Blog

Nov 15, 2026

5min read

Smiling woman in a navy blazer with city buildings behind herGeargina Tan

Why Do AI Pilots Stall, and What Closes the Gap?

Team seated around a white table looking at a shared screen, with sticky notes on the wall behind them

AI pilots stall because the demo runs on clean data and a watchful engineer, while production runs on your messy systems and your busy people. The gap closes when engineers work inside the business, wiring the AI into the tools, handoffs and approvals your team already uses, and stay until it runs without them.

How common is it for AI pilots to stall?

Very. MIT's NANDA initiative published The GenAI Divide report in July 2025. Its headline: despite $30 to 40 billion of enterprise investment in generative AI, 95% of organizations are getting zero return.

Further down, the report found that only 5% of custom enterprise AI tools reach production.

Read that as a statement about delivery, not about models. The models were fine. The tools never made it into the daily work of the people who were supposed to use them.

Why does a pilot work and the rollout fail?

A pilot is built to succeed. Someone picks a tidy use case, cleans a sample of data, and watches the output every morning. It works, because a clever person is standing next to it.

Production removes all three props.

The data is the real data. For a car dealer group, that means customer chats in WhatsApp, quotes in a spreadsheet, stock in the DMS, and leads in a CRM nobody trusts. The engineer who watched the pilot has moved on to the next project. And the people who now have to use the thing have a full day already, so anything that adds a step gets ignored.

Pilots die in the join between the AI and the business. Nobody owned that join.

Why can't we just hand it to the IT team?

You can. Some do it well. But look at who's usually in that room.

Your IT team knows your systems and keeps them safe. They weren't hired to sit next to a sales rep and learn how a quote gets approved on your floor. Your vendor knows its product. It doesn't know your handoffs.

Someone has to hold both halves. That's the job that goes missing, and it's the reason so many pilots look finished on a slide and dead on a Monday.

What is a forward-deployed engineer?

A forward-deployed engineer is a software engineer who works inside the customer's business instead of at the vendor's desk. They learn how your team works, build the integrations and AI workflows that fit it, and stay through rollout and the first months of use.

The person who writes the code sits where the problem is, so the feedback loop is a conversation instead of a ticket.

It's the opposite of a handover. A handover assumes the hard part is the build. In production, the hard part is the fit.

Does embedding engineers raise the odds?

The evidence points that way, with a caveat I'll state up front.

The MIT report found that external partnerships with customized, learning-capable tools reached deployment about 67% of the time, against about 33% for internally built tools. The authors say these figures are self-reported and may not account for every confounding variable. They also say the size of the gap was consistent across the people they interviewed.

So treat it as a strong signal, not a law. It matches what you'd expect. Partners who build the same thing repeatedly have seen the failure modes. Partners who sit with your team see your specifics.

What does it look like in a car dealership business?

Picture a dealer group with five rooftops. Each team chases leads on its own WhatsApp threads. The group's CRM holds a fraction of what's in play.

An AI pilot that summarises one rep's chats is easy. Making the whole group's chats land as structured leads, matched to the right stage of the sales pipeline, routed to the right team, and pushed into the DMS when a deal closes is a different job. It touches systems, people and approvals.

That's the work WTFox's Custom Enterprise is built for. We map how a lead reaches you today, from first message to signed deal. A WTFox engineer then works with your team to connect the platform to your DMS and master CRM, and to build the workflows no off-the-shelf CRM ships with. Your team gets hands-on ChatGPT and Claude workshops as well, so the people using it aren't learning it cold.

The platform underneath is the WTFox CRM that already captures WhatsApp, Telegram and email conversations. The engineer's job is the fit.

How do you judge a partner before you sign?

Ask for four things.

  1. Who is the named person working with your team, and how many days a week?
  2. What does "done" mean in writing, and who signs it off?
  3. How does the AI's work stay reviewable by a human once it's live?
  4. What happens in month three, when the engineer isn't in the room?

A pilot with answers to those four will survive contact with production. A pilot without them is a demo.

Where should you start?

Start with one workflow and one owner. Pick the process where leads or deals leak today, and map it end to end before anyone writes code.

If you run a dealer group and want that mapping done with you, Custom Enterprise starts with a free 1:1 consult. If you'd rather see the product first, the homepage shows what the CRM does. Anything else, get in touch.

Prospect Smarter.
Book Faster.

Start Free