OpenText’s Executive Vice President, Chief Digital Officer, and Chief Information Officer Shannon Bell believes one of the biggest reasons AI pilots fail is that organisations try to automate their most complex workflows too early.
“If you think of an agentic AI deployment, you can’t start with your most complicated business process,” she said during a media conference.
Instead, Bell argued that organisations need to begin with narrowly scoped use cases, understand how those individual agents interact within workflows, and gradually build towards more complex process automation.
Disconnect
Previously, OpenText Chief Product Officer Muhi Majzoub spoke about how AI is creating new jobs instead of eradicating them. Across organisations in Singapore, Bell observed strong interest in AI at the board level. The problem, however, is a disconnect between expected outcomes and actual implementation.
“What I’m hearing from some of the IT leaders I’ve spoken with is that they’re certainly feeling the pressure from their leadership, and they’re starting to move towards agentic AI, but there’s concern about how quickly they’ll see the results. Therefore, there’s a little bit of a mismatch in terms of expectations of outcomes and results, and the length of time that it takes,” she said.
Because organisations do not have the time to complete a full data overhaul before transforming business processes, AI efforts need to focus on specific use cases, Bell noted.
“No organisation today is going to wait for a big bang business process transformation,” she said.
During the initial rush towards AI adoption, Bell said many organisations simply rolled out the technology internally and allowed teams to experiment rapidly.
“You can spend a lot of money on AI pilots and experimentation and not get outcomes. I think the organisations that moved a little slower had an advantage, because in many of those cases, the data wasn’t in the right shape. They didn’t know the outcome they were chasing, or they had a perception of the outcome that wasn’t grounded in reality, and in many cases, tried to do too much,” she observed.
Customer zero
For Bell, becoming “customer zero” allowed OpenText to better understand customers’ challenges, and that approach has continued with the arrival of agentic AI.
“I joined the company two and a half years ago, and the company was always using some of its own technology, but en masse. So we put in place a program to say that every time engineering releases software, we’re going to take it, deploy it for all of our internal users, and then we’re going to provide that closed loop feedback to engineering to drive a better product with AI,” she said.
Realising it needed to accelerate internal AI adoption, OpenText built a no-code, open-model AI platform called Aviator Studio.
“We needed the flexibility to build agentic capabilities and drive adoption ourselves on top of the product, and then see which use cases delivered value and which didn’t,” Bell said.
On the agentic AI side, OpenText relied less on pre-built use cases developed by engineering teams and more on Aviator Studio’s capabilities to experiment and iterate quickly.
“If I ask engineering to build me use cases, I better be pretty sure they’re going to be successful, because you’re not going to want to just turn them off and say, ‘Yeah, this doesn’t work.’ But if you’re driving that innovation with the Studio, it’s much easier to turn things on and off,” Bell explained.
Bell said 30% of her time is spent speaking with customers, sharing use cases, and reviewing business processes. Based on those conversations, she said every CIO is under pressure from management and boards to roll out AI and deliver measurable outcomes, but there’s no magic answer for all.
“The most value I can bring today is sharing our customer zero use cases, offering those to our customers to understand, to see the value, to look at the metrics and the results that we’re delivering to understand where we failed,” she said.
Bell also discussed how OpenText is building what it calls an “agentic genome,” beginning with mapping the organisation in detail.
“We went function by function, job by job, 1,700 different job titles across 21,000-plus employees, and we looked at each and every role. We looked at what the human in the role should do, and what the AI agent could do in this situation,” she said.
“The company then mapped those roles against business processes across OpenText, including hire-to-retire and procure-to-pay processes.”
“We mapped roles, agentic and human capabilities, and our business processes. That was how we started to identify which processes we wanted to target for agentic AI implementation,” Bell said.
Promising use cases
For agentic AI, Bell said she is seeing significant potential in automating L1 help desk functions. According to her, the repetitive nature of calls, issues, and incidents makes the area well suited for automation.
Bell also observed a 25% to 30% productivity gain when developers were equipped with agentic AI and generative AI capabilities for code testing.
“The more complicated ones are when you get into agentic AI for finance. Think of the procure-to-pay lifecycle and what it takes to deliver that. You’re dealing with a complex workflow that touches multiple groups, and you need a mix of human and agentic interactions throughout the process. That’s where people get stuck. If you try to deliver the entire workflow at once, it’s not a fast path to success. You need to build the building blocks first, then orchestrate the flows,” she explained.
Ultimately, Bell said the key to ensuring AI pilots make it into production is investing in data curation and treating even small AI use cases as foundational building blocks.
“We understood our business processes well. We didn’t fix them all or transform them all, but we understood them, and we started very small. Some of the use cases we built agents for seemed so simple that you would wonder whether there was real value in them. But we saw each agent as a building block, and over time, as we applied autonomous tasks to them, we saw more gains,” Bell concluded.
