Many operations teams have purchased AI licenses that remain unused. They made the purchase, sent out an announcement, and anticipated an increase in efficiency. Unfortunately, that increase never happened.
The reality is that the issue of AI adoption in business operations is not about procurement. It’s about change management, but it looks like a procurement issue.
The Shiny Object Trap
When you see something operational that can be improved – such as a bottleneck in procurement, a slow approval chain, or a bloated reporting cycle – the typical reaction is to look for a tool that can help fix it. AI tools are often top of the list since their presentations are truly remarkable, their sales pitch corresponds directly with your source of discomfort, and the subscription fee can easily be justified when compared to a full-time employee.
The issue is that these tools are usually bought without first analyzing the workflow. No one stops to wonder: how does this tool fit in with the current 9 am on a Tuesday reality of my employees? Where does the output of this tool lead? Who is responsible for reviewing the output and what is their next step?
The reality is that as long as these questions are not answered before the purchase is made, they will likely remain unanswered once the contract is signed. The tool will only be more or less sporadically used at best. Or worse, not used at all and only added on top of everything else.
The Integration Bottleneck Is Real
Stand-alone AI tools do not solve the problem of data silos; they replace them with operating silos. If your writing assistant AI doesn’t plug into your CRM, or your meeting summary AI doesn’t talk to your project management tool, or your AI analytics can’t query your operations database, your employees end up in the very copy-paste routine that artificial intelligence was meant to eliminate. Only now, they are copying and pasting data between operational silos.
This is also a massive cognitive overhead for your employees. Switching context between four different separate applications to complete one task is exhausting. These productivity losses will quickly gobble up the gains you expect your AI investment to yield.
The alternative is to start with software tools that already seamlessly plug inside the operational environments where your employees already are. The approach to microsoft 365 ai integration is a good example of how software vendors are beginning to do this. Spreadsheets, email clients, and documents already have the users. Friction is minimal when they are the platform.
You Can’t Automate A Broken Process
One of the most important points that tends to get lost in the conversation around AI and business operations. Putting AI on top of a broken process doesn’t fix the process. It just runs the broken process faster. If your expense request form has five unnecessary approval fields because of a long-ago reorg, no amount of AI-enabled natural language processing is going to fix that. If your new client onboarding takes forever because it’s never really clear which of two departments “owns” the new client, automation will simply proliferate the opportunities for that question to arise. The machines don’t care why you do things the way you do.
Before implementing any AI tool in a production environment, the process it’s designed to help needs to be unraveled and examined. Which steps are there because they deliver real value? Which steps are there because someone started including them years ago and no one has thought to object? This work isn’t terribly liberating but it’s the heavy lifting where the real return on investment is generated.
The Training Gap Nobody Talks About
Purchasing software does not equal adoption. Yet the two are often mistaken for one another, particularly when the metric of success used by a team is how many licenses they’ve been able to assign, rather than how much meaningful output the organization is creating and delivering.
Real adoption – the kind that leads to impact – is reached only when everyone in the organization understands how to work with AI outputs. This means being able to do more than press a button to get something that looks like writing. It means knowing how to write a useful prompt, how to spot when an output is wrong or incomplete, and how to iterate on what comes out of the machine in the service of their goals.
Continuous upskilling is key here. Not a one-hour seminar at launch – a steady, ongoing educational process that changes as the tools change. 72% of organizations have adopted AI in at least one business function, but the value creation gap is growing, blocked by user adoption and implementation challenges and a lack of a clear strategy, not by the technology itself (McKinsey). The people gap remains without systematic investment in bridging it.
Governance Before you go wide
One area where operations leaders tend to move too quickly is data governance. When all of a sudden, your employees are going to start using AI tools, en masse, you urgently need to have a way to answer the question: ‘What data exactly is supposed to go where?’
You don’t want to wait for an employee to access a model they shouldn’t have, or for a news story to break about your firm inadvertently training on some of its most sensitive data before setting these policies and processes up. You need to do it ahead of time, which, if you’re reading this article, you still have.
Measuring What Actually Matters
Measure the time saved, rather than the number of occupied spaces. Measure the reduced time taken for a specific process. Measure how many manual interactions a system has saved over a time window. This is where you will see actual usage and impacts of AI – within the operation.
The number of occupied spaces is a purely purchasing-oriented metric. The rest are operational ones. If your reporting only shows the first, you won’t see the problem until the renewal conversation.
The fact is that AI can truly change how back offices operate, but the software and licensing are about 10% of the story. The rest is change management, and that is a people challenge, not a machine one.
