Industries

Lead generation for manufacturing companies, run by an AI agent on LinkedIn

A US investment casting foundry ran Sellinger on its president's LinkedIn. The agent asked engineers one real question about their parts, followed up every five days and handed every hand raise to the president. Of 879 engineers and buyers messaged, 293 replied (33%) and 42 conversations ended with a buyer side outcome.

Published by Sellinger AI. Numbers and settings checked on 19 October 2026 against the fact register. Report an error.

engineers and buyers messaged after connecting, from 3,345 connection requests: 879
replied, 33% of those messaged (data date 9 October 2026): 293
buyer side outcomes: quote requests, drawings, NDAs, referrals and calls, 4.8% of those messaged: 42
of all replies came from a follow up, not the first message: 44%

Who the agent writes to in manufacturing

Industrial buyers are design engineers, sourcing managers and quality leads at original equipment makers. They ignore vendor pitches, but they answer a peer who asks a sharp question about their parts. That is why the foundry sent from its president's own profile, and why the agent never pitched.

The target list was US manufacturers that use cast parts: valves and pumps, heavy equipment, aerospace and defense controls, oilfield and agriculture. Sales, marketing, HR, consultants and other foundries were excluded in the rules.

Where the 879 prospects came from
Lead source	Messaged	Replied	Buyer side outcomes
Engineers and buyers at named target makers (LinkedIn search by company and title)	449	27.6%	24 (5.3%)
Title and keyword searches	210	30.5%	8 (3.8%)
People engaging with competitor foundries' pages (intent signals)	141	58.2%	6 (4.3%)
Imports and the president's own network	79	29.1%	4 (5.1%)

Intent signal lists replied twice as often, but 57% of those replies came from recruiters, vendors and peers. Named company searches produced 24 of the 42 outcomes.

The playbook the agent ran

The full configuration is on the connect, ask one question, follow up every five days playbook. In short:

1
Day 0
Connection request
Usually without a note (149 of 3,345 had one). 27% were accepted.
2
On accept
One diagnostic question
A median of 33 words, no pitch and no link.
3
Day 5, 10 and 15
Bump, new angle, soft close
One line plus an easy out, then tariffs or lead times, then a polite goodbye.
4
Any reply
Diagnose, then hand off
Part, alloy, volume, timeline, then email, drawings, an NDA or a 15 minute call.

First message, as the agent wrote it to a purchasing lead at an aerospace components group: "Given [company]'s focus on mission critical components for aerospace and defense, I'm curious. Do you ever find it a challenge to source complex investment castings that come fully machined and finished from a single, reliable partner?" The prospect answered after the first bump, five days later: he was about to start a request for quotation.

Results, with their bases

8.8% of all connection requests ended in a reply (293 of 3,345), and 1 in 7 replies became a buyer side outcome (42 of 293).
Follow ups lifted the reply rate from 17.2% to 30.9%. Across 1,699 follow up recipients, 7 first replies were negative.
The median reply time to a prospect was 2.3 minutes since May 2026, and 71% of the messages that arrived in the evening or at weekends were answered before the next business morning.
Of the 42 outcomes, 23 asked for a quote, information or an NDA, 12 referred the foundry to the right buyer, 6 agreed to a call and 1 booked one.

What did not work

A short email test in June (47 leads) got no replies. For this audience LinkedIn won.
Volume fell after July because the lead list ran out, not because the agent slowed down.
Capability questions needed exact guardrails. Until the president wrote down what the foundry does and does not pour, answers on alloys were loose.
Many objections were polite exits: an established overseas supplier, volumes too small, or not my department. The last one produced 14 referrals to named buyers.
Conversion here means a buyer raised a hand. What happened after the hand off to email is not visible to us, so no deal value is claimed.

Compliance and fit

The messages went to US business contacts about a business supply question, on the sender's own LinkedIn account, within the account limits described on the LinkedIn agent page. LinkedIn's own terms apply to any automation, and no tool can guarantee an account is never restricted. The approval switch described in the fact register lets a person read messages before they send.

This page does not fit teams that need to sell to purchasing departments that never use LinkedIn. In that case start with a small email test and read the named owner letter playbook.

Common questions

Can an AI agent really prospect for a manufacturing company?

In the Lamothermic study it did the prospecting work, and a person closed. The agent asked engineers one question about their parts, followed up and answered within minutes. When a buyer raised a hand, the agent collected an email address and the president took over.

Which lead source worked best for manufacturing?

Named company searches produced 24 of the 42 buyer side outcomes. Intent signal lists replied more often but mostly from people who were not buyers.

Does it work by email for manufacturers?

A June email test of 47 leads got no replies in this study. One small test cannot settle the question, so test email separately before relying on it.

How much does it cost?

LinkedIn is priced per agent. The current prices are on the pricing page.

Sources and method

Every number comes from the linked case study, which states its base, its data date and what did not work. Source: Lamothermic case study, Sellinger production data for that workspace, read only, snapshot 9 October 2026. All 293 replied threads were read by hand.