medium’s effect appears with a delay and levels off as spending grows
• The size of an effect is not fixed but changes over time (state-space models, time-varying coefficients)
• The four media influence each other. For example, an ad becomes a topic, is shared on social media, and leads visitors to your site

Among existing methods, a combination of MMM, Bayesian state-space models, time-varying coefficients and dynamic generalized linear models is the closest idea. In this article, we tentatively call the concept a “dynamic generalized model.”

4. The role of AI

AI does not make decisions in place of the model; it helps make the model usable in practice.

• Organize scattered data into one table
• Turn text such as press coverage and word of mouth into numbers (counts, the share of favorable reactions)
• Estimate results when budgets or activities change (scenarios)
• Explain results in words executives can understand

5. Limits and cautions

• Correlation and causation: A rise in sales in a month with ads does not prove the ads worked. Small experiments are needed to check.
• Amount of data: Effects that change over time are hard to capture reliably when observations are few. Remedies include a longer period, coarser aggregation and prior knowledge.
• Difficulty of quantification: Earned and shared media are quantified with proxy indicators such as the number of mentions and the share of favorable reactions.
• The model is a hypothesis: It is not finished once built; it is reviewed regularly.

6. Putting it into practice: five stages

• 1. Take stock of data: Gather records of sales, ad spending, social media, the website and press coverage
• 2. Build a basic model: Start with a simple MMM that includes delay and saturation
• 3. Validate with small experiments: Check effects with tests split by region or period
• 4. Add time variation and interaction: Add the relationships in which the four media influence each other
• 5. Operate: Update regularly and use it for budget allocation decisions

HATA’s approach (as a concept)

HATA LLC. treats this concept as a research theme and is considering how to bring the ideas of econometrics into media strategy based on the PESO model. See Media Strategy (Japanese).

More articles

• How to Make a Recruitment Video: Five Steps from Planning to Release
• Short-Form vs. Long-Form Video: Choosing by Purpose
• What Is IP Business? Expanding the Value of Content
• What Is the PESO Model? Using Paid, Earned, Shared and Owned Media