Most companies now have an AI pilot that impressed somebody in a demo. Far fewer have one that changed how they compete. The pattern is familiar by now: employees finish individual tasks faster, customers wait about as long as they did before, and the CFO is still looking for the return.
Harry Narang, a global AI-native transformation expert, partner at Rockemere Partners and co-founder of the American School of AI, thinks that gap comes down to a handful of leadership choices that rarely get made together: about strategy, learning, people and accountability. Make them as a set, he argues, and AI becomes part of how the business builds products, makes decisions and runs its operations, with people on the hook for the results. Make them one at a time and you end up with an expensive collection of demonstrations.
Start with direction and outcomes, not tools
Narang’s first point is the one audiences least want to hear. “Everyone wants to talk about tools, agents and automated workflows,” he says. “Those are the easy part. The hard part is deciding what all that speed is for.”
An AI-native organization, in his view, has one thing in abundance: speed. It can prototype in an afternoon, draft in minutes and analyze in seconds. What it usually lacks is direction, and speed without direction just gets a company to the wrong place sooner, using the same tools its competitors bought the same quarter.
“If your competitors can buy and use the same AI capabilities, what will make customers choose you?” he asks. He wants leaders to answer that before anyone picks a vendor. The answer might be serving a market the industry has overlooked, delivering a customer experience that is materially better, or changing the economics of something the company already sells. Whichever it is, that ambition should decide where the money goes.
From there, he wants it made concrete: a measurable outcome, a baseline and a named business owner. He also pushes leaders to be honest about which improvements simply keep the company level with everyone else and which could produce something customers can’t get anywhere else. Both are worth doing. They just shouldn’t be confused with each other.
That owner then has to look at the whole workflow, not the piece being automated. Narang’s standard example is document preparation: automate it, leave the week-long approval queue untouched, and the task gets faster while the customer notices nothing. Which is why the business owner needs real authority to change handoffs, priorities and decision rights, not just a budget for software.
This is also where Narang’s optimism comes through. Used this way, AI is more than a productivity tool. It becomes the fastest, cheapest way a company has ever had to test a strategic bet and find out whether it was right. “The organizations that win won’t be the ones with the most AI,” he says. “They’ll be the ones that knew what they wanted it to do.”
Increase the rate of validated learning
For Narang, AI’s real contribution to innovation is not faster building. It is faster learning: how quickly teams test the assumptions that matter and use what they find to make better investment decisions.
“Building a prototype faster is useful,” he says. “Finding out sooner whether customers need it and will pay for it is worth a lot more.”
He wants product, technology and business teams in the same room identifying the riskiest assumptions, running small experiments and looking at evidence from actual customers. AI-generated personas and market summaries are fine for coming up with hypotheses. They are not a substitute for watching what real people do.
On the practical side, he recommends protected time for experimentation and staged funding tied to evidence. Every pilot should have a decision date, clear criteria to scale or stop, and an owner who is prepared to take a successful experiment into day-to-day operation. Without that discipline, he says, you end up with a portfolio of demonstrations and no route to revenue.
Develop judgment alongside capability
Training, in Narang’s view, should teach people to frame problems, challenge what AI gives them and recognize when human expertise is the whole point. That means role-specific learning, hands-on practice and coaching that happens inside the work rather than in a classroom on a Friday.
The needs differ by role. Leaders have to evaluate investment cases and decide where automation belongs. Product teams have to design experiments and acceptance criteria. Frontline staff need practice checking outputs and handling the cases the system gets wrong.
Coaching is the bridge between learning something and actually doing it differently: helping teams apply new skills, look honestly at failures and fix the workflow. Leaders have to make time for that and reward people who experiment responsibly.
He also raises a question most training plans skip. “If AI takes over the work junior employees used to learn from, leaders have to build another path to expertise,” he says. Managers should be assessing judgment and demonstrated performance, not counting course completions.
Give AI context and bounded authority
A capable language model on its own is not an AI system. Narang puts a lot of weight on connecting it to authoritative enterprise data and curated context: the policies, business definitions and workflow knowledge that make an answer right for this company rather than plausible in general.
Someone has to own that knowledge, keep it current and decide which people and systems can reach it. Retrieve an expired policy or a draft agreement, and the model will give a confident, well-written answer that is wrong for the business.
Access to information is not permission to act. Narang wants leaders to define which actions AI may take on its own, which need a human sign-off, and when it must stop or escalate. Those boundaries need to live in system permissions that can be enforced, not in a line of a prompt asking the model to behave.
Before authority expands, teams should test the system against representative situations, including the awkward ones and the failures. Audit trails, ongoing evaluation and fallback procedures that have actually been rehearsed are non-negotiable in his approach. And the human reviewer has to have the expertise, the time and the authority to step in. A name on an approval screen is not oversight.
Turn productivity into retained value
“Time saved becomes valuable when leaders decide what it’s for,” Narang says.
Freed-up capacity could go toward more customers, faster discovery or less overtime. Leaders need an explicit plan for where it goes. If spending stays the same and the business delivers nothing extra, fewer hours on a task don’t turn into profit on their own.
He recommends agreeing with finance up front on how benefits will be verified. Incremental margin, avoided future spend and released capacity should each be tracked on their own so nothing gets counted twice, and the math has to include the full cost: implementation, model usage, data maintenance, human review and exception handling.
Keeping the benefit also takes commercial judgment. Faster delivery can cut revenue under an hourly contract. Lower service costs can make a customer segment that never penciled out suddenly viable. Narang wants leaders testing pricing and packaging alongside the workflow changes, rather than assuming efficiency gains will stay with the company by default.
His blueprint, in the end, asks leaders to connect what they want from AI to the conditions that would actually deliver it. A credible transformation should be able to say which customer outcome will improve, who owns the change, and what evidence will justify the next dollar.
