Why AI Pilots Stall, and the New Role HR Needs to Hire For

Most large companies have plenty of AI pilots, but far fewer get them into everyday work. Our report shows how one operations organization fixed that by adding a new role, and why the decisions that make that role work belong with HR.

The report follows an operations organization at the core of a multibillion-dollar enterprise. At the start of 2025, it had heavy automation investment and dozens of AI pilots. More than 60 improvement initiatives were in flight, and fewer than a third had shipped anything a frontline team used. Within 12 months of bringing in Forward-Deployed Engineers, 11 of 14 builds were in production.

The organization is an illustrative composite. Its figures are representative and aligned with published industry benchmarks.

Why AI pilots get stuck

The people who knew the work and the people who could build were far apart. Frontline experts knew every exception, but they could only write it down in documents. Engineers who could build sat three layers away, working from a central IT backlog.

By the time anything was built, the need had changed or the sponsor had moved on. Fewer than three in ten pilots reached production.

The real process was also different from the documented one, full of workarounds and side spreadsheets. AI built to the documented process failed on the steps nobody had written down.

The role that closed the gap

The fix came from the software industry: the Forward-Deployed Engineer (FDE). An FDE is an engineer who sits inside the team where the work happens and builds working tools with the people who do the work.

The organization placed five FDEs across a workforce of several thousand, one in each major process area. Each one followed three rules:

• Ship a first working improvement within 30 days.
• Map how the work is really done before building anything.
• Count success by what teams actually use, never by demos.

What changed in the first year

Eleven of 14 builds reached production, up from fewer than three in ten before. The results helped revenue as well as cost:

• Order processing time fell by about 40% on covered order types.
• Billing disputes fell by about 25%.
• About 70% of routine deal approvals were cleared automatically within policy, and the approval queue time was halved.

By mid-year, teams were competing for FDE time.

Why this is an HR decision

The FDE model looks like an engineering program, but it succeeds or fails on people decisions. Operations leaders tend to discover them late, so HR should be in the room from the start.

Who to hire. The organization's first hire was technically excellent but avoided working with frontline teams, and it failed. The best FDEs can sit with a team for a day and pull out the unwritten rules while keeping its trust. A shadowing exercise reveals more than a coding test.

Where to find them. External hires mean competing with AI companies at the same pay. Draup data shows 13,511 FDE job postings in Northern America between 2023 and 2026. Internal candidates such as RPA developers and business analysts already have the process knowledge and the trust, but they need engineering training. A strong first group mixes both.

What happens to the team. When an agent speeds up a team's work, its jobs shift toward handling exceptions and customers. Frontline teams share their real process only when they know what happens to their roles, so decide that early and tell them.

Where the career leads. Without a clear next step, FDEs leave and take their knowledge with them. Plan the career path and pay before the first hire.

What to agree with your COO before the first hire

• One FDE per major process area, reporting to a neutral transformation leader.
• A short process-mapping phase before every build.
• Clear rules on what the AI can and cannot do, agreed once with finance, security, audit and HR.
• HR involved at design: the role profile, where to hire from, and the plan for changed roles.
• Success measured by what teams use.