How to lead an AI pilot as a non-technical manager

A non-technical manager leads an AI pilot by choosing one narrow process, measuring it first and fixing a decision date for go, change or stop. The need is visible in the data: of the 200 organisations EY and CII surveyed in November 2025, 47% had multiple GenAI use cases live in production and 23% were still in pilot.

Why do many AI pilots stall before production?

The surveys measure different stages, so they are best read side by side. Microsoft's 2025 Work Trend Index, covering 31,000 workers in 31 countries, found 24% of leaders say AI is deployed organisation-wide and 12% remain in pilot mode. A preliminary MIT NANDA report of July 2025 found 60% of organisations had evaluated custom task-specific AI tools, 20% had reached pilot and 5% had reached production. The report draws on interviews at 52 organisations, a survey of 153 senior leaders and 300+ public initiatives, and its authors note the interview sample may not represent the wider market.

The reasons are less about technology than about scope and guidance. BCG's 2026 survey found a clear AI strategy lifts measurable business impact by 25 percentage points, while better tools alone lift it by about 5. It also found 66% of regular frontline users get limited or no guidance on what to do with the time AI saves them. Gartner predicted on 25 June 2025, as reported by W.Media, that more than 40% of agentic AI projects will be cancelled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls.

People matter as much as the tools. IBM's 2026 CEO study found 83% of CEOs say AI success depends more on people's adoption than on technology, and respondents expect 29% of employees to need reskilling for a different role and 53% to need upskilling in their current role between 2026 and 2028. In the World Economic Forum's Future of Jobs Report 2025, 86% of surveyed Indian employers plan to reskill and upskill their existing workforce to work alongside AI between 2025 and 2030, against 77% globally. Naukri's AI at Work report of 16 July 2026 found 77% of professionals already use AI at work at least sometimes.

The table sets the survey evidence side by side.

The BCG and Gartner findings point to strategy, guidance and risk controls as the gaps. The steps below address those gaps. For the wider adoption picture, see what enterprise AI adoption in India means for managers and 15 ways to use AI at work in 2026.

How can a manager avoid spending months on a pilot nobody wanted?

Start with one process the team already finds tedious. Choose something repetitive and measurable, where a wrong output causes little harm, and avoid anything that touches customers or money directly in the first round. Writing down what the process is today comes next: how long it takes, how often it goes wrong and who handles the exceptions.

Take a hypothetical finance manager whose team matches vendor invoices to purchase orders each month. The baseline is the time a batch takes now, the share of invoices that need a human to resolve, and the errors found after payment. The success criteria are written before the pilot starts, for example the same error rate with less handling time. Writing them afterwards lets any result look like a win.

Expectations should be modest and specific. Brynjolfsson, Li and Raymond found a generative AI assistant raised productivity by 14% on average across 5,179 customer support agents, and by 34% for novices, with minimal impact on experienced workers. A Science study reported by MIT News on 14 July 2023 found ChatGPT cut the time taken by 40% in 20 to 30 minute writing tasks for 453 professionals. Neither covers a given team's process, which is why the baseline matters. A Danish study of administrative records found no earnings or hours effect two years after ChatGPT's launch, so savings may show first as reorganised tasks.

The date for the decision is set at the start as well. A pilot with no end date turns into permanent experimentation, and a fixed date forces the question of value that Gartner lists among the reasons projects are cancelled.

Which pilot roles can a manager staff without engineers?

One way to staff a pilot is with five jobs, and none needs code. One is a sponsor with authority to say go or stop. A second is the process owner, usually the manager. A third is a team member who knows the exceptions, and a fourth is a reviewer who checks a sample of outputs each week. The fifth is a named contact in IT or information security who approves tools and data access.

Microsoft's India findings name AI Workflow Designer among roles 92% of leaders are considering, and the Microsoft Work Trend Index lists AI Trainer (32%) and AI ROI Analyst (29%) among global roles managers are considering. They are intentions, not vacancies.

For small internal tools, a manager can often create AI software without engineers. Microsoft said on 20 August 2025 that customers created over 3 million AI agents using Copilot Studio and SharePoint in its fiscal year 2025, and that more than 230,000 organisations were using Copilot Studio. Those are Microsoft's own global figures, and created does not mean in production. Anything that touches personal data needs the security contact's approval, and the rules are covered in DPDP Act and AI governance for business leaders. Do you need coding to learn AI discusses the skill question.

How much training does a pilot team need?

More than an awareness session. BCG's 2025 survey of over 10,600 workers found employees who receive five or more hours of training are more likely to become regular AI users, yet only 36% feel adequately trained. Naukri's AI at Work report of 16 July 2026 found 79% of professionals are upskilling or plan to start soon, while only 35% say their employer actively provides AI training.

Managers carry part of that burden. In Microsoft's India findings, 63% of managers expect AI training to become a core team responsibility within five years, and 51% of leaders cite upskilling as their top priority for the next 12 to 18 months. Training for a pilot works best when it uses the pilot's own process: sample inputs, the exceptions the team knows, what the tool may and may not receive, and how to flag a wrong output.

BCG's 2025 survey also found 54% of employees would use AI tools even if not authorised, so the pilot team needs the rules for approved tools stated plainly. AI skills for non-technical professionals describes the wider skill set employers ask for.

How should a pilot run, and when should it stop?

Run it supervised. The reviewer samples outputs against the baseline, and every error is logged with its cause. The reason is the finding of Dell'Acqua and co-authors in their 2023 Harvard Business School experiment with 758 BCG consultants: on 18 tasks inside AI's capability frontier, AI users completed 12.2% more tasks and 25.1% more quickly with higher quality, but on a managerial task outside it, AI users were 19% less likely to produce correct solutions. Whether a task sits inside or outside the frontier is found by testing, not by assuming. If the pilot uses an agent that acts, agentic AI in finance and operations covers approvals and limits.

On the decision date, the choice is go, change or stop. Go means the measured result beats the baseline at an acceptable cost and risk, so the pilot widens. Change means the idea holds but the scope, data or checks must be adjusted. Stop is a legitimate result, and a clean stop is worth more than a pilot that drifts on.

The final recommendation should state the baseline, the result against it, the cost per task, the risks and controls in place, the decision requested and the next scope. It should also say what the team will do with the time saved, since BCG found 66% of regular frontline users get limited or no guidance on that point.

Where a taught programme fits

Managers who want a structured route have one option in the AI Officer Programme, a live online programme by Vishlesan i-Hub Foundation, IIT Patna, delivered with Masai, for working managers and senior professionals, covering generative AI, agentic AI, AI governance and responsible AI without requiring coding. Its modules include Agentic AI for the Enterprise and Responsible AI, Governance and AI Business Models. The eligibility is minimum 1 year of full-time work experience, currently working in an organisation, and no coding background required. It grants no IIT Patna degree, credit or alumni status. Tools that vendors name are covered in top agentic AI tools and frameworks to learn in 2026.

Frequently asked questions

What are pilots in AI?

A pilot is a time-limited, small-scale trial of an AI tool on one real process, with a baseline measured beforehand and a decision date fixed at the start. It differs from production, which means the tool is relied on daily. Of the 200 organisations EY and CII surveyed in 2025, 23% were in pilot and 47% had multiple GenAI use cases live.

Can a manager create AI software without engineers?

For small internal tools, often yes. Microsoft said customers created over 3 million agents using Copilot Studio and SharePoint in its fiscal year 2025. Those are global figures from one vendor, and created is not production. Any tool that touches personal or company data should be approved by IT or information security before the team uses it.

How long should an AI pilot run?

The date matters more than the duration: it should be fixed at the start, long enough to see typical and awkward cases and short enough that the sponsor still remembers the question. A pilot without an end date tends to continue without a decision.

What success criteria should an AI pilot have?

Criteria should compare the pilot with the baseline on time, errors and cost per task, and name the risk controls that must hold. Gartner lists escalating costs, unclear business value and inadequate risk controls as reasons it predicts agentic AI projects will be cancelled, so those three belong in the decision.

Does a pilot need engineers?

Not always. Many first pilots use tools a manager can configure, with IT or security approving data access. Engineers become necessary when the work needs custom integrations.

Next step

The AI Officer Programme, Vishlesan i-Hub IIT Patna page sets out the curriculum and how to apply.