How Data Science Is Rewriting Chargeback Strategy

A few years ago, improving chargeback performance meant adding more reviews and more people to the team and having a clear chargeback strategy was unheard of. 

Today, technology and modern solutions make chargeback management easier and importantly, more effective. Merchants reduce manual effort while improving dispute outcomes at the same time. The complexity hasn’t disappeared, it has moved into systems that learn, adapt, and optimize behind the scenes.

Chargeback Context

Across e-commerce, gaming, and SaaS, dispute handling has quietly evolved. It takes approximately 40 minutes for an experienced  dispute manager to review the case, gather all relevant evidence and formulate the dispute. What once was impossible to maintain without a large investment of team effort, now with automation, this process becomes easier, and at the same time, more precise and more productive.

Because the reality is this: a dispute file can easily reach 20 pages. The challenge is not just collecting data. The challenge is making sure the argument is clear, relevant, and convincing for the reviewer on the other side.

The Data Layer: From Structured to Combined Intelligence

For years, dispute systems relied almost entirely on structured data. In the context of chargeback orchestration, structured data refers to clearly defined, machine-readable fields such as transaction timestamp, amount, payment method, billing and shipping details, and delivery confirmation. This data is easy to process and has traditionally formed the backbone of dispute responses.

Unstructured data includes information that does not follow a fixed format, such as customer emails, chat transcripts, support tickets, and issuer communication. Historically, this data was reviewed manually but not systematically used in dispute automation.

As Roenen Ben-Ami, Co-Founder and Chief Risk Officer at Justt.ai, explains, the real shift is happening now, as AI enables these two data types to be combined within a single system. Instead of operating in parallel, structured and unstructured data can now be analyzed together and incorporated directly into dispute arguments.

This changes the quality of the output.

A structured-only argument might state: the transaction was completed and the item was delivered.

When combined with unstructured data, the argument can add: the customer confirmed receipt or continued using the service after delivery.

The difference is small in length, but significant in impact.

According to Ben-Ami, enabling this kind of integration requires systems that can process and connect both data types in real time. This is where he sees a clear gap between traditional tools and platforms like Justt, which were built to unify structured and unstructured data as part of a single decisioning process.

From Data to Argument: Continuous Learning at Scale

A dispute is not won by data alone. It is won by how the strategic presentation of data. That presentation can be optimized using chargeback strategy based on data science.

As Ben-Ami explains, the same underlying data can produce very different arguments depending on:

• ordering of evidence
• emphasis on specific signals
• framing based on reason code or issuer

Now compare two ways of presenting the exact same data.

In one version, the evidence follows a standard template. The delivery confirmation is buried deep in the file, login activity appears as raw logs, and while everything is technically present, it is difficult to follow.

In another version, the same data is structured with clear intent. It opens with a direct delivery confirmation, immediately supported by carrier data, followed by a short note highlighting post-delivery customer activity.

The data is identical, but the outcome can be very different.

According to Ben-Ami, the real differentiation comes from treating dispute generation as a system that continuously learns. Instead of relying on fixed templates, their platform runs large-scale experimentation across dispute variations. They test how structure, sequencing, and narrative affect outcomes across issuers and reason codes.

This creates a feedback loop where every dispute outcome improves the next one.

Ben-Ami explains that building this type of system requires more than adding AI components. It depends on an architecture where data, argumentation, and feedback loops are fully integrated.

What These Systems Actually Optimize

Once data and argumentation are part of a continuous learning loop, the next question is what the system is optimizing for.

From a data science perspective, everything starts with the target function. In modern systems, this target is not fixed. It evolves over time.

Business priorities vary. Some merchants optimize for net recovery after fees. Others focus on maintaining chargeback ratios below network thresholds. In subscription or gaming environments, protecting high-value users becomes part of the objective.

Systems balance these objectives dynamically across segments and over time. They adjust not only decisions, but also the underlying optimization logic. Again, what matters are the overall chargeback strategy.

At the same time, the external definition of performance is changing. As Tarun Singh from Disputed.ai notes, card networks like Visa and Mastercard are expanding beyond traditional chargeback ratios. They now incorporate early risk signals and metrics such as VAMP ratio into how merchant risk is assessed.

This adds another layer of complexity. Systems must adapt not only to internal business goals, but also to evolving network expectations.

Actionable Takeaways for Chargeback Strategy

Reduce manual effort
Manual dispute handling does not scale. The objective of automating chargebacks and using data science to optimize chargeback strategy is to minimize human dependency while maintaining strong outcomes.

Make data usable
Structured and unstructured data should be consistently captured, stored, and readily accessible to fully benefit from modern platforms.

Adapt to changing requirements
Issuer expectations, reason codes, and network metrics evolve. Static processes quickly become ineffective.

Perspective on Chargeback Strategy

Dispute management is evolving into a system that blends data, experimentation, and domain expertise. To drive the next level of performance, the next phase will be defined by how effectively this system learns and adapts.