Publicis Sapient just released its 2026 Global Enterprise AI Report. They surveyed 1,550 AI decision-makers across industries. 73 percent of enterprises say AI is used regularly or across most business processes. Only 10 percent say AI is core to how their business actually operates.
The reported rates differ by 63 percentage points. They describe different levels of use; the aggregate comparison alone does not establish what prevented integration or identify a separate respondent group.
Respondents describe operating constraints
Forty-two percent of respondents said something specific that deserves attention: they see AI as capable while their own structure, workflows, and operating model prevent them from capturing its value.
Twenty-two percent went further and named the way their organization operates as the primary obstacle to AI success: how work moves through the building.
The survey points to a practical constraint: organizations have access to capable AI and still need to change how they work.
An example of use within an unchanged process
Consider a hypothetical team using AI within its existing process. Someone on the marketing team uses ChatGPT to draft copy. A product manager runs meeting notes through a summarizer. An analyst pastes data into Claude and asks for patterns. A developer uses Copilot to autocomplete functions.
All of that counts as “using AI.” In this example, the underlying business processes remain unchanged.
The team still uses its 2024 approval chain, reporting cadence, and handoffs. Individual tasks may take less time, but the example leaves open whether that saves time across the complete process.
Examine the process before automating its steps
The 10 percent figure is a self-description of AI’s role in the business. To evaluate a workflow change, examine the steps and measure what changed.
That means looking at a process, asking which steps exist because a human used to be the only option, and removing or restructuring those steps before automating what remains.
For a five-step approval chain, examine what each step checks. If AI validates inputs at step one and flags exceptions, test whether any later step duplicates that check. Keep the approvals needed for authority, safety, or compliance.
The Publicis Sapient data lines up with what Deloitte found in their 2026 survey of 3,235 leaders. The report found that 62 percent of organizations have not moved AI projects beyond the pilot stage. The pilot-stage figure alone does not establish why those projects remained there.
Assign responsibility for the process change
The 42 percent reporting organizational constraints give leadership a reason to bring process owners into the review with the technology team. Assign responsibility for the workflow change across the relevant division heads and operations leaders.
A reporting line assigns authority; it does not determine the outcome. Give the owner responsibility for process changes and results as well as the software purchase.
The U.S. numbers in the report are telling. Seventy-one percent of respondents expect significant progress scaling AI in the next 12 to 24 months. But only 20 percent say their organization is fully equipped today to meet those expectations. Closing that confidence gap requires changes to the operating model alongside time and tools.
Test a change in a recurring workflow
If your team uses AI and your business runs the same way it did before they started, focus on integration into daily work. That requires the people who own the operation as well as the technology team.
Pull up a mid-tier workflow your team runs every week that involves at least three people and at least two handoffs. Identify the verification, formatting, routing, or exception handling an agent might take on. Then specify what people would do differently and how you would check that change. Retain approval steps that serve a real control requirement.
For that trial, start with the work itself. Map where time goes, where decisions stall, and where information sits idle between handoffs. Then rebuild those points with AI as a design constraint from the start.
Compare end-to-end completion time, review effort, and errors before and after the trial. That record can show whether the change improved this workflow without treating survey categories as proof of a particular deployment method.