Case study

Somes Oy: a lead list that sorts itself

Somes Oy is a Finnish video and social media agency that needed a fast way to tell which companies on its list are worth contacting first. I built them an AI system that watches the agency's own CRM, finds and verifies each company's social channel and writes the result straight into the sales view. It runs on its own in the cloud and the client owns it outright. From kickoff to handover took about four weeks.

  • Client Somes Oy
  • Industry Video and social media
  • Service Signal monitoring
  • Duration
  • Handover
Book a free discovery call
Starting point

What was the starting point?

Somes makes video and social media marketing for companies. Finding new clients starts with a list, and the list is long. Before anyone gets a call, you need to know whether the company has a social channel at all and whether it is alive. Without that, the salesperson is calling blind.

Until now a person did that research. One company at a time, browser open, result typed into the CRM by hand. The work cannot be skipped, but it eats exactly the hours that should go to clients.

Somes knew what it wanted. A list that tells you, by itself, who to call first.

Solution

What was built?

1

Watches the CRM

When a new company lands in the CRM, or an old row needs refreshing, the system picks it up automatically. Nobody starts anything.

2

Finds the company's social channel

The system searches by company name, collects candidates and gathers more detail on the most likely ones.

3

The AI chooses, it does not guess

It gets a list of candidates and either picks the right one or says that none of them is the company's own channel. It is not allowed to invent anything outside the list.

4

Measures whether the channel is alive

The date of the latest post and the view counts tell whether the channel is active or has gone quiet.

5

Writes the result into the sales view

The result goes into five agreed fields in the client's own CRM. The sales view sorts itself from hottest to coldest, and nobody has to open a separate tool.

The same principle works with any list. The register can be a CRM or a spreadsheet, and the thing to check can be a website, LinkedIn or job postings instead of a social channel.

Day to day

How does it work day to day?

Runs on its own

Runs from the cloud at regular intervals on weekdays. A successful run is silent. If something fails, the client gets an email about it.

Uncertain cases go to a human

When the AI is not sure, the company gets the status Check and a person looks at it.

Never overwrites a human

The system never touches a value corrected by hand, and it never writes anywhere except the five agreed fields.

The client owns everything

Code, scheduling, documentation and accounts are all in Somes Oy's name. My access can be removed with one click and nothing stops.

Feedback

Client feedback

Working with Ilmari went really well from start to finish. The project stayed on the agreed schedule, and Ilmari was able to explain even the more technical parts clearly, so that how they work and what they are good for was easy to understand. He actively took on every question that came up and always followed through on the things that did not have a ready answer straight away.

The solution Ilmari built has made our client acquisition concretely more efficient. We operate in a fairly small market, so being able to identify the most promising clients as efficiently as possible matters to us. The automation significantly reduces manual research and processing, so we can spend more time on actual client work and sales.

Somes Oy Written client feedback, August 2026. Translated from Finnish.
Timeline

How did the project run?

1

Kickoff Week 1

We went through the CRM and agreed on two use cases and the five fields the system is allowed to write to.

2

First version Weeks 1 and 2

A read-only version whose results we reviewed together before anything was written to the CRM.

3

Refining on real data Weeks 2 and 3

The selection logic was rebuilt when the first model turned out to guess too much. Results were checked by eye, row by row.

4

Handover Week 4

The whole system moved into the client's own environment, with a user guide written in Finnish. It has run without me ever since.

Updated

Frequently asked

Frequently asked questions

Who owns the system?

The client. Code, scheduling, accounts and data belong to Somes Oy, and the handover included a user guide in Finnish. I build every system so that it works without me. Maintenance is agreed separately, if it is wanted.

What happens when the AI is not sure?

The company gets the status Check and a person looks at it. The AI may choose from ready candidates, but it may not invent an answer or write over a value that was corrected by hand.

How long did the project take?

About four weeks from kickoff to handover. The first working version was ready in a couple of weeks. The rest went into refining on real data and the handover itself.

Would the same work for us?

If you have a list of companies and someone checks things about them by hand before reaching out, probably yes. The system is built into the register you already use, whether that is a CRM or a spreadsheet, and the source to check can be a website, LinkedIn or job postings instead of a social channel. The discovery call is free and the price ranges are public on the consultancy page.

Start here

Do you also have a list someone goes through by hand?

Thirty minutes, no obligation. Tell me what you need to know about the companies on your list before reaching out, and we will work out together whether it is worth automating.

Book a free discovery call