Nexly AgentsBlog

Nexly Agents › Blog › Manufacturing

Manufacturing

The installer calls at 5:40 pm. The manufacturer answers on Monday

Published 2026-10-08·4 min read·Author: Jakub Milewski, Nexly Lab

Photo: Unsplash

Do not start automating the source of your revenue with AI. Start where your company receives the calls nobody is there to answer.

Friday, 5:40 pm. An installer is standing on site at a controller that has been showing "no communication with module" for an hour. The customer is standing next to him. He calls the manufacturer, support works until 4 pm, voicemail picks up. The answer is in the manual, somewhere in the middle of several hundred pages, and he does not know that. Even if he knew, he would not search for it on his phone. On Monday morning the manufacturer's engineer will spend an hour with him on the phone, starting with "please describe what you see".

That is what a week looks like at a manufacturer whose equipment is installed and serviced by third-party installers: heat pumps, building automation, machinery, controllers, photovoltaics, valves. The support department takes hundreds of questions a month. A large share have an answer in the documentation, only nobody knows on which page. The rest of the knowledge sits in the heads of two or three engineers. In the United States, a company that answers phones for businesses counted on its own data that 47% of calls come outside nine to five on weekdays. In Poland I know of no measurement I would trust, but every head of support knows the Friday voicemail.

When such a company hears "deploy AI", the first thought goes to the configurator, product selection, quoting. That is where revenue is made and what is easy to show the board. It is also where a mistake costs the most, and where the data to train the system still has to be built. MIT counted last year that about half of generative AI budgets go to sales and marketing, while the largest documented savings in the report were in customer service and document processing. Those savings came without headcount cuts, from cutting external spend. The OECD added the voice of small manufacturers: they cannot find a vendor that builds anything for a company their size.

Start where the knowledge is already written down. Manuals, datasheets, service bulletins, firmware notes. An agent trained on that documentation answers the installer at 5:40 pm the same way it does at ten in the morning. It answers only from the documentation, and what the documentation does not cover it writes down together with a description of the situation on site. On Monday the engineer gets a complete ticket instead of a call from zero.

In a technical industry three conditions come with this and none can be dropped. The installer has to know he is talking to a machine; from August 2026 the AI Act requires it. Settings that affect people's safety or the warranty always go to a human, even when the answer is in the manual. And someone in the support department has to own the agent: read the report, fix the documentation where it is silent. MIT states plainly that tools which do not learn from use stay in pilot, because people describe them in one sentence: "useful for a week, then it repeats the same mistakes". Beyond that one person, nobody changes how they work.

After a few weeks the more interesting part begins. Every conversation from the field is kept and someone reads it every night, except that someone is an agent. In the morning the support department knows which error message came back five times this week, and that is a signal for production, not for service. It knows which questions the documentation is silent on. It knows that designers and wholesalers are asking about selection and parameters. The agent was meant to handle service, but it recognises a sales signal in that, and the salesperson gets the contact details in the morning together with the whole conversation.

After a few months the company knows agents from the inside. It knows from its own conversations where the agent helps and where it gets things wrong. It is the least invasive way to get comfortable with AI, because beyond one person nobody touched any process. Only from there does the company look for the next areas: complaints, product selection, quotes. The source of revenue comes last, once the company knows what it wants from AI.

Ask your head of support one thing: how many of last week's calls were about something that is in the manual, only nobody knew where. If the answer is "about a third", you have the place to start. Write that number down and come back to it after three months. Thomson Reuters found that only 18% of companies check whether AI paid off. The rest cannot say a year later whether it worked, and the project fades out.

Want to see this on your own documentation? Enter your website address and e-mail in the demo form. An agent for a manufacturer learns from manuals, not from the website, so after you sign up we will get in touch for the materials, and your head of support can ask it his own questions.

Sources

See the AI agent on your own website

Give us your website address and e-mail, and in 20 minutes you get an agent trained on your offer, ready to be quizzed.

Get a free demo

Read also

Your people already use AI. Just not the one the company bought Marketing starts AI with content. Meanwhile the customer is waiting for an answer Don't start AI with a revolution