The most expensive part of AI isn’t the model, it’s the habits
November 12, 2026
reading time 7 minutes
A familiar scenario: a company buys AI licenses, holds a webinar for colleagues, and gives them access. A few months later, it turns out that only a few enthusiasts use the tool, while everyone else works as before: spreadsheets, chats, “send me the report.” The money has been spent, but the habits remain the same.
The problem is not a bad model. BCG surveyed more than 11,000 employees and found a paradox: a clear strategy with weaker tools delivers more than the best tools without a strategy. In the first case, 80% of people see a noticeable impact; in the second, 60%. And in a Gartner survey, data team leaders cited resistance from people more often than a lack of money.
You can buy technology in a week. Changing culture – meaning “the way we do things here” – is much harder. Its visible part is habits: where people look for numbers, whom they ask, and what they discuss in meetings. Deeper beliefs change more slowly, but it is habits that determine whether AI pays off.
The systematic work with people during such changes is called change management: its goal is to make a new way of working a habit rather than leave it as a pilot. Below, we explore why people hold on to old ways, who can actually change this, and where to start.
Key points
Your people are already using AI
often secretly and without rules
Resistance starts where people do not see a problem
They think “things are fine as they are,” so they see no reason to change anything
AI does not work by order – a leader’s example does
You can require people to track tasks in a CRM, but you cannot require them to think with AI
Without shared definitions of data
AI destroys trust faster than it builds it
Shadow AI: people are already using it, they just aren't telling you
One of the largest studies of trust in AI – by KPMG and the University of Melbourne, with more than 48,000 respondents across 47 countries – paints this picture:
58%
of employees regularly use AI at work, but around 70% do so with free public tools, and only 42% use tools provided by their employer
61%
do not say they used AI to do their work
66%
use AI answers without checking them, and 56% have already made mistakes because of this
44%
have used AI against company rules – although only 34% of organizations have rules for generative AI
So the problem is not that people are afraid of AI. They are reluctant to talk about using it.
In a culture without rules or permission to make mistakes, AI becomes a personal habit for a few individuals rather than a new habit for the company. Meanwhile, work data ends up in services the company does not control.
“Why fix what isn't broken?”
Resistance to change is usually seen as sabotage. But most often, people resist not the change itself, but the fact that they do not understand why it is needed.
“For many years, I implemented SaaS solutions in companies: CRM, project management, customer support, and so on. Every time, the hardest part was not integrations, but people. For them, everything was already working: they did not see a problem and therefore saw no reason to change anything. Back then, a simple rule helped us: if a task is not in the system, it has not been assigned; if it is not closed in the system, it has not been completed. In meetings, we discussed only what was in the system. With AI, resistance looks different. People do not reject it; they use it in their own way: secretly, in their own apps, often without checking the answers. And the team's shared way of working does not change. You cannot fix this with an order: you cannot force someone to think with AI. So the leader's role is different here – not to control, but to demonstrate. To ask the assistant questions in meetings, discuss its mistakes openly in front of everyone, and agree on which answers can be trusted. When this becomes normal, there is no longer any reason to hide.”
– Vitalii Safaniuk, CBDO gloozd
The CEO explains “why”, the manager shows “how”
Change needs two voices. People want to hear from the CEO why change is needed – and see that the CEO believes in it. But they want to hear from their direct manager how the change will affect their work. It is this manager who shapes the team's habits every day. In a 2026 Gallup study (more than 23,000 employees in the US), the difference is striking: where managers actively support AI, 33% of employees say AI has changed their work. Where managers do not support it, only 4% say so. Meanwhile, half of HR directors at Fortune 500 companies are not confident that their managers can teach teams to work with AI at all.
BCG sees the same thing from another angle: only 33% of employees say leadership clearly explains its plans for AI. 88% expect they will need significant retraining, while only 36% feel sufficiently trained.
Data is the foundation of trust in AI
Trust in AI can break in two directions: some people do not believe any number, while others do not check anything. And most often, it is not the models that break that trust, but the data they work with.
Imagine a retail chain manager asking an AI assistant: “How many active customers do we have?” For marketing, an active customer is someone who has made at least one purchase in the past 12 months. For the loyalty program, it is someone who has taken any action with their card, even without buying anything. For finance, it is someone who generated revenue in the reporting month. No one is wrong, but the numbers are different. The assistant confidently gives one number. Two departments see the “wrong” number – and the rollout is over before it has even begun.
Shared definitions of metrics and clarity about who owns them are not bureaucracy – they are a way to protect trust. The business defines what a metric means; IT makes sure the data is reliable.
We have been through this ourselves: gloozd grew out of the internal data team at !FEST Emotion Holding. Analytics there was rebuilt around three levels: BI reporting, self-service analytics in Looker, and an AI assistant that can answer questions in natural language (demo, case study).
!FEST
Previously, a typical request to an analyst took 2–3 working days, with around 150 such requests per month. Shared metric definitions were set in Looker's semantic layer, so an “active customer” means the same thing in a dashboard and in the assistant. Within a year, managers started building around 60% of reports themselves, and getting an answer now takes minutes instead of days.
The main change was not in the tools: the data team stopped being the only gateway to the numbers, and managers stopped writing “prepare a report.”
read the full caseWhere to start
1. One process that already hurts
Not “implement AI across the company,” but cut the time needed to evaluate a campaign from a week to a day. People accept change fastest when it addresses a specific pain point, and a first quick result shared with everyone is more convincing than any presentation.
2. Bring AI out of the shadows
Simple rules, approved tools, and an honest conversation about what already works. People who have been using AI secretly should not be judged; they should be made mentors.
3. Start with managers
Middle management often resists change the most – and it is this group that determines whether AI becomes a team habit. Train managers first and explain why this matters to the company.
4. Show, don't order
The leader asks AI questions in meetings and openly discusses its mistakes. And agree on the ground rules: which AI answers must be checked and who is responsible for decisions made with its help.
5. Put definitions in the system, not in a document
A definition in the semantic layer works in every dashboard and every AI answer. A definition in Confluence works as long as someone reads it.
6. Measure habits, not licenses
Time to an answer, disagreements over numbers, the share of data-driven decisions – not the number of activated accounts.
AI is a change project, not a purchase
Models are getting cheaper and smarter every quarter. Habits are not. Where a company redesigns processes instead of simply handing out licenses, 67% of employees see measurable results, compared with 43%, according to the same BCG research.
You cannot change culture with an order. But you can change its visible part – habits – one at a time: through a rule, a meeting, or a leader's example.
So before your next AI project, ask not only “which model should we choose?” but also “what needs to change in how our people work and make decisions?” The answer to the second question will determine whether the investment pays off.
Want to find a process to start with? Write to us – and we will work through it together with your team.
Contacts
hello@gloozd.agency
Reference:
gloozd – a data & AI agency that grew out of the data team of the !FEST Emotion Holding. The company builds data infrastructure, implements AI solutions, and researches customer behavior using sociological methods, working in three directions: data bureau, R&D AI lab, and cx analytics. A certified Google Cloud partner. More details: gloozd.agency
// ready to start?
