AI agents
Contact center speech analytics
CONTACT.WORK speech analytics transcribes conversations and scores every call against your company's criteria. Managers see who skipped a required step, why customers call, and where sales fall through, and the AI writes a short summary of each call. Scoring looks at the meaning of what was said, not just keywords. A person reviews disputed cases.
Summarydelivery moved to FridayAIRating86 of 100scorecardMissedrecap at the endstep 5TopicdeliveryautoTools
What speech analytics does
Runs on every call, no sampling.
Examples
What it solves in practice
Five common situations and what the manager does in each.
Find the call behind a complaint in minutes
A customer complains that an agent was rude or promised a discount. The manager looks the call up by the caller's number or a phrase from the complaint, jumps to that line in the transcript, and sees the scorecard result beside it. If the complaint holds up, searching for the same phrase reveals whether it happened on other calls as well.
- Search by phone number or phrase
- Jump to the right line
- Similar cases
Call2:32 PM, by numberfoundLine02:41 discount promisedviolationRating61 of 100scorecardSimilar7 more callsfor the monthThe report shows where you lose customers
Call volume is the same, but deals are down. The stage report shows that the same share of calls still reaches the offer, but reps stopped handling the price objection. The manager opens those calls, compares them with calls from top performers, and then walks the team through the answers that work.
- Share of calls at each stage
- Rep comparison
- Winning answers from top reps
Needdone in 92%no changeOfferdone in 71%no changePricehandled in 48%−17%Top rephandles in 85%referenceA growing topic shows up in fresh calls
By morning, there are twice as many calls about website payments. Speech analytics spots this in fresh calls, and the manager hands the problem to the people who can fix it without waiting for the weekly report. While the website is being fixed, the frequent question can go to the IVR or an AI agent.
- Topic growth within a day
- Sample calls
- Frequent question goes to an AI agent
Online paymentup ×2since 9:00 AMExamples24 callslistenSentto web developerstaskTemporaryIVRreplyEvery call a new hire takes is scored
A new agent's calls are checked from their first day on the line, with no manual listening. Their mentor sees which steps they skip most often, reviews specific calls with them, and plays recordings of how an experienced colleague handled a similar case. New hires see their own scores and reviews, and within a couple of weeks it's clear whether the mistake is gone.
- Every call scored
- Agents see their own scores
- Examples from experienced colleagues
Rating54 → 78over 2 weeksMost viewedno recappost-mortemExampleGreg P.'s calllistenCallsall 312reviewedThe right phrases said, nothing extra promised
A bank, insurer, or clinic needs agents to give a recording notice (required in all-party consent states), get consent, and avoid promising anything the company doesn't guarantee. Speech analytics checks this on every call and shows who has violations and how many, so they get fixed before an auditor asks.
- Recording notice
- Consent
- Prohibited promises
Bookingdisclosed98%Consentreceived95%Restrictionsapproval promised3 callsViolationsby 2 agentspost-mortemOne scorecard covers all five situations – no separate setup for each.
How it works
From recording to score
Speech recognition
You choose where speech is recognized
Recordings come from the platform's telephony or from your phone system. Each speaker must be recorded on a separate track (dual-channel recording) so the system can tell who said what. The platform's telephony is set up this way out of the box.
- Platform telephony or your phone system
- Dual-channel recording required
- English speech
00:04agent: greetingtrack 100:09customer: where's my ordertrack 200:31agent: confirmed addresstrack 1Bookingtwo tracksrequiredScoring
Counts what was said in the agent's own words
A step counts even if the agent phrased it their own way. Managers correct wrong scores, and those corrections are used to fine-tune the model.
- Meaning of what was said
- Manager corrections
- Learns from your calls
Phrasehow can I help?needsAlternativewhat happened?needsDisputedfixed by managerto trainingModellearns from your callsfine-tuningIntegrations
Where scores go
The call summary, topic, and score are sent to your CRM via API, and agent reports are built right in the platform.
- CRM
- Reports in the platform
- Review tasks
CRMoutcome and score in dealvia APIBitrix24call topic in recordvia APIReportsagent scoresin platformTasksreview for managerautoLaunch
How we launch speech analytics
- 01RecordingsWe connect the platform's telephony or your phone system and make sure each speaker is on a separate track.
- 02ChecklistWe set up your criteria: call stages, required and prohibited phrases, and weights.
- 03TopicsWe build the list of call reasons you need in reports.
- 04CalibrationWe compare the system's scores with your specialists' scores and refine the rules.
- 05LaunchWe turn on scoring for every call and the review of flagged calls.
- 06Fine-tuningWe fine-tune the model based on manager corrections and expand the scorecard.
Requirements
Requirements and pricing
- Recordings
- Each voice on its own track
- Language
- English
- Deployment
- Our cloud or your server with a GPU
- Cloud pricing
- $0.015 per minute, lower at volume
- Your own server
- No per-minute fees
- Implementation
- No license fees; our work is billed in support plan hours
Preparation
What to confirm before you start
Q&A
Frequently asked questions
How is speech analytics different from call transcription?
Transcription gives you only the text of the call. Speech analytics also scores the call, identifies the topic and the customer's mood, and shows why the customer said no and where a sale was missed.
How accurately does the system score calls?
Individual calls are sometimes scored wrong, usually because of background noise or vague scorecard wording. That's why scores are used to find recurring problems and pick calls for review, while managers make the final call on disputed cases.
Can you score calls against our scorecard?
That's how it's meant to work: we set up your scorecard with its stages, required phrases, and weights. There are no license fees; installation, migration, and setup are done by our engineers and billed in support plan hours.
How soon after a call is the score ready?
Shortly after the call ends, so managers learn about a violation the same day, not during a scheduled review weeks later.
Which calls are scored?
All of them: inbound, outbound, and internal calls between employees. You don't need to pick calls for review.
Does speech analytics work with our phone system?
It does, as long as your phone system records the two speakers on separate tracks – otherwise the lines can't be separated. If calls are currently recorded on a single track, recording needs to be reconfigured. Our own telephony records on separate tracks by default.
Can you analyze old recordings?
As long as the recordings still exist. You set the retention period for recordings and transcripts: extend it ahead of time, and you can score calls retroactively.
Does it check chat messages too?
Chats are scored by AI chat analytics using the same kind of scorecard, adapted for messages. It is added separately and billed per conversation.
Can we process recordings only on our own servers?
You can. Recordings are then transcribed and scored inside your company with no per-minute fees. You need a server with a GPU.
Can scores be sent to our CRM?
The call summary, topic, and score are sent via API to the deal or the customer record.
Can you tailor it to our business?
The model is fine-tuned on your company's recordings and on manager corrections to disputed scores. The more nuances your business has, the longer setup takes.
Will it replace our quality team?
No. You still need QA specialists, but instead of picking calls at random, they review flagged calls and coach agents.
How do we measure ROI?
Compare the month before launch with the month after: how many hours your quality team spent listening to calls, what share of calls had violations, how many calls led to a deal, and how often customers called back about the same issue.
How much does speech analytics cost?
The cloud option starts at $0.015 per minute, with lower rates at higher volumes; call length is counted by the second. On your own server, there are no per-minute fees.
Glossary
Terms on this page
- Speech analytics
- The system analyzes recorded calls automatically: topics, script adherence, customer emotions. Learn more →
- Dual-channel recording
- Each speaker is recorded on a separate track. Learn more →
- Cloud deployment (cloud)
- The platform runs on the vendor's servers, and agents connect over the internet. Learn more →
- Model fine-tuning
- An existing model is trained further on company materials so it speaks the company's language. Learn more →
- CRM – customer relationship management
- Where a company keeps its customers, deals, and full communication history. Learn more →
- API (application programming interface)
- The way programs exchange data with each other without a person in the loop. Learn more →
- IVR (interactive voice response)
- A phone menu: the caller hears a greeting and picks the right department. Learn more →
- Speech-to-text
- Turns spoken conversation into text. Learn more →
- On-premises deployment
- The system runs on the company's own servers, not in the vendor's cloud. Learn more →
- Customer record
- The screen of customer data an agent sees during a conversation. Learn more →
Updated: