Customer support · Production

Dovetail improved word error rate by 36% with AssemblyAI to power customer intelligence

The problem

Customer conversations ingested by Dovetail are inherently messy and unstructured — full of overlapping dialogue, informal phrasing, and unique accents — making accurate, diarized transcription a prerequisite for delivering reliable customer insights.

Workflow diagram · grounded in source
1
Customer data ingestion
trigger
“It ingests mountains of feedback from interviews, support calls, product reviews, and more”
2
Speech-to-text transcription
ai_action
“AssemblyAI offers both asynchronous and real-time (streaming) speech transcription, as well as speech understanding models, at industry-leading accuracy built on cutting-edge AI research and accessible through a simple API”
3
Speaker diarization
ai_action
“~10% improvement in speaker diarization (speaker labels)”
4
Insight generation
output
“transforming it into clear, shareable insights that drive decision-making across organizations”
Reported outcome

Switching to AssemblyAI delivered a 36% improvement in word error rate and approximately 10% improvement in speaker diarization, along with a notable increase in customer sentiment tied to transcript quality.

Reported metrics
word error rate (WER) improvement36%
Speaker diarization improvement~10%
Customer sentimentNotable increase in customer sentiment tied to transcript quality
Reported stack
AssemblyAI
Source
https://www.assemblyai.com/customers/dovetail-customer-story
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Switching to AssemblyAI delivered a 36% improvement in word error rate and approximately 10% improvement in speaker diarization, along with a notable increase in customer sentiment tied to transcript quality.

What tools did this team use?

AssemblyAI.

What results were reported?

word error rate (WER) improvement: 36%; Speaker diarization improvement: ~10%; Customer sentiment: Notable increase in customer sentiment tied to transcript quality (source-reported, not independently verified).

How is this customer support AI workflow structured?

Customer data ingestion → Speech-to-text transcription → Speaker diarization → Insight generation.