The trust deficit is the missing layer beneath every AI ROI dashboard

A new report is making the rounds this month, and it is worth reading closely if you sign the AI invoices. On 8 July 2026, Andus Labs published findings that put the blame for weak AI returns somewhere most dashboards never look. The claim is blunt: enterprise AI returns are decided by how a company operates, not by which models it buys, and most generative AI pilots deliver no measurable financial impact because of outdated workflows, decision rights and incentives, not technology (Source: Andus Labs, July 2026).

The part that stuck with us is what they named the top pattern blocking returns. They call it the Trust Deficit. Leaders treat probabilistic AI as a deterministic search engine, then call it broken when it doesn't behave like one (Andus Labs). It is an expectations problem, and measuring it is how you find it. It sits underneath every ROI number you are trying to defend.

Why the dashboards keep missing it#Link to this section

This week the standard fixes are everywhere: spending caps, usage dashboards, forward-deployed engineering teams parachuting into departments. None of that is wrong. But all of it measures deployment, and deployment is not the same as return. You can hit every adoption target and still change nothing about how decisions get made.

The scale of the write-off is real. According to S&P Global Market Intelligence, the share of companies abandoning most of their AI initiatives reached 42%, more than double the year before, and the average organisation scrapped 46% of its proof-of-concept projects before reaching production (Source: S&P Global Market Intelligence, 2025). Andus Labs is direct about the cause: incentives that still reward old behaviour, and teams quietly routing around tools the organisation never re-staffed for (Andus Labs).

There is a belief problem sitting on top of the incentive problem. Gallup found that only 13% of US employees use AI daily at work, and just 28% use it a few times a week or more (Source: Gallup, April 2026). And a 2026 FlexJobs survey of more than 4,400 workers found that 42% fear AI is coming for their role (Source: FlexJobs, 2026). People who feel threatened use a tool to look compliant and keep working the way they always did. A dashboard reads that as adoption. Your P&L reads it as spending with no return.

What good measurement actually tracks#Link to this section

The encouraging news is that measurement discipline is becoming normal. Wharton found that 72% of business leaders report tracking structured, business-linked ROI metrics such as profitability, throughput and workforce productivity (Source: Wharton Human-AI Research and GBK Collective, October 2025). The question is no longer whether you measure. It is whether you measure the thing that actually moves returns.

If the Trust Deficit is the real blocker, then counting licences and logins tells you almost nothing. Here is what we would track instead.

Decision quality, not deployment counts. Pick the decisions a tool is meant to improve, then measure whether those decisions got faster, cheaper or more accurate. Good looks like a shorter cycle time on a named decision, with the error rate holding or falling. If nobody can name the decision the tool changed, you are funding activity, not outcomes.

Workflow change, not tool access. Ask whether the work itself was redesigned around the tool, or whether the tool was bolted onto the old process. A tool inside an unchanged workflow produces the Trust Deficit by design, because it is being asked to behave like the deterministic step it replaced. Good looks like a documented before and after in how a task actually runs.

Expectation calibration. Ask people whether the tool does what they were told it would do. A large gap between the promise and the lived experience is the Trust Deficit showing up as a number, and it predicts abandonment long before your renewal date.

Unused seats, out loud. Track the tools people pay for and quietly route around. A usage dashboard shows the idle seats. What it can't show is the busy seat whose owner quietly works around the tool.

The cleanest way to read all four is straight from the people doing the work, anonymously and often, so the signal isn't filtered through the person whose budget is on the line. Wharton's own framing is that ROI is now measured, and people, not tools, set the pace (Wharton Human-AI Research and GBK Collective). If people set the pace, people are also your measurement instrument.

The point#Link to this section

Spending caps and dashboards are fine. They sit on top of the operating model, and the operating model is where returns are won or lost. Before you buy the next tool or build the next dashboard, measure whether the last one changed a decision or a workflow.

The GAiGE covers the people side of this: whether each tool does what people need, and how often they have to fix its output.