💻 computer science

Data valuation model for non-monetary exchanges

This paper proposes a normative, choice-based data valuation model for non-monetary intracompany exchanges that quantifies value through user selection behavior and formalizes it as a cooperative game with a Shapley value to ensure fair, uniqueness-rewarding allocation of data product worth.

Original authors: Julia Blyumen, Eitan Farchi

Published 2026-06-05

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Julia Blyumen, Eitan Farchi

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Problem: How Do You Price a Ghost?

Imagine a world where you can copy a song, a recipe, or a piece of software infinitely without it ever getting used up. In the real world, if you sell a chair, you have one less chair to sell. But with data, once you make it, you can give it to a million people, and you still have the original.

Because data is so cheap to copy, traditional ways of pricing it (like "how much did it cost to make?" or "how much are people willing to pay?") break down. It's like trying to price water in a desert when you have an infinite supply of it right next to you.

Furthermore, many companies don't sell data; they share it internally for free. There is no money changing hands. So, how do you tell a data creator, "Great job, your product is valuable!" without using a dollar sign?

The Old Ways That Failed

The paper points out two common ways companies try to measure value, and why they are flawed:

The "Popularity Contest" (Downloads/Views):
- The Idea: "If 1,000 people downloaded this, it must be great."
- The Flaw: This only rewards the "superstars." It ignores the "long tail"—the unique, niche tools that only a few people need but are critical for them. It's like saying a movie with 10 million viewers is 100 times more valuable than a brilliant documentary watched by 100,000 people. It also encourages people to make "copycat" products just to get more clicks, rather than making something truly new.
The "Bundle" (The All-You-Can-Eat Buffet):
- The Idea: "Let's just sell a big package of 50 data products for one price."
- The Flaw: This hides the value of individual items. If you get a package with 1 amazing tool and 49 junk tools, the creator of the amazing tool gets the same credit as the junk. It kills the motivation to make high-quality, unique things because everything gets lumped together.

The New Solution: The "Attention Score"

The authors propose a new way to measure value based entirely on choice. They ask: How much did a user have to give up to choose this specific data product?

Think of a user's attention like a limited amount of money in their pocket. They can't buy everything.

If a user buys only Product A, they are saying, "I love this so much I didn't buy anything else." That is a huge vote of confidence.
If a user buys Product A, B, C, and D all together, they are saying, "I like all of these, but none of them are special enough to make me drop the others."

The Formula (Simplified):
The value of a product is calculated by adding up points from every user who chose it. But here is the twist: The more other things a user chose, the fewer points that user gives to your product.

Scenario A: User picks only your product. You get 1 full point.
Scenario B: User picks your product and 9 others. You get 0.1 points (1 divided by 10).

This means a product that is chosen by a small, dedicated group of people who only want that product can actually score higher than a product chosen by a huge crowd that also grabs 50 other things.

The "Fairness" Engine: The Shapley Value

The paper uses a fancy math concept called the Shapley Value (from Game Theory) to prove this system is fair.

Imagine a group of friends trying to split a pizza. The Shapley Value is a rule that says: "You only get credit for the slice you actually helped create, and that credit is shared fairly based on how many people were eating with you."

In this data model:

The Players: The data products.
The Team: The users.
The Rule: Every user contributes 1 unit of "value." If a user chooses 5 products, that 1 unit of value is split 5 ways. If they choose 1 product, they give the whole unit to that product.

This ensures that uniqueness is rewarded. If you make a product that is the only one of its kind for a specific need, you get a high score. If you make a product that is just a copy of something else, your score drops because users are splitting their attention between the two.

What This Means for Creators

The paper argues this system creates a healthy ecosystem:

It stops the "Me-Too" trap: If you make a copy of an existing popular tool, you won't get as much credit as the original because users will split their attention.
It saves the "Long Tail": Niche tools that only 50 people need, but need them desperately, will be valued highly because those 50 people likely won't be distracted by 50 other options.
It measures "Gross Data Value": The total value of all data in a company grows when you have more unique products and more people using them. It's like a garden: the more unique flowers you plant, and the more bees that visit them, the more beautiful the garden becomes.

The Bottom Line

This paper suggests that in a world where data is free to copy, attention is the currency. By measuring how selectively people choose a product, we can fairly value data without needing money, prices, or sales figures. It rewards the creators who make things that are truly distinct and necessary, rather than just popular or duplicated.

Technical Summary: Data Valuation Model for Non-Monetary Exchanges

1. Problem Statement

The paper addresses the critical challenge of valuing data products within intracompany exchange platforms where traditional monetary pricing mechanisms are absent. Existing valuation approaches face three primary limitations in this context:

Non-Rivalrous Nature of Data: Unlike physical goods, data can be consumed repeatedly at near-zero marginal cost. This "perfect elasticity" of supply renders traditional supply-demand equilibrium models ineffective for pricing.
Limitations of Bundling: While bundling is a common strategy to price digital goods, it obscures individual product value, making it difficult to attribute revenue or value fairly to specific producers. This lack of visibility discourages innovation and differentiation.
Flaws in Popularity-Based Metrics: Relying on aggregate access counts (views, downloads) fails to distinguish between high-value unique products and low-value duplicates. Furthermore, popularity metrics disproportionately favor "big-head" products, failing to incentivize the creation of "long-tail" products which, collectively, represent a significant portion of total economic value.

The authors argue that in non-monetary environments, valuation must be decoupled from cost, competitive pricing, and ROI, relying instead on a normative metric derived solely from user selection behavior.

2. Methodology

The proposed solution is a normative, choice-based metric that quantifies the value of a data offering based on the discriminative nature of consumer attention.

2.1 Core Concept: Attention as Value

The model posits that consumer attention is a finite resource. An offering's value is inversely proportional to the number of other offerings a consumer selects alongside it.

Discriminative Choice: If a consumer selects a single offering out of many available, that offering receives a high value score.
Indifferent Choice: If a consumer selects all available offerings, the marginal value contributed to any single offering is diluted.

2.2 Mathematical Formulation

The system is defined by a set of $n$ offerings $O$ and $m$ consumers $C$ . A subscription relation $sb: C \to \mathcal{P}(O)$ maps each consumer to the set of offerings they subscribe to.

The value score $v(o_i)$ for an offering $o_i$ is defined as the sum of the inverse of the subscription set sizes for all consumers who selected that offering:

$v(o_i) = \sum_{o_i \in sb(c_j)} \frac{1}{|sb(c_j)|}$

Where:

$sb(c_j)$ is the set of offerings subscribed to by consumer $c_j$ .
$|sb(c_j)|$ is the number of offerings in that set.

2.3 Cooperative Game Theoretic Foundation

The authors demonstrate that this metric is formally equivalent to the Shapley value in a cooperative game.

Players: The data offerings ( $o_i$ ) are treated as players.
Coalitions: Consumers ( $c_j$ ) represent the coalitions.
Value Function: Each consumer contributes one unit of value, which is shared equally among all offerings they subscribe to.
Interpretation: This formulation provides a principled, fairness-based allocation of value, ensuring that an offering's score reflects its marginal contribution to the collective utility of the consumers.

3. Key Contributions

A Normative Valuation Metric: Introduction of a formula that calculates data product value without requiring external market data, cost structures, or experimental preference elicitation.
Theoretical Grounding: The formalization of data valuation as a cooperative game with a closed-form Shapley value, ensuring the metric adheres to axioms of fairness and efficiency.
Incentive Alignment: The model is designed to correct the biases of popularity-based metrics by rewarding uniqueness and discriminative consumption.
Gross Data Product Value (GDaP): Definition of a system-wide metric ( $V = \sum v(o)$ ) that tracks the total value of the data ecosystem, showing how it responds to changes in production volume and consumption frequency.

4. Results and Observations

Through theoretical analysis and illustrative examples, the paper demonstrates the following behaviors of the metric:

Incentivizing Long-Tail Products: The metric successfully assigns higher relative value to "long-tail" products that are uniquely chosen by specific user groups, compared to "big-head" products that are popular but selected indiscriminately alongside many others.
Discouraging Duplication: The model penalizes redundancy. If a consumer subscribes to multiple identical or highly overlapping offerings, the value score for each is diluted. Conversely, unique offerings that capture exclusive attention receive higher scores.
Strategic Producer Behavior: The metric incentivizes producers to create products that fulfill unmet needs or offer distinct value propositions rather than simply increasing volume or mimicking popular products.
Growth Dynamics:
- The total Gross Data Product Value ( $V$ ) increases with both the number of offerings and the frequency of consumption.
- The paper suggests that in mature markets, the system will naturally exhibit logarithmic growth due to the saturation of consumer attention and organizational capacity, shifting focus from expansion to optimization.

5. Significance and Claims

The paper claims that this metric provides a principled and fair framework for measuring data product value in internal, non-monetary exchanges. Its primary significance lies in:

Decoupling Valuation from Pricing: It enables organizations to make insights-driven investment decisions regarding data products without relying on unavailable cost or competitive pricing data.
Fair Attribution: By utilizing the Shapley value, it offers a mathematically rigorous method for attributing value to individual producers based on their actual contribution to consumer utility, rather than arbitrary popularity counts.
Ecosystem Health: It promotes a diverse data ecosystem by validating the economic importance of niche ("long-tail") data products, ensuring that the total value of the organization's data assets is not underestimated.

The authors conclude that while the current model focuses on individual valuation, future work should explore bundling strategies and the quantification of product complementarity (e.g., how base products and complementary products interact to multiply value), leveraging the same game-theoretic framework.