AI Chat Scoring System for Operator Quality Control

Development of an AI System for Automatic Operator Chat Scoring

Manual evaluation of chat quality is the weak link in most contact centers: analysts check only 2–5% of dialogues. AI scoring analyzes 100% of chats and provides objective evaluation against standardized criteria. We have implemented such systems in 50+ projects and guarantee: scoring accuracy is at least on par with humans at 100x speed. AI model accuracy reaches 95% with correlation to experts >0.9, outperforming the average analyst (85–90%). Implementing automatic scoring can save up to 60% of the QA department budget. For a contact center with 50 operators, annual savings can reach 2 million rubles.

What Problems Does AI Scoring Solve?

The main pain point is subjectivity and low sample size. A person gets tired by afternoon, attention wanes, criteria fluctuate. AI is stable: equally strict in the morning and evening. The second problem is missing systemic errors. If an operator is rude to every tenth customer, it goes unnoticed with a 5% sample. AI will detect the pattern within a day.

The third issue is scaling. Hiring 10 analysts for 500 operators is expensive and time-consuming. An LLM model can be fine-tuned in a week and processes chats in real time.

How We Configure Evaluation Criteria

The typical checklist is just a foundation. We add specifics of your business: for support—correctness of diagnostics, for sales—success of cross-sell, for logistics—accuracy of deadlines. Criteria are weighted: a mistake in the solution may cost 0.7 points, while a missing greeting costs 0.1.

Why Model Calibration Matters

Without calibration, the model is a black box. It may evaluate politeness rather than usefulness. Our process: experts label 300–500 chats, the model learns from these ratings, then we compare on a held-out set. Target: correlation >0.85. If the metric drops, we update the benchmark or retrain. Regular quarterly calibration maintains accuracy.

Implementation of AI Scoring

Base class in Python with Pydantic:

class ChatQualityScore(BaseModel):
    greeting_score: float        # 0-1
    problem_understanding: float
    solution_accuracy: float
    communication_quality: float
    procedure_compliance: float
    empathy_score: float
    overall_score: float         # weighted sum
    highlights: list[str]        # specific examples from chat
    improvement_areas: list[str] # what to improve

def score_chat(dialog: list[dict]) -> ChatQualityScore:
    # Pass the entire dialogue + evaluation criteria
    return llm.parse(
        build_scoring_prompt(dialog),
        response_format=ChatQualityScore
    )

The model returns a structured result automatically loaded into CRM or BI systems.

Comparison of Manual Evaluation and AI Scoring

What the Work Includes

1. Audit of current standards — analyze 100 chats, identify criteria and typical mistakes.
2. Model tuning — fine-tune LLM (GPT-4 or Llama 3) on your data using LoRA.
3. Calibration and testing — check correlation with experts, adjust weights.
4. Integration — connect to your chat solution (Zendesk, LiveChat, Bitrix24) via API.
5. Reports and dashboards — set up Google Data Studio or Grafana with auto-generation once a week.
6. Operator training — conduct a workshop on working with the system’s recommendations.

Timeline: 2 to 8 weeks depending on integration complexity. Cost is calculated individually—contact us for a project estimate.

Feedback for Operators

Automatic weekly reports for each operator: strengths, growth areas, trends. Gamification: team rankings, badges for improvement. Management dashboard: heat map of problematic criteria across the team—where training is needed.

Our Experience and Guarantees

We have implemented 50+ projects for retail, telecom, and fintech. We provide a model card with accuracy metrics, guarantee SLA response time—p99 under 2 seconds per chat. After deployment, we maintain the system: quarterly calibration, retraining when scripts change.

Example operator report: 120 chats for the week, overall score 0.87. Strengths: solution accuracy (0.92), empathy (0.90). Growth areas: greeting completeness (0.65), CRM entry (0.70). Monthly trend: +0.05.

Get a consultation: we'll explain how to adapt AI scoring to your processes. Order a pilot project on 100 chats—evaluate accuracy before purchase.

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.