A Netherlands-based AI automation agency: Selling automation in a market where everyone sells automation

Client
a Netherlands-based AI automation and development agency
Product
AI automation agency, B2B services, Netherlands
Engagement
Ongoing since August 2026

A founder-led AI automation and custom development agency in the Netherlands that rebuilds one manual workflow at a time inside the systems a company already uses.

About the client

Its buyers are mid-sized, non-technical companies in the United Kingdom, Ireland, the Netherlands, Germany, Denmark, Canada and Australia that still run admin and compliance work by hand.

A founder-led AI automation and development agency in the Netherlands came to us in August 2026 with one LinkedIn profile, a seven-message sequence and a reply rate under 1 percent. The founder had tried to automate the outreach himself. We reviewed the campaigns, said the lists and the copy were not good enough, and rebuilt both.

Target markets: mid-sized, non-technical companies in the United Kingdom, Ireland, the Netherlands, Germany, Denmark, Canada and Australia that still run admin, coordination and compliance work by hand.

The Opportunity

The agency sells one workflow at a time: the manual work between the systems a company already uses, rebuilt as an automation, with nothing replaced or migrated. The buyers exist. Mid-sized companies still hire people to do data entry, follow-up and reporting by hand, and they say so in public. A job posting describes the manual work, the team that does it and often the system it happens in.

Nobody was reading that signal. Lists were built in LinkedIn Sales Navigator from industry and headcount filters, which find company types, not manual work. The founder set one metric for the rebuild: reply rate, not sales. A reply means the message touched a real pain and reached the right person.

  • Job postings are a public, dated record of manual work a company wants done.
  • Mid-sized companies without an internal tech team cannot automate that work themselves.
The Challenge

Everyone sells automation now. The founder's own inbox is full of the same pitch, and most of it reads like the same AI-written copy. Buyers have learned the pattern. Openers like "saw you do" or "noticed your company is growing" are read as automated in the first line.

The agency's own first sequence was part of the noise: seven generic messages with a 70 to 80 percent efficiency claim and no company-specific line, sent by hand from Sales Navigator without research. Replies stayed under 1 percent.

  • Real research costs credits. One company can post 50 jobs, and running AI on every posting burns a Clay plan on duplicates.
  • Personalisation had to come from the prospect's own words, not from a compliment or a guess.
Our Approach
Start from the pain, not the industry

We built a Clay workbook that searches LinkedIn job postings for keywords that signal manual work. The first export in August 2026 held 13,076 companies, and a second in November 2026 added 17,106. Together that is about 30,000 companies and about 105,000 job postings reviewed across seven countries.

One row per company, before any paid step

A company that posts 50 jobs appears once. The workflow removes duplicate companies first, so no credit is spent twice on the same firm.

Qualify before researching

Clay checks the employee count and keeps mid-sized companies. An AI column then judges whether the company is tech-savvy enough to automate on its own. Only companies that are not move forward. Medical and health companies are excluded.

Find the people before spending on research

Clay people searches for senior and head-level roles returned 39,881 candidates. A company with no reachable decision maker stops here.

Research what the message needs

AI columns on the OpenAI API read the posting and the website. They return the core work of the business, three manual tasks, the system in use and the steps a trigger event sets off by hand. A last AI column picks the three people most likely to own both the decision and the manual work. The final table holds 6,521 people.

Write with the research, not about it

The first message asks how much of the team's day still goes into their three named tasks instead of their core work, and says their existing system stays. No "saw you do", no compliment, no claim about results. Four short steps: a connection request with no note, then three messages. The last one asks to be pointed to whoever owns the task.

Send from four profiles with human pacing

The sequence runs in lemlist from four LinkedIn profiles, and Clay pushes each person with their variables straight into the campaign. We tested Prosp.ai in parallel and dropped it in October: it checked for replies every four hours, so follow-ups fired after someone had already answered. lemlist checks every 30 minutes. Expandi handles list exports from Sales Navigator.

Work the replies

Three weeks in, replies arrived faster than the founder could answer them. We added tags that separate "not interested" from "not the right person", so the second group is asked for the right contact, and one inbox with a morning digest. A weekly review looks at the replies and changes one thing at a time. Clay credits ran out in October 2026 and the account had to be rebuilt before the next segment could run.

The Results

<1% to 4.5%

LinkedIn reply rate, Aug to Nov 2026

25.8%

Connection requests accepted

30,000 to 6,521

Companies screened to people worth a message

Numbers from the client's lemlist team, AI and workflows campaign, 29 August to 8 November 2026.

  • 2,433 people contacted so far from four LinkedIn profiles, out of 6,521 shortlisted.
  • 599 connection requests accepted (25.8 percent).
  • 110 replies, 4.5 percent of people contacted, up from under 1 percent in August 2026 on one profile with generic copy.
  • 19 replies marked interested. The best single profile reached 7.2 percent replies in September 2026.

A second campaign for software companies ran with lighter research over the same period: 14.7 percent acceptance and 2.1 percent replies on 757 people contacted. Less research, fewer replies. We do not claim meetings, deals or revenue: replies and research are our part, closing is the client's. Beyond the numbers, the engagement created:

  • A research engine the agency owns: job posting in, qualified company, named manual tasks and system, three people, a message only that company could receive.
  • A credit-safe Clay build in which the expensive steps run last.
  • A reply process with tags, one inbox and a weekly review, so no reply is left unanswered.
The Outcome

In three months the agency went from one profile and generic copy to four profiles and a message built from each prospect's own job posting. The reply rate went from under 1 percent to 4.5 percent. At the first review the founder said the agency finally had an engine rolling.

The result is a way to be heard in a crowded category without a bigger budget. The specificity comes from the prospect's own posting, which a competitor with a larger ad budget is not reading.

The same research now feeds an email version of the sequence, drafted in lemlist for 2,577 contacts and not yet launched, and a smaller lemlist test that uses its AI personalisation for higher-value offers, too early to report.

Research first, copy second is how we run every engagement. See what a monthly engagement costs, read how we built a cold email engine for a programmatic advertising platform, or book an intro call.

30,000Companies screened
6,521People worth a message
2,433People contacted so far
599Connections accepted
110Replies
4.5%Reply rate, up from under 1%
Shortlisted
6,521100%
Contacted
2,43337.3% so far, still running
Accepted
59925.8% of requests
Replies
1104.5% of contacted
Interested
19marked interested