Client story

Every call scored 0–5 — how AI turned phone conversations into measurable data

Sales was conducted mostly by phone, and the most critical moment of the process left no data behind. We integrated the phone system with the CRM so that the recording, transcription, and quality score land in the client record, while a language model generates a note and a suggested reply. We also solved the correspondence personalization problem by simplifying rather than expanding template logic.

← All client stories

The challenge

The most critical moment of the process left no trace in the data

Sales was conducted predominantly by phone, and the conversation was the key moment of the process. At the same time, there was no record of what happened in it. The consequences spanned three areas. Management had no basis to assess the team's work beyond sales results — a lagging measure dependent on many factors. New hires learned without access to examples of well-conducted calls. And the content of arrangements was lost — notes were written irregularly, and when returning to a client after weeks, context had to be rebuilt from scratch. Outgoing correspondence also required manual personalization, including correct grammatical forms depending on the recipient, which discouraged the team from using prepared templates.

What we did

Telephony, CRM, and a language model in one flow

  • Integrated the phone system with the CRM — incoming numbers match to client records, and the contact history receives the recording, transcription, and call quality score on a six-point scale.
  • Built an analytics layer over transcriptions — a language model processes the call transcript and generates a system note and a suggested reply to the client.
  • Designed cyclical feedback for salespeople — periodic summaries with conclusions, limited to calls below a defined quality threshold.
  • Solved the correspondence personalization problem through simplification — instead of building an elaborate system of variables and conditions, we pass the full template plus recipient data to the model and let it handle grammatical form matching.
  • Launched automated contact with inactive clients — after a defined period, the message is sent independently unless the salesperson takes action in the system.
  • Prepared an extension for the mobile channel.
The result

What changed

Every call leaves a trace in the client record: recording, transcript, and quality score. Call summaries turned out to be the most valuable element in practice — useful even when the transcription contains inaccuracies. Management has a quality measure independent of sales results, available on an ongoing basis. Outgoing correspondence is personalized automatically, without maintaining elaborate template logic.

“In phone-based sales, the most critical moment usually leaves no data behind. The decision on personalization: passing the full template with recipient data to a language model gives the same result at far lower maintenance cost — provided the output is verified before sending.”

Maciej PuchałaManaging Director, Celvaron
Next step

Let's write down how your company actually works.

  • 45 minutes, and no preparation needed
  • An honest read on where you sit today
  • The first process worth automating — and why that one
Maciej Puchała
Maciej Puchała Founder +48 729 0 87 87 0 contact@celvaron.com