Discovering Best Practices from Call Center Dialogues Using AI

Discovering Best Practices from Call Center Dialogues Using AI

You’ve likely noticed that some operators close tickets with CSAT 5/5 in 5 minutes, while others struggle with CSAT 3/5 for 20 minutes. Over our practice, we’ve analyzed thousands of dialogues. It turns out that top performers intuitively use techniques not documented in standard procedures. Our AI system identifies these patterns and turns them into replicable best practices. This allows you to quickly elevate the performance of all agents to the top 10% without lengthy training. We use a combination of NLP and RAG to analyze dialogues. On a typical project, we process at least 1000 dialogues and identify up to 30 unique patterns. The result: a 20% reduction in handling time and a 0.4-point CSAT increase within 3 months. Based on the discovered practices, we create training modules for your coaches. Ready to assess your project? Contact us—we’ll tailor a solution for you. Clients save up to $50,000 per year after implementation.

How to Determine the Best Dialogues with AI?

A dialogue is considered exemplary based on a set of signals. We use a composite score based on:

• CSAT ≥ 4/5 (Wikipedia: Customer satisfaction) — post-interaction survey score
• First Call Resolution = true
• Handling time ≤ 7 minutes (if the group average is no more than 10)
• No repeat contact within 48 hours

Sample: top 10% of dialogues by composite score. For comparison, we take the bottom 10% to maximize contrast.

Pattern Extraction: Linguistics, Structure, Context

The AI analyzes the corpus of top dialogues and identifies three groups of patterns.

Linguistic patterns — what phrases top operators use at critical moments. For instance, when handling a complaint: "I understand how unpleasant this is" coupled with an immediate solution proposal. Poor operators spend 30+ seconds apologizing without providing specifics.

Structural patterns — how a successful dialogue unfolds step by step. Frequency comparison between top 10% and bottom 10% shows that top operators get straight to the point after the greeting, while poor operators make 2–3 unnecessary confirmations.

Problem-specific patterns — how to handle typical objections ("too expensive", "I’ll think about it", "not suitable"). Our system extracts successful scripts for each type of inquiry.

def extract_best_practices(
    top_dialogs: list[Dialog],
    bottom_dialogs: list[Dialog],
    topic: str
) -> list[BestPractice]:
    prompt = f"""Compare successful and unsuccessful dialogues on topic '{topic}'.
Identify 5 specific practices that distinguish successful dialogues.
For each practice: description + a quote from a dialogue as an example."""
    return llm.extract_structured(prompt, top_dialogs, bottom_dialogs)

Algorithm details: For each group of dialogues we compute TF-IDF vectors and cosine similarity. Statistically significant differences (p < 0.01) go into the final report. Verification is carried out on a hold-out sample—at least 500 dialogues.

Why Are Monthly Practice Updates Important?

The service landscape changes quickly: new products, seasonal fluctuations, policy updates. What worked a few months ago may be obsolete today. For example, when we started with one retailer, we identified 15 strong patterns, but after a few months 3 of them lost effectiveness due to script changes. The system automatically recalculates practices every month—you always use current approaches.

Deliverables Included in the Work

We provide:

• Report with identified practices — at least 20 specific patterns with examples from your dialogues.
• Documentation — description of each pattern, how to implement it, and how to measure impact.
• API access — you can use the model for real-time analysis of new dialogues.
• Team training — a webinar or workshop for your trainers and QC specialists.
• Support — 3 months of accompaniment, with practice updates every 4 weeks.

Efficiency Comparison: AI vs. Manual Analysis

AI analysis is 40 times faster than manual analysis.

Our Track Record

We have implemented similar systems in 12 contact centers. Our team has over 5 years of experience in dialogue analytics. We guarantee data confidentiality. On average, after implementation our clients see:

• CSAT +0.4 points,
• FCR +15%,
• handling time reduced by 20%. Clients typically save $50,000 annually after implementation. The probability of discovering non-obvious practices is 4 times higher with AI compared to manual methods.

Implementing the AI System in Your Contact Center

The process consists of four stages:

1. Analytics — collect and label 1000+ dialogues (2–3 weeks).
2. Design — configure the model to your specifics (1–2 weeks).
3. Pilot — run on 10% of traffic, compare with control group (2 weeks).
4. Full deployment — integrate with your CRM, train users (1–2 weeks).

Total timeline: 6 to 9 weeks end-to-end. Pricing is calculated individually—we’ll assess your project for free. Just reach out to us.

Order a free analysis of 100 of your dialogues—discover what practices are hidden in your data. Get in touch for a custom quote.

NLP Development: Text Classification, NER, Embeddings, and Information Extraction

We often receive a task: process 50,000 support tickets — currently all manual. Dataset — 3,000 labeled examples, 12 categories, imbalance: one category occupies 40% of the sample, three at 1-2% each. Baseline accuracy — 78%. Sounds decent until you look at recall for rare classes: 0.31, 0.44, 0.28. These classes — complaints and churn threats — are most important to the business.

This is a typical NLP development project. The problem is not the algorithm but that accuracy is the wrong metric. Our experience across 30+ projects shows: we start by analyzing business metrics and only then choose the model.

Why accuracy is not the right metric for rare classes?

Accuracy ignores imbalance. If the "churn" class appears in 2% of cases, the model can predict "all good" and get 98% accuracy — but the business loses clients. Solution: F1 macro (averaged over all classes) or weighted F1. For NER — strict entity F1 (exact matches only). We guarantee: after choosing the correct metric, model quality becomes measurable and predictable.

Text Classification: From BERT to Distillation

BERT-like models are the standard for classification. ruBERT-base or ruBERT-large from DeepPavlov for Russian. multilingual-e5-large — for multiple languages in one pipeline. XLM-RoBERTa-large — a strong multilingual backbone.

Fine-tuning for classification: add a classification head on top of the [CLS] token, train for 3-5 epochs with lr=2e-5, weight decay=0.01. For imbalance — weighted CrossEntropyLoss or focal loss with gamma=2.0. Contact us — we will show a code snippet.

Imbalance case study. Dataset — 3,000 examples, imbalance 1:20. Solution: class_weight via sklearn + CrossEntropyLoss. Additionally — augmentation of rare classes via backtranslation (ru→en→ru through MarianMT). Recall for rare classes rose from 0.31 to 0.67 with a slight drop in accuracy (76%→74%). Full NLP development end-to-end took 3 weeks.

Distillation for production. BERT-large gives F1 0.89, but inference on CPU — 180ms. Distillation into DistilBERT or ruBERT-tiny2 reduces latency to 25ms with F1 0.84. Export to ONNX Runtime provides an additional 1.5-2x speedup. DistilBERT achieves 7x lower latency than BERT-large with only a 5% drop in macro F1 – a typical production trade-off.

How to choose between BERT and LLM for your task?

For most classification and extraction tasks, BERT-sized models offer the best trade-off between cost and performance. Shift to LLMs only when the task demands generation, complex reasoning, or zero-shot generalization.

NER: Named Entity Recognition

NER — extracting persons, organizations, locations, dates, amounts, document numbers. For general categories (PER, ORG, LOC), pre-trained models work well. For specialized ones (medical terms, legal concepts) — fine-tuning is needed.

Data annotation. The main cost of an NER project. For a quality model — 500-2,000 labeled sentences per entity type. Tools: Label Studio (open source) or Prodigy (by spaCy creators). IOB2 format — standard.

Architecture. Token classification on top of BERT: each token gets a label (B-PER, I-PER, O). spaCy 3.x with transformer pipeline — a convenient production choice.

Nested entities. Standard IOB models cannot handle nested entities (organization inside an address). For such tasks — span-based NER: SpanBERT or SpERT. More complex but correct.

Post-processing is mandatory. The model predicts tokens — normalized entities are needed. Date — dateparser. Amounts — regex + validation. Names — deduplication via rapidfuzz. Included in our standard delivery.

Sentiment Analysis and Opinion Mining

Binary classification positive/negative works out of the box with BERT. Complexity — aspect-based sentiment analysis (ABSA): "the restaurant has good food but terrible service." For ABSA: aspect extraction (NER) + sentiment per aspect. Joint models BERT-for-ABSA — quality on Russian data is lower due to dataset scarcity. RuSentiment, SentiRuEval — main resources.

For production with simple positive/negative/neutral: distil models are enough. Three classes, balanced dataset, 2,000+ examples — F1 macro 0.82-0.87 in 1-2 days.

Text Summarization

Extractive summarization (select sentences) — TextRank or BM25 without training. Fast, no hallucinations. Good for long documents.

Abstractive (generates new text) — seq2seq: mT5, mBART, FRED-T5, ruT5-large. For production via LLM API (GPT-4, Claude) — often the best cost/quality/speed trade-off.

Embeddings: Vector Representations of Text

Embeddings are the foundation of semantic search, deduplication, clustering, RAG. Quality critically affects downstream tasks.

Models. E5-large-v2, BGE-M3, multilingual-e5-large — strong multilingual embedders. sentence-transformers/paraphrase-multilingual-mpnet-base-v2 — fast option. For Russian: ru-en-RoSBERTa (Skoltech) performs well on semantic textual similarity.

Embedding quality evaluation uses the MTEB benchmark as standard. But top results on MTEB don't guarantee success on a domain dataset — we build domain-specific eval.

Fine-tuning embeddings. If standard models don't give the required Recall@k — contrastive learning on domain pairs with MultipleNegativesRankingLoss. How to perform this for domain data:

1. Collect 500–2,000 semantically similar pairs from your domain.
2. Apply MultipleNegativesRankingLoss with a batch size of 32–64.
3. Train for 1–3 epochs using AdamW (lr=2e-5).
4. Evaluate Recall@k on a held-out domain test set.

This approach yields a 5–15% improvement in Recall@k in practice.

Dimensionality and storage. E5-large: 1024 dim, float32 — 4KB per vector. For 10M documents — 40GB. Quantization int8 reduces to 10GB. FAISS IVF_PQ — more compact but with losses. Included in our deployment recommendations.

Information Extraction

Structured extraction is a frequent task. Examples: key contract terms, technical characteristics, dates and amounts from invoices.

1. Regex + rule-based. For INN, OGRN, amounts, dates — more reliable than neural networks. No data required.
2. NER + post-processing. For variable formats.
3. LLM with structured output. GPT‑4 / Claude with JSON schema — for complex documents. Cost: minimal per document. For 10k+ documents/day — we calculate the economics.

We guarantee a hybrid: regex/NER for typical fields + LLM for edge cases. Our guarantee is backed by years of production experience and more than 30 projects.

Work Stages

Prototype on existing data — 1-3 weeks. Production system with CI/CD — 1.5–2.5 months. Cost is calculated individually — get a consultation for a project estimate.

What's Included

• Model and pipeline architecture documentation
• Access to the model via REST API (FastAPI + ONNX)
• Client team training (2-hour webinar + Q&A)
• Accuracy guarantee on the agreed test set
• Months of post-delivery support (bug fixes, adaptation to new data)

Our Experience

Years of NLP projects from classification to RAG systems. The team includes ML engineers experienced with Hugging Face, spaCy, LangChain, MLOps. We use vLLM, Kubeflow, Weights & Biases — a production stack, not toys. Contact us to evaluate your NLP project within two days — request a free consultation on your text processing pipeline.