Fallstudie · B2B Services · November 2025

Lead-data enrichment for the sales team of a workplace coffee and water supplier

To a working proof of value
3 days
In production
11 months
Companies enriched
54,000+
Automated data enrichmentLLM pipelineAPI integrationManaged AI subscription

Client

#

A supplier of coffee machines and water dispensers for workplaces across German-speaking Europe. Its sales team decides which companies to approach, and with which offer, from a purchased company database.

Challenge

#

A dental practice, a logistics company and a software studio each need something different from the same catalogue. The sales team’s main signal for who hears which offer is each company’s industry code.

The purchased database left many companies without one. Filling the gap by hand meant finding the company, reading its website and picking the right entry from the official German classification of economic activities, five levels deep with hundreds of categories at its finest level – several minutes per company, across tens of thousands of companies.

And the codes the database did contain had never been checked.

Our approach

#

Within three days of the first call, a first version of the AI agent was running on a sample the client chose, and we compared its results with the purchased codes and with records checked by hand. Four principles shaped the agent:

Research the way a person would. The agent starts from a company’s name and address, finds the real business behind it, reads what that business publishes about itself and only then assigns a code. When the evidence is unclear, it flags the company for a person instead of guessing.

A fixed list, not free text. The model can only choose from the official classification, so every answer is a real category the sales team already works with.

Names, not numbers. To a language model, a numeric code is an arbitrary string; the words behind it carry the meaning. So the model chooses between category names, and the service maps its choice to the code afterwards.

Inside the CRM. The client’s CRM sends records to the agent, and the industry codes are written back into the fields the sales team already uses. Nobody has to open a separate tool.

Once the results met the sales team’s expectations for accuracy, we deployed the agent as a managed service – a one-time integration fee, a price per record and a monthly service fee – with separate development, test and production environments.

A new lead with only a company name and address arrives in the client’s CRM. The AI agent finds the real business behind the name, reads what it says about itself and chooses a category from the official classification. The industry code is written back to the company record in the CRM; cases with no clear match are flagged for a person rather than guessed.
Every new lead is researched the way a person would, coded from the official classification and written back into the CRM. Unclear cases are flagged rather than guessed.

Results

#

In production for eleven months. The agent has enriched more than 54,000 companies, and nearly every company in the client’s CRM now carries an industry code. New leads are enriched as they arrive, so the sales team works from complete records instead of researching companies by hand.

What transfers

#

Check the data you bought. A purchased database is an accuracy claim that is rarely tested. A hand-checked sample shows how far to trust it.

Automate the clear cases, flag the rest. An agent that hands over what it cannot decide earns more trust than one that always answers, and the sales team keeps the final word on the edge cases.

Prove it in days, then run it as a service. A three-day test on the client’s own sample settled the question before any integration work, and the same agent is still running today.

Is your CRM missing information that sits on your prospects’ own websites? We can fill it in automatically and run the service for you. See Agents-as-a-Service, or Start a conversation.

Als PDF herunterladen

Soll KI in Ihrem Unternehmen arbeiten?

Sagen Sie uns, welchen Prozess Sie verbessern möchten und wo die Daten liegen. Ganz unverbindlich.