AI projects are clearing launch and missing ROI

A lot of AI projects are learning the corporate art of being both successful and in trouble at the same time.

That is the uncomfortable picture in Laivly’s 2026 AI Deployment Index, released today. 

The Winnipeg-based company, which sells AI software to contact centres, surveyed 200 contact centre leaders across the U.S. and Canada and found 65% called their most recent AI project a success, even as 43% of projects were delayed or stalled and 53% had exceeded budget.

The useful question is what “success” means when customers are getting stuck, front-line staff don’t trust the tools, revenue is leaking, and the budget is blown.

The revenue hit is already showing up. Respondents who say AI has cost them money because the tools can’t handle customer complexity came in at 28%. If you signed off on the vendor contract and told the board the rollout was on track, those numbers land differently.

One in five said revenue loss is happening but they can’t quantify it. Which is arguably worse.

The report found 43% of boards or senior leadership are dissatisfied with AI progress, and that impatience is producing rushed deployments built to show activity rather than results.

“CX leaders have been pushed to use AI and it resulted in companies that deployed without really having the expertise to operationalize and scale their ideas,” said Jeff Fettes, CEO at Laivly.

Close to half of the companies surveyed said their AI tools are driving customer friction, including negative sentiment, repeat calls, and churn. Among companies reporting significant AI-related friction, 57% said they’re losing 5% to 10% of sales.

At the same time, 78% of companies still expect savings through contact centre staff reductions, and 44% plan cuts within the next year. Laivly found the companies cutting most aggressively are also reporting higher customer friction, greater revenue leakage, and higher AI project costs.

More than a third of the leaders surveyed say their teams struggle with AI tools that lose context between customer interactions. A similar proportion say the tools create compliance and tone problems. When the people on the phone don’t trust what the AI is telling them, they work around it, and 36% of the companies surveyed saw turnover rise over the past year.

The disconnect isn’t limited to customer service. KPMG Canada’s Global AI Pulse survey, published in May, found that while 70% of Canadian organizations say AI is delivering meaningful business value, only 3% have achieved measurable returns on their investments.

The companies doing better seem to have picked boring problems first.

Half of the leaders who called their AI deployments successful said one reason was choosing work that could show measurable ROI within 90 days. That forces the team to answer the basic question before launch, not six months later when everyone is pretending the dashboard looks fine.

Did calls get shorter? Did repeat calls drop? Did frontline staff use the tool? Did customers stop getting bounced around?

If nobody can answer that, the project isn’t a success yet. It’s just live.

Final shots

• AI success metrics need to survive contact with the customer, the frontline worker, and the budget.
• Staff cuts are a weak savings story if the same teams are also reporting more friction, more revenue leakage, and higher project costs.
• A 90-day ROI test gives leaders a cleaner way to separate useful AI from software that only looks good on a board deck.