Despite the promise of increased productivity and profitability, a vast majority of enterprise generative AI initiatives are canceled during the pilot phase due to inappropriate use cases, overambitious deployments, and governance issues.
Adoption of AI, particularly generative AI, has become popular among enterprises because of a promised step change in productivity and profitability. But the shocking truth is that most generative AI initiatives are getting quietly canceled while still at pilot stage.
A recent MITrevealed that around 95 percent of enterprise AI projects don't deliver any measurable returns and get binned before they reach full scale production. Why do so many of these pilots, launched with high hopes and based on powerful technology, fail to go the distance? Dr. Lou Bachenheimer, CTO Americas with SS&C Blue Prism, identifies a mix of reasons.
"The most common mistake is that the chosen use case is just not an appropriate one for generative AI," he warns, suggesting that the rule of thumb should always be to use the simplest tool for any given task. "If there's a cheaper and faster deterministic or machine learning-based model that you could deploy, then that would generally be more logical than using generative AI just because you can.
" Another common failing is deploying generative AI across an entire use case rather than taking a more tiered route and using it for the part of a task where it can best deliver. "Generative AI is great in certain roles, such as adding structure to unstructured data," says Bachenheimer. "Once accomplished, you can hand off the resulting data to simpler but more established and auditable methodologies so that they can complete the job.
"The excitement factor around generative AI can blind people to issues of governance until it's too late. What starts as a flawlessly great idea in the mind of its originator looks different to a legal team looking for regulatory risk.
"Part of the problem may be that generative AI, unlike most enterprise software, is something that ordinary people use for themselves all the time," believes Bachenheimer. There's a big difference between someone using AI to create a cake recipe, vs using it in a business context. As a technology based on natural language, generative AI will inherit any biases and flaws built into its training data, and hallucinate about what it can't find.
Deployment teams must also ensure that its decisions are auditable and explainable. This is more achievable with modern reasoning LLMs than it was with pre-LLM classical machine learning systems, but it's something that must be on the to-do list, especially in regulated sectors. Great efforts are ongoing to develop LLMs that are more consistent and less prone to bias. But any system that contains a whiff of fallibility or is tricky to audit must have a strong governance layer.
Data privacy concerns around the information being fed into the models exacerbates enterprise challenges. SS&C has skin in the game, being active in the tightly regulated financial services sector. It knows the sting of AI pilots that flop under scrutiny. That is why before deploying AI in a full production environment, it decided to go about building a specialized governance gateway.
All calls to the LLM are first run through this gateway which imposes certain guardrails, checking for sensitive data leakage, toxicity or malicious prompt injection before providing a verdict back to where the request came from.
"It's a way to eliminate bias and hallucinations," says Bachenheimer. "And it means we end up with an audit log to prove that we have done everything in our power to ensure that there aren't issues cropping up. We used this for a full year before deciding it was something our customers could benefit from and turning it into a product.
"ROI is also a factor. A good system architect takes the most efficient approach to automation, only using generative AI when it can add true value. Not least among the reasons for this kind of structured approach is the need to generate return on investment in a reasonable timescale. Generative AI is a powerful tool, but it can be expensive, potentially killing off a project's rationale before the green shoots of any dividends can materialize.
Dr. Bachenheimer explains how SS&C has gained this kind of hard-won wisdom from its own experiences:"We were early adopters of agentic and generative AI-based technologies," he says.
"We realized early on that you can significantly reduce costs by only using generative AI where it's actually needed, and you can reduce risk by separating out any sensitive data and segregating it from the AI completely. " "Many of the governance and guardrail solutions you'll see on the market are somewhat piecemeal, not designed for specific regulated industries," warns Bachenheimer. "We've solved that, so you don't have to go and build it yourself.
" Without this kind of governance, a heavily regulated industry like finance or healthcare runs the risk of getting stuck at square one when it comes to deploying generative AI. Strong governance is fundamental, but without orchestration, even the most advanced technologies struggle to deliver value. SS&C Blue Prism WorkHQ is the result, an agentic automation platform that brings AI, people, and business systems into a single, governed workflow, providing a unified environment to orchestrate and streamline work.
It represents a major evolutionary step forward from basic robotic process automation software and simple automation, integrating every workflow seamlessly and letting enterprises scale new AI-driven use cases quickly. The strength of a unified platform becomes clear when you are deciding what tool is best for the use case you have in mind. The right platform can orchestrate that, ensuring that AI fits in where it makes sense, and doesn't get thrown at every problem.
There are various forms that this orchestration can take, explains Bachenheimer:"There's long term orchestration where you are using some kind of business process management technology," he says.
"Secondly, there's the kind of orchestration where you are dealing with different sets of tools, perhaps APIs or something that calls out to different AI agents. You need to take all those tools and put them into a flow.
" Thirdly comes the exciting bit where you move beyond deterministic models and start building something agentic. True agency starts where you're providing the AI agent with a list of tools and a goal and letting it do its thing. That's when you're getting real power out of the agentic landscape, believes Bachenheimer.
"You need all three types of orchestration in an agentic platform for it to function properly," he concludes. "Without them, any platform is fundamentally flawed. Most platforms out there are missing the element of user interface automation. Without that you are not getting results that are auditable, affordable or low in risk.
"WorkHQ is battle-tested across SS&C's own global operations. In fact the company is the largest consumer of its own software with over 3,400+ automations in production, between them generating substantial returns.
"We learned a lot in the process of scaling out the required capabilities," says Bachenheimer. "We combined all that into our latest platform. You can now keep all your automation in one control plane, manage it, orchestrate it and govern it in one area.
" The technology has been put to the test in heavily regulated financial services and healthcare settings. All this maturity is now combined into one platform giving adopters the ability to leverage new capabilities quickly and start to see real benefits in an actual regulated setting. This ability to deliver swift returns is critical in a context where AI doesn't have the luxury of two or three years of 'wait and see'.
The ax is liable to fall well in advance of that. Generative AI can be put to some pretty advanced tasks right out of the blocks. Many current use cases focus on simple data digitization or summarization of data. WorkHQ enables you to apply AI to real business use cases, involving both scale and complexity.
"It offers the right platform at a time when the gains from applying AI are starting to move on from a matter of greater efficiency to actual revenue growth," claims Bachenheimer. "It gives you AI that moves the needle and results in actual returns. You can start to grow the business rather than just make people more efficient.
" First step, he says, is a paradigm shift away from trying to use AI to fix everything to a mindset of using it where appropriate. When you want the right tool for the right job, you need a platform that can achieve that safely, reliably and cost effectively.
Generative AI AI Adoption AI Failure Digital Transformation Enterprise Technology
United States Latest News, United States Headlines
Similar News:You can also read news stories similar to this one that we have collected from other news sources.
Maple Gold Mines Reports Increased Gold Resources at Douay and Joutel ProjectsMaple Gold Mines announced a significant increase in the resource estimate for its Douay and Joutel gold projects in Québec, Canada, exceeding 5 million ounces of gold. The update reflects successful drilling, improved modelling, and positive economic assumptions, including a maiden high-grade underground resource at Joutel. The company's stock has seen substantial growth over the past year.
Read more »
Kris Boyd Predicts Pressure for Danny Rohl if Rangers Fail to Win TitleFormer Rangers striker Kris Boyd suggests Danny Rohl will face increased scrutiny if Rangers don't win the Premiership, particularly if they finish third. A recent defeat to Motherwell has significantly hampered their title chances, leaving them trailing Hearts and Celtic.
Read more »
'I've worked with David Attenborough for years - he's brilliant to work with'A film producer has worked with Sir David Attenborough for a decade over five projects.
Read more »
Three NI special school projects approved as SEN demand hits record highA new 500-pupil special school in Belfast as well as increased pupil numbers at two other sites are being planned
Read more »
Trump admin pays wind developers to quit, back fossil fuel projects: DoI offers up to $885M if they abandon offshore wind projects
Read more »
Lizzo Celebrates 38th Birthday in Bold Style, Promotes New Album and Upcoming ProjectsLizzo marked her 38th birthday with a daring fashion statement, celebrating with close friends and promoting her upcoming album Bi***. The Grammy-winning artist also shared updates on her Netflix special and an upcoming biopic, showcasing her multifaceted career.
Read more »