If you need to understand modern Japan, look at how people make decisions. And if you want to know how Japanese consumers decide what to buy, where to eat, or where to work, chances are they are using Kakaku.com.

Founded in 1997, Kakaku.com operates some of Japan’s most popular sites. Its flagship price comparison platform attracts tens of millions of monthly users, while Tabelog, its restaurant review site, covers hundreds of thousands of eateries. It also runs leading platforms for job (KyujinBox) and apartment (Sumaity) searches.

However, even as one of Japan’s top tech players, Kakaku.com hit a wall in late 2023. The company was eager to scale AI across every part of the business—but speed, security, and scale were holding them back.

Challenges

• Fast wasn’t fast enough. Business teams had strong ideas, but with limited AI engineering resources, most PoCs took too long to move forward. Even validated concepts were delayed by development backlogs, security reviews, and infrastructure setup.

• Security felt out of control. Different teams were using different tools—ChatGPT, Claude, and others—with little central oversight. IT had to manage security risks, inconsistent data handling, and rising costs across multiple systems.

• Scaling AI meant scaling chaos. Without a unified platform, each new initiative required custom support. Business teams couldn’t build on their own, and the IT department couldn’t support demand at scale, which blocked the company’s vision of democratized AI adoption.

Facing these challenges, Kakaku.com needed a platform that could democratize AI application development while maintaining enterprise controls.

Solution by Dify Enterprise

After evaluating multiple options, they discovered Dify, an AI solution development platform that turns anyone into an AI builder through visual workflows and plugins. But at their scale, the community version wouldn’t cut it. They needed enterprise.

Self-Serve AI for Every Team

Instead of overloading central IT, Kakaku.com used Dify Enterprise’s multi-workspace architecture to give each team a secure, self-managed space, where different departments could build AI solutions grounded in their own data and domain expertise. With separate apps, users, and budgets, cost tracking was clear, experimentation easy, and access tightly scoped. They even set up a sandbox workspace to let new users start building right away with no approvals or delays.

Login Once, Secure Everywhere

Employees logged in Dify with their existing Azure AD credentials through SSO so there were no extra passwords to manage, and when someone left the company, their access disappeared automatically. Public link sharing was disabled across Dify, with Google Cloud Armor enforcing the rules at the network level. These controls gave IT confidence that sensitive data stayed protected without slowing anyone down.

Production-Ready Infrastructure

Running on GKE with Helm charts, Dify scaled automatically with demand and rolled out updates without downtime. The Admin API turned repetitive tasks into code: provisioning users, creating workspaces, and updating permissions became scripts that ran in seconds. Better yet, IT teams could plug Dify into their existing workflows and tools, and thus avoided the overhead of building new processes from scratch.

Central Control, Local Freedom

IT kept the keys to the critical pieces: infrastructure, model approvals, security policies. But each Dify workspace had its own administrator who handled the daily operations like adding users and managing permissions. This decentralized setup allowed teams to move at their own pace within the established guardrails. What’s more, regular audits of these administrators ensured compliance with internal policies, which prevented unapproved models or tools from slipping through.

Results

With Dify’s enterprise infrastructure in place, teams across Kakaku.com started building solutions that actually moved the needle. Dify gave them the tools; what they built with those tools redefined how the company operates.

Here’s what happened when employees got their hands on enterprise AI:

• 3-Hour Launch to Live System: The product data extractor powered by Dify went live in just three hours and now processes thousands of products daily.

• 30% Became Builders: Within a month, 30% of employees had signed up for a Dify account and built over 70 applications—not through mandates, but because this platform made practical sense.

• 950 Apps+ Built: Today, this employee-led innovation launched nearly 950 internal applications, with 75% of all employees registration of Dify and support from a strong, sustained user base.

Looking Forward

The impact was clear: time saved, speed gained, and shadow IT transformed into innovation in plain sight. But perhaps more importantly, the collaboration between Dify and Kakaku.com proved that enterprise AI success is not about hiring more AI engineers, but enabling anyone to build fast.

Dify is honored to have partnered with Kakaku.com on their successful AI adoption, and is excited to support their ongoing innovation.