We are based in Osaka. That's not incidental to what we build or why we build it. The Kansai business environment — its industrial character, its relationship with information technology, its particular version of organizational caution around data — shapes the problems we encounter and the solutions we think are worth building.
This post isn't market analysis. I'm not going to claim precise statistics about AI adoption rates in the Kansai region. What I can offer is observation from working directly in this environment: the conversations we've had, the objections we hear, and the patterns that emerge when you spend time with IT teams and operations leads in Osaka's manufacturing and logistics base.
Osaka Is Not a Startup City
The dominant narrative in Japanese AI coverage centers on Tokyo: the startups in Shibuya and Minato-ku, the venture capital clustering around Marunouchi, the government digital transformation initiatives aimed at public services. Tokyo is where the AI news happens. That's where the conferences are, where the investment announcements land, where the press coverage is.
Osaka's economic character is different. The Kansai region, and Osaka specifically, is Japan's second-largest economic center, with a GDP equivalent roughly comparable to a mid-size European country. Its economic base is substantively industrial in a way Tokyo's isn't: pharmaceuticals (Takeda, Shionogi — founded here, many operations still here), chemicals (Daikin, Sumitomo Chemical), food and beverage, automotive parts suppliers in the wider Kansai-Tokai corridor, and an enormous logistics infrastructure tied to Kansai International Airport and the Port of Osaka.
These are not technology-native industries. They are industries where the dominant concern is operational continuity, where IT systems have been running for 10 to 20 years and no one has a compelling reason to replace them, and where the phrase "AI adoption" triggers questions about risk and data security before it triggers questions about capability or efficiency.
When we started building dodoAI in 2024, we made a deliberate choice to position in this environment rather than in Tokyo's startup-adjacent enterprise segment. The problems here are harder and less fashionable. But they're also real, and the companies that need solutions are not being well-served by products designed for cloud-native, US-headquartered enterprise IT teams.
Why Manufacturing and Logistics Move First
Across the conversations we've had, manufacturing and logistics organizations are meaningfully ahead of, say, financial services or healthcare in evaluating AI for process automation. This seems counterintuitive — finance and healthcare have more obvious high-value decisions to automate. But there are structural reasons why manufacturing leads.
The first is process legibility. Manufacturing operations run on defined workflows. Procurement follows purchase order procedures. Supplier qualification follows documented criteria. Receiving inspection follows quality checklists. These processes exist as written policy, and the decisions within them are bounded: a PO either meets the approval criteria or it doesn't. That clarity makes the agent configuration work tractable. You can map the policy into logic rules with reasonable confidence that you've covered the main cases.
Finance and healthcare processes often have more judgment-intensive steps where the "rule" is genuinely ambiguous or where exceptions are common. Automating those is possible, but it requires a more sophisticated approach to exception handling and escalation. Manufacturing approval processes, while still complex, tend to have fewer genuine gray areas.
The second reason is competitive pressure from supply chain efficiency. Osaka-area manufacturers that supply into larger automotive or electronics assembly chains face upstream pressure on lead times and procurement cycle times. If a Tier 1 assembler in Aichi requires purchase order confirmation within 24 hours but a supplier's internal approval process takes three days, that creates a real operational problem. Process automation has a clear value proposition in that context — not as a technology investment, but as a supply chain competitiveness requirement.
Logistics organizations face similar timing pressure: inbound customs documentation, carrier rate approvals, freight claim processing — all of these have external SLAs that internal approval processes need to meet. Automation reduces the variance in cycle time, which directly reduces missed SLAs.
The Data Sovereignty Conversation in Kansai Industry
In every substantive conversation we have with Osaka-area enterprises, data sovereignty comes up before capability. This is not the same conversation that happens at Tokyo-based startups evaluating AI tools, where the concern is more about compliance checkbox than genuine data protection. The concern in Kansai manufacturing is substantive and grounded in specific risks.
The dominant concern is procurement data. Purchasing prices, supplier terms, quality agreement details, sole-source relationships — this is genuinely competitive information. For a mid-size manufacturer that has spent years developing supplier relationships and negotiating favorable terms, having that data processed through a shared cloud AI service represents a real risk. It's not hypothetical. The question "where does our data go when the AI processes it" has a definite answer with cloud services, and enterprises here don't like that answer.
A secondary concern is supply chain structure. For manufacturers with complex multi-tier supply chains — which describes most of the Kansai manufacturing base — the structure of the supply network is itself competitive intelligence. Knowing which suppliers a company uses, at what volumes and frequencies, reveals production capacity and strategic sourcing decisions. They don't want that information leaving their perimeter.
This is why on-premises deployment is not a "nice to have" for our customers — it's the baseline requirement. We do not have an alternative commercial conversation. If a prospect asks whether we have a cloud option, the answer is no, and that's by design. The customers who call us are the ones for whom that constraint is non-negotiable.
Organizational Decision Patterns: IT and Operations Alignment
Japanese enterprise IT has a specific organizational dynamic that's worth describing because it affects how AI adoption actually happens. In many Osaka-based manufacturers, the IT function is conservative, understaffed relative to operational complexity, and focused primarily on maintaining existing systems. There is no "innovation team" or "digital transformation office" with dedicated budget and authority to adopt new tools. The CIO, if there is one, is also typically overseeing a shared services function that includes HR systems, ERP maintenance, and internal helpdesk.
AI adoption proposals that require the IT team to deploy and operate new infrastructure meet significant resistance — not because the IT team is opposed to automation, but because they don't have the capacity to manage another system. This is one reason we spent considerable effort making dodoAI deployable by a general-purpose IT team rather than an AI specialist. The deployment target is RHEL or comparable Linux on existing bare-metal or VMware infrastructure, using container-based packaging that the IT team already knows how to operate. We didn't want our deployment to require hiring an ML engineer.
The effective entry point for AI automation in these organizations tends to come from the operations side — procurement managers, logistics coordinators, finance team leads — who have a specific problem they want to solve and who have enough organizational standing to sponsor a pilot. The IT team's role in those cases is less about initiative and more about security review and infrastructure provisioning. We've learned to engage both sides early: the operations lead who defines the problem and the IT lead who approves the deployment architecture.
What's Not Working Yet
Not everything is moving quickly. A few areas where Osaka enterprise AI adoption remains genuinely stuck:
HR and personnel data automation is almost universally off the table. Japanese employment law is complex and the penalties for procedural error in personnel decisions are significant. No enterprise IT team we've spoken with is ready to put autonomous agent decision-making anywhere near HR workflows, regardless of how well the technology works. That's probably the right call for now.
Multi-company data sharing is complicated by the Keiretsu legacy. Some Kansai manufacturers operate within supply chain networks where there's significant cross-ownership and longstanding inter-company process integration. These aren't formal Keiretsu in the classic postwar sense, but the relationships are close enough that data flows between companies in ways that create APPI complexity. Automating processes that touch cross-company data requires legal review that slows adoption significantly.
The skills gap is real. Not in AI — the question isn't whether the companies understand AI. It's that the internal process documentation quality needed to configure rule-based agents is often poor. Procurement approval policies exist as implicit knowledge distributed across several people's heads, not as written documents that can be translated into logic. The rule mapping work we do at the start of every engagement is as much an organizational process documentation exercise as it is a technical configuration task. That work takes time, and not every organization is ready to invest in it.
Building here isn't glamorous in the way that building AI products in Tokyo or San Francisco is. The enterprise cycles are longer, the procurement processes are slower, and the organizations we work with are not going to announce a partnership or write a press release about a pilot deployment. But the problems are real, the value created when automation works correctly is measurable, and the organizations that succeed here tend to be the ones that are serious about building durable operational capability rather than chasing the next technology trend.