AI ROI: A Practical Path Forward
After publishing the first two parts of this series, I heard the same sentiment from leaders across industries: “We use AI and it’s very useful, but we can’t scale the results.”
The good news is that most organizations are not far behind their peers. The bad news is that the gap will widen quickly for those who don’t move with intent.
There is a practical progression. It’s grounded in data, human behavior, and organizational dynamics. It’s not a five-year transformation program, but it’s also not a month-long pilot. It’s a progression that builds confidence, generates momentum, and compounds its own success.
Organizations doing this well are already seeing that momentum build.
Here’s how to chart a realistic path forward, one that grows with your organization instead of overwhelming it.
Step 1: Start Where People Are Already Pulling
If you do nothing else, start here.
Every major study, from McKinsey to MIT, shows the same thing: AI traction begins with individual behavior, not corporate mandates. You’ll know you’re in the right starting place when employees are already using AI to remove friction from their daily work.
Look for early indicators like:
- People voluntarily using AI for recurring tasks
- Teams asking for access rather than being forced to adopt
- Hearing, “I always use it for…”
The strongest leading indicator of capability formation is deliberate engagement. When people are learning what AI is good at and where it breaks down, they are discovering where the real opportunities hide.
What to track now: At this stage, you are measuring readiness, not return. The goal is to understand where AI is becoming habitual and trusted.
Focus on:
- Adoption depth: weekly active users, frequency of use, and repeat usage by the same individuals
- Emergence of repeatable patterns: the number of recurring use cases employees return to without prompting
- Learning signals: qualitative feedback on when AI helps, when it fails, and where humans still intervene
- Sentiment and confidence: whether employees describe AI as “useful,” “reliable,” or “part of how I work”
Avoid over-weighting time-saved estimates or productivity claims. Usage patterns matter far more than efficiency gains. You are looking for signals that people are learning how to work with AI, not proof of ROI.
Step 2: Formalize What’s Working
Once you see consistent usage patterns, you’re ready for the first real lift: operationalizing AI inside one or two processes that matter.
The mistake here is chasing theoretical ROI. The better approach is choosing processes where:
- Judgment is the bottleneck
- Inputs are recurring
- Outcomes are measurable
- At least one champion is comfortable with AI
McKinsey’s analysis of top-performing AI organizations shows that early success comes from improving process-level outcomes, not attempting enterprise-wide transformation.
These early wins often look modest: reducing cycle time by 15–20%, cutting error rates, improving response quality. But they matter because they teach the organization how to deploy AI, how to test it, monitor it, support it, and intervene when needed.
What to track now: At this stage, measurement shifts from individual productivity to process performance. The goal is to establish whether AI improves outcomes in a way that is stable, repeatable, and explainable.
Track metrics such as cycle time, throughput, error rates, exception rates, customer satisfaction, and cost per transaction.
Early wins may look modest. A 10–20% improvement is often enough. What matters more is consistency: predictable performance, known failure modes, and clear handoffs between AI and human judgment. These metrics tell you whether AI can be embedded into real workflows without increasing operational risk.
Step 3: Move Toward Revenue Only When the Foundation Holds
This is the step everyone wants to start with. It’s also where the most AI programs derail.
Organizations that succeed here understand that revenue experiments are not operational projects. They invest in what McKinsey calls the capability stack: data readiness, workflow integration, talent depth, and governance.
What to track now: Revenue-facing AI requires patience and discipline in measurement. These initiatives rarely produce immediate lift, and attribution is often indirect.
Focus on leading indicators that correlate with long-term revenue impact: conversion rates, win rates, deal velocity, average deal size, retention, expansion ARR, and customer engagement trends.
The goal is not to prove short-term ROI, but to determine whether AI is influencing customer behavior and delivering value. Treat these metrics as signals, not verdicts.
Step 4: Know When You’re Ready to Scale
You’re ready to scale when three things are true:
- You see organic pull from multiple teams Not mandates. Pull. Multiple departments asking, “Can we use AI for this?” is one of the strongest predictors of scalable value.
- Your operational deployments behave predictably Processes have clear baselines, stable performance, and known exceptions. AI doesn’t collapse when conditions change.
- You can explain the value story in plain language Organizations that scale aren’t the ones with the best models. They’re the ones who can clearly articulate:
- What’s working
- Why it’s working
- What it unlocked
- What the next logical use case should be
The Real Takeaway
Scaling AI isn’t about moonshots. It’s about sequencing.
Give people room to experiment. Use their behavior to guide your first operational bets. Use operational success to build trust and competence. Then take on revenue and push the envelope.
The companies winning with AI aren’t lucky. They’re deliberate. They start small, learn quickly, measure honestly, and scale only when the system, not just the model, is ready.
If your AI program feels chaotic or unclear, it doesn’t mean you’re behind. It probably means you’re normal.
And if you want to know exactly where to step next, look at what your people are already doing. The path forward is usually hiding in plain sight.
Part 3 of 3
