AI offers the potential for operational efficiencies, reductions in human error, a solution for employee attrition, and split-second analytical recommendations, but the reality is that no one — in the business, outside the business, or even CIOs themselves — knows whatto expect from AI.
And still, businesses are aggressively pursuing it, in many cases in hopes of transforming their organizations.
What we do know for sure is that AI deployment is like other IT projects in one important respect: Its success depends on employee, customer, and management acceptance. Ultimately, the degree to which AI is accepted by these constituencies “level-sets” how far AI can be adopted.
Common areas of level-setting include employee willingness to adopt AI in their work, customer willingness to accept AI-based customer service, management trust in AI decision-making, and
AI integration with systems of record.
In traditional IT projects, there was no level-setting — only user acceptance testing to let IT know whether its user-customers were satisfied (or not) with new systems and how they worked.
AI projects aren’t like this. They require multiple layers of acceptance, such as: Does the functionality work? Are we getting the results from the system that we expected? Do we likethe system? Do we want to workwith the system?
It’s vital for IT and business management to understand this at a time when MIT reports that despite nearly 80% of organizations piloting and using AI engines such as ChatGPT to enhance individual user performance, the addition of these tools has done little to impact corporate P&L in a positive way. The MIT report further concludes that 95% of organizations piloting or deploying AI are getting little or no return on their investments.
“Enterprises are piloting GenAI tools, but very few reach deployment,” said MIT. “Generic tools like ChatGPT are widely used, but custom solutions stall due to integration complexity and lack of fit with existing workflows.”
Problematic fits with workflows and system integration take us back to the rudiments of system functionality, business results, user satisfaction, and users’ willingness to adopt and work with a new system. And it is the concept of “level-setting” that can turn tepid AI outcomes into more productive gains.
Here’s how.
Continuous model training
A West Coast cardiology clinic obtained a new AI system to assist with diagnostics of heart conditions, as well as recommendations of treatments and medications for patients. The system was brought in with baseline functionality that had been trained with US data, and the clinic wanted to further customize this data to fit the profile of its patient population.
It did this by obtaining permission from patients to input their cardiac information, medications, and treatments into the system. Some patients agreed to share their information, and some did not. Meanwhile, some cardiologists remembered to ask for and record the information, and some did not. The end result was an imperfectly trained AI system that could skew outcomes because it didn’t have access to all the information.
This situation is true for most enterprise AI systems. It is never possible to guarantee that these systems are fully trained, because the reality they operate in is constantly changing.
Level-set: The best way to level-set projects like this is to continuously monitor systems for accuracy of results. If their accuracy level is 95% at the time they are deployed and they begin to lose accuracy, it’s time to retrain and re-level-set them to the expected accuracy level.
This can best be done in AI projects by making AI improvement continuous. In other words, don’t end the project on the day the system is installed. Instead establish the degree of accuracy you want to maintain, and collectively engage users and IT to maintain that accuracy, entering into an AI model retraining phase when results accuracy declines.
Workflows that flow
Is the new AI fitting comfortably into the business workflow it was designed for? Does it seamlessly blend with work processes, or chop them up and upend them? Can users easily work with the AI, or do you see them working around it?
The end objective of any AI business process infusion is that the AI disrupts business in a positive way — not that it disrupts business processes.
Level-set: If there is too much business process disruption and too little business gain, it’s time to stop, re-level-set and move forward. This time the achievement level for the AI and its user interface might be more gradual than it was the first time around.
User acceptance
WalkMe, an SAP company, reported in April 2026 that “over half (54%) of workers bypassed AI tools and completed tasks manually at least once in the past 30 days. A further 33% haven’t used AI at all. Rather than friction, the research describes outright rejection.”
This worldwide study of 3,750 executives and employees at enterprises with 1,000 or more employees also found, “Only 9% of workers trust AI for complex, business-critical decisions, compared to 61% of executives — and that 88% of executives are confident employees have adequate tools, [but] only 21% of workers agree.”
User acceptance is a major red flag for AI projects, because there is an inherent fear of job loss and change that biases many of users against AI in the first place. And when users refuse to accept AI into their workflows, AI projects fail, despite what executives think.
Level-set: When it comes to user acceptance, level-setting targets in AI projects should be incremental. Maybe users have a hard time letting AI fully process an invoice from start to finish, but they will let the AI enter into the process as a starter to verify whether the terms of an invoice are being upheld, or if the party the invoice is being paid to is correct. Once this level of acceptance is achieved, the project can be extended to incrementally add automation step by step, with each step going through a new level of user acceptance.
Given that AI performance isn’t always good, and users may be anxious about AI, this step-by-step approach also enables users to see they are integral to the AI adoption process — a key realization to being motivated to move forward collaboratively with AI.
The human side of your AI agenda
At first glance, it might seem to CIOs that they already have “level-setting” built into their projects. “We call this a new project phase, or enhancement requests for systems in production,” you might think.
Not so. Why? Because level-setting might not have anything to do with system functionality or enhancement.With AI, companies are now working with the man-machine interface in business processes, so level-setting becomes more a question of how much AI humans are able to ingest as they see their own jobs change.
This is a humanistic issue that every business should consider.
Read More from This Article: AI upends key pillars of project management
Source: News