You can't build AI in a walled garden

Over the last several years, I’ve had countless conversations with enterprise data teams building analytics platforms, operational reporting, and AI systems. The pattern is increasingly familiar: a company wants to unify customer data, build an AI-powered support assistant, or improve forecasting only to discover that replicating its own data into a centralized location reliably and in a performant manner is the hardest part of the project.

This challenge to centralizing data is a walled garden, where a vendor makes it difficult or expensive to use your own data outside the platform where it was created. And the restrictions that vendors typically utilize often show up in several ways:

• API access only available on premium plans or paid add-ons
• Per-call or credit-based pricing models that are prohibitively expensive for data replication
• Rate limits that work for automations but make large-scale extraction impractical
• Requiring certification with fees for any applications that connect to APIs
• Terms of service that restrict exports, persistent copies, indexing, or AI use cases
• Requirements to translate data out of proprietary formats in order to use outside of your platform 

Individually, these restrictions may seem reasonable. However, together across multiple enterprise applications, they mean customers technically own their data, but vendors increasingly control whether they can afford to use it in the customer’s data stack.

[CTA_MODULE]

The evidence is everywhere

The idea of a walled garden isn't theoretical. Across the software industry, SaaS vendors are increasingly introducing technical, financial, and contractual controls that determine how customers can access and use their own data. 

Consider a few examples:

• SAP updated Note 3255746 in 2024 to restrict third-party use of the ODP-RFC API, an interface that supported an entire ecosystem of SAP extraction tools. Existing approaches that customers had used for years became non-compliant, and future support increasingly routes through SAP-priced alternatives.
• Slack updated its API terms in 2025 to prohibit bulk exports via API, persistent copies/archives/indexes, and the use of Slack data for training LLMs. It also introduced new rate limits that significantly constrain large-scale extraction and indexing workloads.
• Workday is moving toward a consumption-based Flex Credits model, where API usage consumes credits and throughput remains tightly controlled. Large-scale extractions can become both slow and expensive.
• NetSuite charges separately for analytics-grade access through SuiteAnalytics Connect and governs throughput through service tiers and concurrency limits.
• Salesforce continues to preference its own offerings and gate API access and usage through editions, limits, and paid capacity expansions while simultaneously increasing costs for ecosystem partners that help customers move and replicate Salesforce data.
• Rippling requires customers to purchase extra products to access certain integrations and blocks access to specific data sets entirely.  With the launch of Rippling Data Cloud, Rippling introduces another attempt to keep data inside its systems. 
• Xero has tiered monthly API fees and explicit egress allotments with overage pricing. 

I’ve seen the exact details of these restrictions differ, but the outcome is the same: accessing data increasingly requires navigating layers of pricing, throughput limits, permissions, policies, and vendor-controlled alternatives. And in many cases, the technical challenge isn't building the solution; the challenge is gaining reliable access to the data in the first place.

Why AI changes the stakes

These restrictions are arriving at exactly the moment companies need broader and higher performance access to data than ever before. SaaS providers are adopting this defensive posture because they see their core business models under threat from AI-native upstarts and customers increasingly DIYing alternatives, forcing them to aggressively lock down customer data to maintain/grow their market share.

A decade ago, most organizations could work around data silos. An analyst could export a report from Salesforce, pull data from SAP, combine it in a spreadsheet, and eventually answer the business question. It wasn't elegant, but it was manageable.

AI changes that equation entirely.

The organizations I speak with aren't trying to build AI on top of a single application. They're trying to build agents that combine data from CRM, support interactions, billing data, product usage, their codebase, and much more. They're building forecasting models that need financial, operational, and customer data. They're building agents that need to reason across the entire business, not just one system.

The more valuable the AI use case, the more systems it typically needs to access (and to do so more frequently). That's why I believe walled gardens have become one of the biggest obstacles to enterprise AI adoption and value creation. They don't just make data movement harder — they make it harder for organizations to assemble the complete picture that AI requires.

Why Open Data Infrastructure should be part of every software buying decision

One thing I've learned from watching customers navigate these challenges is that you can't solve a walled garden by expecting your vendors to do the right thing for you. Every software purchase is also a data decision, and vendor policies are not static. API policies will change. Pricing models will change. Rate limits will change. Every few years, the industry finds a new way to put additional controls around data access. That's why I believe the long-term answer is having open data access as a contractual requirement and picking tools that promote Open Data Infrastructure.  

That fundamentally changes the relationship between organizations and the software they depend on. Instead of asking vendors for permission every time a new AI use case emerges, companies need to build this into their purchase agreements and into their architecture. 

From my perspective, that's ultimately what this discussion is about. The companies that move fastest will be the ones that do not have to renegotiate data access every time the business wants to build, automate, or analyze something new. ODI gives customers the control they need to keep innovating on their terms.

[CTA_MODULE]