Register Now
HomeCustomers
Case study

How Mercari Rebuilt Its DLP Program with Data Lineage

Overview

Mercari operates Japan’s largest consumer-to-consumer marketplace alongside fintech businesses with strict compliance requirements around data loss prevention. Each business holds critical customer information that needs strong security controls to protect it, and both verticals have their own regulatory challenges on the kind of controls needed. Mercari sought a solution that could provide both solid controls and meet strict regulatory requirements while maintaining flexibility and productivity in the environment.

Industry

E-commerce marketplace and fintech

Company Size

2,100+ employees across its global footprint

Mercari operates Japan’s largest consumer-to-consumer marketplace alongside fintech businesses with separate regulatory requirements, each holding customer information that needs protection.

Before introducing Cyberhaven, the Enterprise Security team had limited visibility into where files were moving. Greater visibility was the starting point. But greater insight from DLP would inevitably increase the number of alerts that need review. Mercari needed a solution that could both follow the data and keep monitoring operations effective and efficient.

“We wanted to implement a solution with low friction on employees that could provide us with deep insights on data flows, and the ability to set meaningful and effective policies. We found that in Cyberhaven.”

Jason Fernandes

VP of Security and Privacy, Mercari

At a glance

Four Requirements, Four Outcomes

Requirement

Result

Data origin

The team needed better visibility into where files were moving. Detection coverage alone could not tell them whether a given file movement was risky, so they needed a way to handle data differently based on where it originated.

→

Datasets defined by origin and policies by destination, so legitimate business workflows and genuine exfiltration can be told apart.

Accounts

Being able to provide strong controls while also maintaining some flexibility in the environment, such as the ability to distinguish corporate Google Workspace use from personal accounts on the same service, was also a requirement.

→

Corporate and personal accounts are distinguished within the same service, such as Google Workspace, allowing personal use to be restricted without blocking a service the business needs.

Shadow AI

Rapid generative AI adoption needed an equally scalable solution that could support a more comprehensive inventory of which AI tools were in use, and from which accounts they were being accessed.

→

Unapproved AI tool usage made visible, with warnings and blocking applied by risk tier.

Operations

Deploying DLP inevitably increases the number of alerts, so deploying a solution that could be tuned and operated easily was a requirement.

→

The volume of detections brought within a level the team can operate day to day.

Data lineage

Following Where the Data Came From

Problem

Comparing DLP products tends to become a contest over which data types can be detected. What Mercari needed was not only coverage, but the ability to follow where the data came from. Two files with identical content are not the same thing if one came out of an HR system and the other was written by the employee, yet on content alone they look the same.

How Cyberhaven solved it

Cyberhaven’s data lineage preserves a file’s origin as it moves. Mercari used this to define datasets by source rather than content (HR systems, support tools, the analytics environment), so sensitivity follows origin and content together, not a keyword. Region-specific detection rules are provided and tuned, but rules alone have limits; origin is the second axis that carries the judgment.

Same file, two ways of looking at it
Content only
No origin context
budget.xlsx
Personal cloud
Missed
With lineage
HR system
budget.xlsx
Personal cloud
Incident

An HR export, copied into a new file, renamed, headers removed, and values edited. Content rules have nothing reliable left to match. Lineage follows the copy and paste, not the content, so it still knows where the data came from. Illustrative example, not a specific Mercari incident.

“You can keep chasing coverage of data types and rulesets but you hit a ceiling on this kind of detection alone. Knowing where the data comes from provides important context that allows you to gain insight into whether or not the flow of data constitutes an incident.”

Hiroyuki Haruki

Enterprise Security Team, Mercari

See how data lineage works→

Corporate vs personal accounts

Keeping Approved Services Available While Separating Personal Accounts

Problem

Many services employees use are company-sanctioned, such as Google Workspace. The same service can often be reached from a corporate account or a personal one, and platform-side controls for telling those apart are limited. Separating the two while keeping the environment flexible and productive was a requirement: without that distinction the review scope widens to every use case, which is not realistic for any team.

How Cyberhaven solved it

Cyberhaven identifies the account that data moves into, distinguishing corporate accounts from personal ones and internal domains from external ones within the same service. That let Mercari draw the line at the account rather than the service: what the business needs stays available, while warnings and blocking apply only to transfers into accounts and domains unrelated to work.

One service, two accounts
Google Workspace
Corporate accountinternal domain
Allowed
Personal accountexternal domain
Warn / block

“Inside the same service we can separate the corporate account from the personal one, which allows us to set stricter policies while still allowing flexibility for certain cases, as well as reduce the burden of monitoring operations.”

Hiroyuki Haruki

Enterprise Security Team, Mercari

Explore modern DLP→

Shadow AI

Turning Shadow AI From a Blind Spot Into a Managed Risk

Problem

Mercari is moving toward AI-native workflows, and employees are adopting generative AI at a rapid pace. An approval route existed and the list of approved tools was published, but there was a need to go beyond manual processes and implement automated shadow AI detection to cover blind spots.

How Cyberhaven solved it

Mercari evaluated several candidates and chose Cyberhaven for discovery accuracy, administrative usability, and a light touch on development environments. Its Agentic Data Security product finds the AI tools in use and separates corporate AI accounts from personal ones (a corporate Gemini tenant versus a personal login, for example), then applies policy by risk tier, blocking the highest risk and warning at the point of use for the rest. Again, what matters is which data moved, not which site was visited.

AI tool inventory, controlled by risk tier
ToolAccountControl
GeminiCorporate tenantSanctioned
GeminiPersonal loginWarn
Unreviewed AI toolPersonal loginBlock

“We used to take a more manual approach to managing which AI tools were in use. Now there is a list, and every tool on it has a control attached. Because the risk is visible, approving a tool is a much easier decision.”

Hiroyuki Haruki

Enterprise Security Team, Mercari

Explore secure AI adoption→

Results

The Results and Looking Forward

Overall, the way Mercari’s Enterprise Security team works has changed. Data is now judged with its origin taken into account, not by content alone. Controls land at the account rather than the service, so approved tools stay available while personal use is restricted, and shadow AI is visible enough that approving a new tool is a straightforward decision.

Deeper visibility did bring more to review, but tuning keeps it within a realistic operating load.

Cyberhaven continues to strengthen its capabilities across DLP, DSPM, Insider Risk Management, and Agentic Data Security, and Mercari is looking to broaden its own coverage in step with that. The rollout of cloud sensors across the collaboration and mail platforms is part of it and is in progress.

At the same time, the operational side has to mature: reducing false positives, revisiting exclusions, and automating. Each round of review feeds back into which datasets, thresholds, and destinations warrant a warning, a block, or nothing at all, so the queue becomes sharper rather than larger. Alongside that, the team continues to push AI governance further.

“Leveraging both Cyberhaven’s endpoint and browser sensors has improved our ability to manage DLP risks without relying heavily on network-based defenses.”
Yu Sasaki
Engineering Manager, Enterprise Security, Mercari