Realizing AI ROI Begins With People And Trust

They pour money into AI, and the returns virtually never arrive. Almost never is the fix another model or a better data set. The quickest route to ROI is human: Build trust, reimagine workflows with people in them, and measure the outcomes that matter to the business.

It's Not the Model, It's The ROI Problem

Studies done by MIT Sloan Management Review have long observed that most organizations find it difficult to scale AI pilots into performance. Meanwhile, Gartner predicts that approximately half of internal decisions will be automated or made by AI agents. The paradox is plain; adoption soars but value lags.

This is evidenced by a corporate survey that SAS and IDC sponsored last year. Most organizations say they already use AI, and another significant chunk plan to begin. But the high confidence in AI expressed by many leaders isn’t matched by adoption of governance and explainability for it, with less than half saying they have implementation in place. That mismatch — enthusiasm but not guardrails — comes to bear in stalled projects, shadow tools and tepid returns.

Trust Is The Infrastructure We Need For AI ROI

Trust is not a feeling; it’s an operating system. The SAS–IDC study highlights three persistent filters that currently stand in the way of ROI: brittle data and cloud footings, flimsy governance, and an AI talent gap. Each is solvable, but no prompt solves any of them. When those tools aren’t available, teams rely on AI sparsely or overwrite outputs or simply dodge using the bait.

The gulf of trust is not only within the enterprise. Outputs from consumer-facing generative features, for instance, AI smarts in search, are still suspect to many people. If end users are distrustful, your employees will be too — unless you give them reliability to see and accountability to hold.

Three Barriers You Can Eliminate Now To Boost ROI

• Strengthen the data plumbing. Think of data quality, lineage and access controls as a product with an owner, backlog and service level objectives. Reliable inputs yield reliable outputs — and reduce the cost of human review.
• Adopt governance by design. Leverage established frameworks such as the NIST AI Risk Management Framework and ISO guidance to govern testing, drift monitoring, bias evaluation and model lifecycle. Publish decision logs, failure modes and fallback rules so teams know when to trust and escalate.
• Close the skills gap — where work happens. The majority of ROI comes from frontline workers designing their own work, not a central lab. Pair short, job-specific training with templates and checklists. Develop “AI champions” within each function able to translate models into better workflows and train other members of their function.

Humanize AI And Make It Accountable For Outcomes

People naturally trust systems that are conversational-feeling. Findings from the SAS–IDC report reveal that individuals frequently trust generative models more than traditional models even though it is easier to explain how traditional methodologies arrive at a decision. That bias is powerful—and risky.

Lean into the human, but calibrate it. Add confidence indicators, show sources and suggest the cost of being wrong. Demand a brief justification for high-stakes outputs, and make it easy to contest the output. Human-in-the-loop review should be a feature, not a bottleneck — especially in customer-facing use cases or regulated decisions.

Measure What the Business Leaders Care About

AI ROI lags when model metrics rather than business performance are how teams measure success. Establish baselines and counterfactual scenarios prior to the launch. Focus on a short list: cycle time, cost to serve, error rate, customer satisfaction and revenue lift. Instrument each step, and decouple automation improvements from shifts in demand or seasonality.

Run controlled pilots with explicit stopping criteria and success boundaries. If most of the gains are due to process not the model, document that and retain changes. ROI is ROI, whether it’s sourced from smarter algorithms, cleaner data or lighter workflows.

Proof From The Field: Real ROI In Human-AI Workflows

Customer service can be a good place to start. Banks and retailers increasingly route routine queries through virtual assistants and present suggested responses to agents. The machine predicts; the person chooses. Companies tell us that with generative prompts alongside playbooks, source citations, and automated call summaries, they’re experiencing faster handle times and fewer escalations.

Content operations show similar patterns. CarMax has said it uses generative models to summarize vehicle reviews, with editors tweaking tone and accuracy. The hybrid workflow — machine first, human final — scaled up production while safeguarding brand voice, a blueprint any marketing department could follow.

Risk functions see gains too. Financial services firms like American Express have for some time merged machine learning with human analyst oversight to prevent fraud, cutting down on false declines while not opening up risk. The lesson is the same: clear accountability, and a human owner of the decision, means AI enhances performance rather than undermining trust in it.

A Simple, Human Battle Plan To Begin With ROI

The right thing to do is choose one high-friction workflow with measurable cost or delay. Designate a business owner and an AI champion. Map the steps, describe what “good” is supposed to look like, and figure out where it is that the model suggests versus a person approves. Operationalize NIST-based controls, get the process instrumented, then train the team for one hour on cues, traps and escalation.

Ship in weeks, not months. Publish the results. Celebrate the humans who reimagined the work, and not just the model that fueled it. Do that a few times, and your AI will have an ROI — because your people will trust it, use it and improve it.