CallScribe for BPO operations

Why GCC BPO is structurally different from offshore BPO

A Manila or Bangalore BPO serving US clients optimises for English-language QA, US-time-zone shift coverage, and cost per agent-hour. A Riyadh or Manama BPO serving GCC clients optimises for Arabic-language QA across multiple dialects ([/dialects/khaleeji](/dialects/khaleeji), Levantine, Egyptian), Saudization or Emiratisation hiring quotas, and regulatory compliance in the buyer's sector — banking, telecom, government. The QA problem space is genuinely different. English-first speech-analytics tools (NICE Nexidia, Verint, CallMiner) handle Arabic as a secondary language with degraded accuracy on Khaleeji content, which is the dominant dialect. The result is BPO QA teams falling back to manual sampling because their automation does not work on the audio they actually have.

CallScribe is built Arabic-first and prices flat per project rather than per minute, which matters for BPO economics. A 1,000-agent BPO operating across 4-6 client programs may transcribe 5-8 million call minutes per month — at typical per-minute pricing from English-first vendors, the bill is prohibitive. A flat-rate Scale plan plus enterprise tier brings the cost per call into the cents range, which is what BPO P&L can absorb.

Vision 2030, Saudization, and the BPO scale-up

Vision 2030's contact-centre localisation push has reshaped the GCC BPO landscape. The Public Investment Fund (PIF), through subsidiaries like ccc by stc and partnerships with global operators, has built capacity for hundreds of thousands of agent seats in Saudi Arabia. Saudization quotas — initially 30% Saudi nationals in BPO operations, rising toward 50%+ — change the workforce composition: a higher proportion of Najdi and Hijazi Khaleeji speakers, with English fluency variable across the Saudi national cohort. QA teams need to assess Arabic-language calls handled by Saudi nationals and English-language calls handled by expatriate agents on the same scorecard, with dialect-appropriate sentiment models for each.

The Saudi Communications, Space and Technology Commission (CST, formerly CITC) regulates licensed contact-centre operations and increasingly expects retention and retrievability of customer interactions for dispute and compliance review. CallScribe's retention configuration supports per-program rules so a BPO running banking, telecom, and government client programs can apply appropriate retention windows to each without manual reclassification.

100% transcription replaces 5% sampling

Industry standard QA practice is to sample 2-5% of calls per agent per month — a figure constrained by what human QA reviewers can audit, not by what produces statistically valid agent performance scores. At 5% sampling, a 200-agent program with each agent handling 1,500 calls/month produces 15,000 audited interactions out of 300,000 — but the sample is heavily biased toward whatever criteria the QA lead used to pull calls (often "calls that look problematic from disposition codes", which themselves are unreliable).

CallScribe transcribes 100% of calls and runs an automated QA scorecard on every transcript: greeting verification, customer authentication, scripted disclosure detection, hold-time tracking, call-closure phrasing, profanity and policy-violation detection. Human QA shifts to auditing model-flagged calls (low scores, high-disagreement zones, regulator-relevant interactions) rather than pulling random samples. A 100-strong human QA team that previously audited 15,000 calls/month now reviews 50,000-80,000 model-flagged calls with deeper attention because the routine scoring is automated. See [/use-cases/support-qa](/use-cases/support-qa) for the workflow detail and [/use-cases/coaching](/use-cases/coaching) for how QA outputs become coaching evidence.

Multi-tenant BPO operations and per-client isolation

A BPO running 6 client programs cannot mix data across clients — contractual isolation is a hard requirement, often backed by client-side audit rights. CallScribe's tenancy model uses PostgreSQL Row-Level Security to enforce per-project isolation; a BPO admin can create separate projects per client program with independent retention rules, scorecards, vocabulary, and access controls. Client-side auditors can be granted scoped access to their program's transcripts only, with audit logs of access exposed for compliance review.

Per-client custom vocabulary matters operationally. A telecom-client program needs telco-specific brand names, plan codes, and technical jargon; a banking-client program needs financial-product names and regulatory phrasing. Vocabulary uploads are per-project and apply to both the acoustic-language model bias and downstream search-and-analytics indexing.

Agent coaching at BPO scale

A 200-agent program produces a coaching backlog that no team-leader population can manage manually. CallScribe assembles per-agent monthly evidence packs: 5 strongest calls (high CSAT, fast resolution, full QA score), 5 weakest calls with specific scorecard misses highlighted, aggregate metrics on talk-listen ratio, filler frequency, and dead-air. Team leaders spend coaching time reviewing evidence rather than collecting it. For BPO operations directors, agent-attrition reduction tied to better-quality coaching is one of the harder ROI numbers to attribute but consistently shows up as a 3-5 percentage point improvement in 90-day retention when coaching becomes evidence-driven.

KPIs BPO operations directors track

AHT decomposed into talk/hold/wrap (not raw PBX duration); FCR from transcript content rather than dispatch metadata; sentiment-derived CSAT covering 100% of calls rather than the 5-10% who respond to IVR surveys; per-agent QA scorecard compliance; per-program scorecard compliance trends; regulator-relevant flag rates (mis-selling, missed disclosures, PII leakage); shrinkage and adherence inferred from talk-time analysis. CallScribe outputs all of these via API or CSV export for ingestion into the BPO's WFM and BI stack (NICE WFM, Verint, Genesys Workforce Engagement, or in-house tools).