Would you expect an employee who can’t access company information to transform your business?

Imagine this: You hire a new employee with impressive credentials. On their first day, you tell them they can’t attend any meetings. They’re not allowed to talk to other departments. They can’t access your product catalog or financial data. And they’re completely in the dark about your company’s history and how things operate.

Sounds ridiculous, right? Yet this is exactly what we’re doing with AI when we trap it inside individual SaaS applications.

The average enterprise runs an astonishing 367 SaaS apps. Given the fast rate at which SaaS vendors are sprouting bespoke AI agents, the average enterprise will soon have 376 instances of AI as well.

None of those agents will transform your business. The LLM’s replies are only as good as the data it retrieves. It won’t give contextual, deep answers unless it can access interoperable, accurate data from diverse sources. Only then will AI agents go from additive—helping engineers vibe code, assisting marketers in generating blog content—to transformational.

A SaaS-y predicament

After years of raising prices to stay afloat, SaaS vendors are hungry to convert AI agent use into revenue. Salesforce Agentforce, for example, charges $2/conversation. Until SaaS AI agents can provide more value, however, my guess is that most IT departments will balk at the operating costs. Those two-dollar conversations quickly add up when compared to the $20/month charge of ChatGPT or Claude, or DeepSeek’s $1.10 per million chat tokens.

Also, pre-defined, SaaS-specific API calls only let an agent do so much. It’s like hiring a new employee who doesn’t know what products you sell, doesn’t attend meetings, and has no idea how the company operates, but is really good at telling you sales metrics. You wouldn’t expect that person to perform well. Yet we put AI into the same situation when we silo it within one app or function. 

An agent is only as good as the data retrieves

Business transformation is a function of profitability, which itself stems from ROI. You get ROI when you create new revenue or cut costs. While AI agents do make some human jobs more efficient, companies ultimately must look at the infrastructure behind AI agents to change how the business generates profit. The answer lies in auditing SaaS implementations, unifying data into a centralized place, adding a knowledge graph for context, and only then building agents.

It pays to re-architect the data stack. Without interoperability, data remains siloed, leading to superficial insights that may appear impressive but are fundamentally flawed upon closer examination. A proof of concept will look good until it becomes clear that the facts are totally wrong. Or the AI will answer accurately for two questions, then stop being useful. Worst case, poor data access results in AI deployments that fail to deliver on their promises or even violate ethics. 

How to stay relevant in an AI-native world

There’s still more fear than excitement about AI in business. Even if it’s just a starting point, having a strategy helps restore confidence. Don’t fixate on SaaS vendors’ bespoke AI agents. They are only designed to help the point solutions with which they’re paired. Instead, design an interoperable, unified data solution that includes context. Then, when you piggyback AI agents on that contextual, broad data, you will be able to not just make your business more efficient, but transform it to keep up with our AI-first era.