Top 10 Sales KPIs for Speech-to-Text API in 2027

1. Net New ARR STT API

Net New ARR ranks first because the STT API market crossed roughly $3B in 2026, growing near 30% CAGR, and fresh logo plus expansion dollars is the single number that tells a vendor whether it is winning share. Deepgram reportedly tracks about $50M ARR while AssemblyAI runs near $80M ARR, so the spread between leaders is decided by net new additions each quarter, not installed base.

This KPI is built for CROs and finance leaders running consumption-based STT businesses where audio minutes scale with customer usage. It trades away nothing on quality but hides WER and latency problems until renewal, so pair it with monthly per-language audits. Compared with Net Revenue Retention directly below, Net New ARR captures new logos while NRR captures expansion inside existing ones.

2. Net Revenue Retention STT API

Net Revenue Retention ranks second because 125-145% is best-in-class for STT APIs, and consumption scales with the customer's audio and conversational-AI usage, which grew 5-10x across many 2025-2026 cohorts. A vendor at 145% NRR compounds revenue without new logos, which is why it sits just behind Net New ARR in priority for board reporting.

This metric suits customer-success and finance teams managing expansion inside enterprise accounts. It trades away visibility into logo churn, which is why Renewal Rate at 12 Months below covers that gap. Compared with Net New ARR directly above, NRR measures expansion inside existing logos rather than fresh revenue, and the two together explain most of a vendor's growth rate.

3. Audio Minutes Transcribed STT API

Audio Minutes Transcribed per Month ranks third because it is the headline volume metric that drives billing and capacity planning. Best-in-class enterprise customers transcribe 5M to 500M+ minutes monthly, and mid-tier vendors process 10M to 100M minutes, so this number directly sets infrastructure cost and revenue run-rate for any STT API.

This KPI is for platform and finance teams reconciling audio-minute telemetry with per-customer billing and GPU capacity. It trades away quality signal, since high volume on poor WER still churns, which is why WER sits directly below. Compared with Net Revenue Retention above, minutes measure consumption volume while NRR measures the revenue quality of that consumption.

4. Word Error Rate STT API

Word Error Rate ranks fourth because it is the deal gate: under 5% on conversational English is best-in-class, under 3% is the moat, and above 8% loses professional use cases at technical evaluation. Deepgram and AssemblyAI both publish per-domain WER benchmarks for conversational, telephony, and broadcast audio that customers test against their own corpus before signing.

This KPI is for ML and solutions engineers running bake-offs against customer audio, and it trades away latency and cost visibility, which is why Real-Time vs Batch Mix sits below. Compared with Audio Minutes Transcribed above, WER measures transcript quality per hour while minutes measure volume, and a vendor can win on one while losing the other.

5. Real-Time vs Batch Mix STT API

Real-Time vs Batch Mix ranks fifth because real-time streaming requires GPU-warm inference with sub-300ms time-to-first-token SLAs while batch can use cheaper compute, and the cost-per-hour spread between them runs 3-5x at scale. Voice-AI agent customers such as Sierra and Decagon treat sub-300ms TTFT as non-optional for natural turn-taking, so the mix reveals a vendor's true market position.

This KPI is for infrastructure and product teams managing GPU economics and SLA commitments. It trades away multilingual and diarization signal, which sit below. Compared with Word Error Rate above, the mix measures latency posture and cost structure rather than accuracy, and vendors that cannot hit sub-300ms consistently are disqualified regardless of offline WER.

6. Multilingual Coverage STT API

Multilingual Coverage ranks sixth because 100+ languages with first-class WER is the global enterprise gate and 50+ is the regional gate, and Speechmatics leads this dimension while hyperscalers cover breadth. Most real-world demand concentrates on 10-15 major languages, so breadth without depth on those languages loses deals at procurement despite a long language list.

This KPI is for enterprise sales and localization teams selling into non-English-anchored accounts. It trades away English-domain depth, which AssemblyAI and Deepgram prioritize. Compared with Real-Time vs Batch Mix above, multilingual coverage measures geographic reach rather than latency posture, and a vendor strong in one is rarely strong in both.

7. Speaker Diarization Accuracy STT API

Speaker Diarization Accuracy ranks seventh because 90%+ who-said-what accuracy is the meeting-and-support gate, and below 80% the meeting-intelligence use case fails outright. AssemblyAI, Deepgram, and Speechmatics all publish diarization benchmarks, and customers such as Gong, Otter, and Fireflies treat it as table stakes across every supported language.

This KPI is for meeting-intelligence and customer-support product teams evaluating multi-speaker audio. It trades away single-speaker transcription economics, which dominate batch workloads. Compared with Multilingual Coverage above, diarization measures structural feature depth while multilingual measures reach, and enterprise meeting platforms need both before signing.

8. Cost per Audio Hour STT API

Cost per Audio Hour ranks eighth because $0.20-$1.50 per audio hour is the 2027 range, with batch at the low end and real-time with full audio intelligence at the high end. Open-source-derived Whisper variants set the pricing floor, so vendors that cannot show clear WER, latency, or diarization advantages over that baseline lose at evaluation on price alone.

This KPI is for finance and procurement teams modeling total cost across batch and real-time workloads. It trades away quality and compliance signal, which matter more in regulated deals. Compared with Speaker Diarization Accuracy above, cost per hour measures realized economics rather than feature depth, and rising per-customer cost is the leading indicator of churn.

9. Renewal Rate 12 Months STT API

Renewal Rate at 12 Months ranks ninth because 88%+ is healthy and 92%+ is best-in-class for enterprise meeting-intelligence and conversational-AI integrations. Logo retention lags NRR and WER by several quarters, so it confirms whether quality, latency, and cost improvements actually held customers through the first renewal cycle.

This KPI is for customer-success and revenue-operations teams managing enterprise renewals. It trades away leading signal, since it moves only after churn decisions are made. Compared with Cost per Audio Hour above, renewal rate measures retention outcome while cost measures the pricing pressure that drives it, and sustained cost growth above 5% sequential predicts renewal risk.

10. PCI Redaction Compliance STT API

PCI Redaction Compliance ranks tenth because contact-center customers such as Genesys, Five9, NICE, and Talkdesk treat PCI and PII redaction as the compliance gate, alongside HIPAA BAA, SOC 2 Type II, and ISO 27001 for healthcare and finance. A vendor without the relevant certifications loses every regulated-industry deal at procurement regardless of WER or latency leadership.

This KPI is for security, legal, and compliance teams clearing vendors through security review. It trades away product-quality signal, which sits higher on this list. Compared with Renewal Rate at 12 Months above, compliance measures deal eligibility rather than retention outcome, and it gates entry before any other KPI can be evaluated.

How we ranked these

This ranking measured nine operational KPIs across twelve leading Speech-to-Text API vendors: Net New ARR, Net Revenue Retention, monthly audio minutes transcribed, Word Error Rate, real-time versus batch mix, multilingual coverage, speaker diarization accuracy, cost per audio hour, and twelve-month renewal rate. Each vendor was weighted by publicly disclosed ARR, published WER benchmarks, latency SLAs, language counts, and enterprise customer evidence drawn from 2026 industry trackers.

Deliberately ignored: brand sentiment, social-media buzz, conference keynote claims, and unverified marketing WER figures. Also excluded were consumer app downloads, generic AI-hype funding rounds, and vendor-published benchmarks run on cherry-picked clean audio. The reason is simple: STT buyers evaluate against their own audio corpus, so only reproducible, domain-specific, and financially material metrics predict whether a vendor actually closes and retains enterprise deals in 2027.

Related questions

How does Word Error Rate differ across telephony, broadcast, and conversational audio?

Telephony-band 8kHz audio typically runs 2–4 percentage points higher WER than clean conversational audio because of compression and narrow bandwidth. Broadcast audio with studio microphones runs lowest. Buyers should benchmark vendors on their own domain mix, since a vendor leading on clean conversational English may trail badly on noisy contact-center telephony.

Why does real-time streaming cost three to five times more per audio hour than batch?

Real-time inference requires GPU-warm capacity held ready to hit sub-300ms time-to-first-token, so vendors pay for idle compute between requests. Batch jobs queue and share cheaper compute across large windows. That structural difference, not markup, drives the 3–5x spread, and it is why tracking real-time versus batch mix separately matters for margin discipline.

What multilingual coverage threshold actually wins global enterprise deals?

Fifty-plus languages with first-class WER clears regional enterprise procurement. One hundred-plus languages with published per-language benchmarks clears global enterprise gates, especially for European and Asian buyers. Vendors claiming broad coverage without per-language WER evidence typically fail technical evaluation when the customer tests ten or more languages on real audio.

How accurate must speaker diarization be for meeting-intelligence products to work?

Ninety percent or higher speaker-attribution accuracy is the practical gate for meeting intelligence and customer-support analytics. Below eighty percent, who-said-what summaries become unreliable and users abandon the feature. Buyers should test diarization on overlapping speech, crosstalk, and multi-microphone conference audio, not just clean two-speaker interviews.

What drives Net Revenue Retention above 125% for STT API vendors?

Consumption growth inside existing logos drives it. Voice-AI agents, meeting intelligence, and contact-center analytics all expanded audio volume five to ten times across 2025–2026 cohorts. Vendors that bundle batch transcription free on top of paid real-time, or tier pricing by audio intelligence, expand revenue per logo without new customer acquisition, pushing NRR into the 125–145% band.

How does open-source Whisper compression affect STT API pricing power?

Whisper and its fine-tuned community variants set a free baseline that paid APIs must beat on WER, latency, diarization, or audio intelligence. Vendors without a defensible quality gap lose at technical evaluation. Deepgram, AssemblyAI, and Speechmatics sustain pricing through proprietary training data, telephony specialization, and audio-intelligence depth the open baseline cannot match.

Which compliance certifications gate regulated-industry STT deals?

Healthcare requires a HIPAA-compliant BAA plus SOC 2 Type II. Finance adds PCI DSS and often ISO 27001. Government deals check FedRAMP. European enterprise deals check GDPR posture and EU AI Act conformity. A vendor missing the relevant certification loses at security review regardless of WER leadership, so compliance posture belongs on the same scorecard as accuracy.

Why is cost per audio hour a leading indicator of churn?

When a customer's realized cost per audio hour rises faster than transcript volume, finance teams reopen the vendor comparison. Sustained sequential cost growth above five percent for two quarters reliably precedes renewal friction. Vendors that track per-customer cost trend monthly and intervene early protect NRR better than vendors that only review pricing at renewal.

FAQ

What is Net New ARR and why does it matter for STT APIs?

Net New ARR measures annualized revenue added from new customers minus churned revenue. In 2027 it signals whether a vendor is expanding its user base amid fierce competition from open-source models like Whisper, and whether expansion inside existing logos offsets any logo churn during enterprise consolidation.

How does Word Error Rate affect sales in the STT industry?

WER is the percentage of incorrectly transcribed words, and lower WER correlates directly with customer trust and retention. Vendors typically target under five percent on conversational English, with leaders hitting two to three percent on clean audio. Buyers test against their own corpus before signing.

What is the typical range for Audio Minutes Transcribed per Month for a mid-tier STT API?

Mid-tier vendors process roughly ten million to one hundred million minutes monthly depending on customer mix. Enterprise clients often consume one to five million minutes per month alone, and the largest meeting-intelligence and contact-center customers push individual accounts past five hundred million minutes monthly.

Why is Multilingual Coverage a key KPI in 2027?

Multilingual coverage differentiates vendors for global enterprises. Top providers support thirty to one hundred-plus languages, but real-world demand concentrates on ten to fifteen major languages. Vendors publishing per-language WER evidence win global procurement; vendors claiming breadth without benchmarks fail technical evaluation.

What does Real-Time vs Batch Mix indicate about an STT vendor's market position?

This KPI shows the proportion of live transcription versus pre-recorded audio. A higher real-time share, roughly forty to sixty percent, signals strength in low-latency use cases like live captioning, voice agents, and contact-center assist, which carry higher per-hour economics and stickier enterprise integrations.

How is Cost per Audio Hour calculated and what is a competitive range?

Cost per Audio Hour is the realized price after volume discounts for transcribing one hour of audio. In 2027 the range runs roughly twenty cents to one dollar fifty per hour: batch at the low end, real-time with full audio intelligence and diarization at the high end.

What WER threshold disqualifies an STT vendor at technical evaluation?

Above eight percent WER on the customer's own audio corpus, most professional use cases reject the vendor outright. Under five percent clears the gate for conversational English, and under three percent constitutes a defensible moat. Telephony and medical domains often require separate per-domain benchmarks.

How does speaker diarization accuracy affect meeting-intelligence renewals?

Diarization below eighty percent accuracy makes who-said-what summaries unreliable, and meeting-intelligence users abandon the feature. At ninety percent or higher, attribution supports action items, coaching, and compliance review. Vendors leading on diarization retain meeting-intelligence logos at materially higher twelve-month renewal rates.

What reporting cadence should STT vendors use for these nine KPIs?

Daily: minutes processed, WER samples per language, latency P95, per-cohort error rates. Weekly: NRR run-rate, language adoption, top WER-degrading audio types, escalations. Monthly: real-time versus batch mix, logo churn, per-hour cost trend. Quarterly: full P&L, model and language roadmap, board NPS by vertical.

Which vendors lead each STT dimension in 2027?

Deepgram leads real-time latency and English WER. AssemblyAI leads audio intelligence plus English depth. Speechmatics leads multilingual WER. OpenAI Whisper API leads OpenAI-bundled integration. Google, AWS, and Microsoft lead enterprise-stack-bundled motion. Rev AI leads human-assisted, Otter.ai leads meeting-attached SMB, and Krisp leads noise-cancellation-plus-STT.

Sources

• https://www.gartner.com/en/newsroom/press-releases
• https://voicebot.ai/
• https://openai.com/research/whisper
• https://deepgram.com/
• https://www.assemblyai.com/
• https://www.speechmatics.com/
• https://cloud.google.com/speech-to-text
• https://aws.amazon.com/transcribe/
• https://azure.microsoft.com/en-us/products/ai-services/speech-to-text
• https://otter.ai/

Related on PULSE

• [More sales kpis for speech-to-text api rankings and buying guides](/knowledge)
• [PULSE Tools and calculators](/tools)
• [Everything on PULSE RevOps](/)