Estimated reading time: 7 minutes

Workato CTO Adam Seligman explains why internal AI initiatives must tie directly to real business outcomes, and how deeper integration via MCP was their big game-changer.

In late 2024, low-code app builder Workato rolled out Claude for Business to all of its 1,300 internal employees. At first, adoption was modest: just a few hundred or so employees used it for sporadic tasks, like summarizing emails or analyzing data.

For a while, their use remained at about 150-200 chats a day company-wide. This improved some workflows, but its impact was limited without deeper integration into business systems.

It wasn’t until they introduced Model Context Protocol (MCP) servers into the mix, in October 2025, that Claude usage skyrocketed. With read and write access to business apps like Snowflake, Salesforce, Gmail, calendars, and others, Claude became a near-overnight sensation.

“We went from 200 employees to every employee using Claude almost every day,” says Adam Seligman, Workato’s CTO. “We saw a 700% increase in usage over a 60-day period.” 

“We were caught by surprise,” he adds. “We knew that Claude was incredibly capable and we knew that MCP would unlock it, but the creativity of our employees just stunned us.” Workato employees began using MCP-enabled Claude to surface high-value business context, such as pulling together dashboards to identify customers at risk of churn.

Then they turned on write functions with MCP servers, opening up a new class of automation. A Workato support engineer built a workflow to automatically respond to customer tickets with human-in-the-loop approval, while a sales operator mined customer stories to guide feature development. From sales call preparation to financial analysis, MCP expanded how teams accessed and acted on business context.

The thing is, all this enthusiasm comes at a cost. “Our top engineers are using around two billion tokens a month,” says Seligman. With a market dominated by hype and vendor noise, how do you keep a level head and justify such massive AI use? 

For Seligman, it was a case of pivoting the conversation to real business outcomes. “We have to get out of the tech mumbo jumbo and vendor news and just get to the business,” says Seligman. “What can AI do to the business that will get value?” 

For him, this means focusing on tangible results, making AI as actionable on business processes as possible, and continuing to share internal stories and knowledge. In this way, measuring the success of AI really becomes a measurement of the success of the business at large.

Ground AI usage in KPIs

In the past couple of years, most organizations have raced ahead with spending on internal AI tools. Often, this was done hastily, without setting baseline metrics or key performance indicators (KPIs) to measure the before and after effects of introducing AI.

“Now, organizations are doing their 2026 budget, and they’re not always seeing results in their business,” says Seligman. “If your AI projects aren’t designed to have an influence on KPIs, they might not be material.”

Tracking metrics is now imperative to avoid falling into the grave of AI pilots that failed to deliver a return on investment. “My advice is to look at the metrics and think of how you report to your board,” says Seligman.

For instance, how quickly are you finding new customers and onboarding them? How fast are you building new projects? How are you reducing operational costs? These are the real metrics that investments in AI initiatives should be designed to influence from the outset. “Your business lives and dies with these KPIs,” adds Seligman.

Give AI to the entire business

At Workato, AI has moved beyond the realm of software engineering. Employees use Claude to plan marketing events, aid targeted outreach for sales journeys, and prepare quarterly business reviews (QBRs).

From a leadership perspective, guiding the right culture is still a critical element of successful adoption. At Workato, this has occurred naturally and has been encouraged by top-down support. Their CIO Carter Busse often demos experimentation with AI tools, and similarly encourages others to share screencasts of how they’re using it in practice.

All you can eat business data

Measuring KPIs and sharing AI with the entire company is all well and good, but that doesn’t fill an important void – access to relevant external data and systems that business users actually use.

AI tools can hallucinate, and without access to real business data and read/write access to external platforms, their functionality is quite limited. “Businesses need ways of applying AI in their core processes, but doing it safely and reliably,” says Seligman. 

The recent growth of internal AI usage at Workato: per-day Claude chats rose sharply once MCP was introduced in October 2025.

For instance, an engineer could ask Claude to update a Jira ticket based on a sales call in Gong, a sales platform that Workato customer representatives use. In some cases, external platforms, like Workday and Salesforce, don’t yet have MCP servers for Workato’s needs, which has forced engineers to build their own MCP servers.

Seligman points to Workato’s outcomes such as business growth, increased lead and pipeline generation, faster responses to opportunities, and even operational expansion – including new data centers – as leading indicators of success. “We’re kind of in a growth whirlwind right now,” he adds. 

Share context and knowledge 

Lastly, AI models and practices are constantly changing, and it takes continuous effort to refine a successful corporate AI strategy, from engineering to sales teams.

“We’re moving faster than ever before,” says Seligman. AI in software development has progressed significantly in short order, from in-line code generation, to using IDEs with assistance, and fully autonomous agents. “Now, engineers are devising goals and plans with agents and then letting them work for hours uninterrupted,” he says.

Workato performs a lot of exploratory work: they use AI agents to test options, such as for new technical features, before committing to a direction. They also use AI agents to generate in-line commenting so that engineers retain knowledge of the code they’re creating.

All this monthly token consumption can rack up quite a bill, but Workato does set guardrails and techniques to optimize usage. For instance, they’ve been improving AI outcomes by feeding the LLM engineering context, helping an LLM engage with the nuances of Workato’s codebase and platform semantics. Seligman shares that they’ve done this with reusable “skills,” which are custom capabilities that help Claude operate with domain-specific knowledge.

One such skill involved teaching Claude how Workato’s “datapills” work – a key part of their platform’s low-code field-mapping system backed by complex JavaScript. “We had a learning curve of a week or two to understand this internal thing,” says Seligman.

The upside down year

We’re at an inflection point in how software engineers use AI tools like Claude Code, GitHub Copilot, or Google’s Jules to increase coding productivity. However, it’s not always clear how changes in lines of code equate to end business results.

“We have to make good use of AI so that it’s easy to operate, and doesn’t add more debt than it’s worth,” says Seligman. “That’s a big step, but we’re now seeing that in other functions and business processes.”

According to Seligman, getting access to corporate data and processes in a secure way through enterprise-grade MCP servers was the real turning point at Workato for reaping clear value from AI tools. You get faster access to business applications and governance at the same time, he says. 

This newfound ability has the power to reset the conversation. “2026 is going to feel like an upside down year,” says Seligman. CIOs used to need to constantly play a balancing act between speed versus control, but recent advances in safe enterprise MCP use flips the script, he says. 

Now, the focus is on using AI to deliver business value. Perhaps it’s finding concrete areas where AI can be used to reap rewards, like improving customer satisfaction, or reducing operating costs. “Anchor on those things, look at the processes across the company that’ll shape them and then use technology there.”

And even if it means high AI bills, for executives like Seligman, it’s worth it to continue evolving and delivering new customer experiences. “Look, I’m happy to spend the money to keep up with customer demands.”