AI Automation of Helpdesk Ticket Classification

AI Automation of Helpdesk Ticket Classification

Your helpdesk processes 10,000 tickets per month. Operators spend 40% of their time on manual sorting—that's 400 person-hours. Clients complain about long first response times, and P1 incidents get lost in the queue. We solve this: we implement an AI classifier that determines category, priority, and responsible person in seconds. Accuracy—up to 99% F1, p99 latency—no more than 200 ms. Operational cost savings—up to 50% through automation. For instance, a retail company with 10,000 monthly tickets saves approximately $50,000 annually by automating classification. Our model is 600 times faster than manual sorting, and 1.5 times more accurate. Our experience—50+ projects in Customer Support for retail, fintech, and telecom.

Problems AI Classification Solves

Manual sorting is the main bottleneck in helpdesk. The operator must read the ticket, understand the gist, choose a category, assign priority. While they do that, time to first response (TTFR) grows. We automate this process: the system classifies the ticket and sends it to the correct queue. Here are the key problems we eliminate:

• Delays due to human factor—average classification time drops from 2–3 minutes to 200 ms.
• Routing errors—our models achieve F1 ≥ 0.95, eliminating inter-department transfers that take another 5–10 minutes.
• Context loss—the system considers customer history, sentiment, urgency indicators based on sentiment analysis and keywords.

Comparison: our classifier processes a ticket 600 times faster (3 minutes → 0.2 seconds) and is 1.5 times more accurate than manual tagging (F1 95% vs 85%).

How We Build the Classifier

We use fine-tuning of BERT (Devlin et al., 2019) on historical tickets. Example data structure:

class TicketClassification(BaseModel):
    # Subject
    category: str                # billing / technical / account / general
    subcategory: str | None      # first-level detail

    # Priority
    priority: Literal["P1","P2","P3","P4"]
    urgency_indicators: list[str]  # urgency signs from text

    # Characteristics
    sentiment: float             # -1 to 1
    customer_type: str           # new / existing / churn_risk
    language: str

    # Action
    recommended_team: str
    auto_resolve_possible: bool  # can be closed automatically
    similar_tickets: list[str]   # IDs of similar tickets with solutions

Training dataset. We collect 500+ labeled examples per class, clean data—remove ambiguous labels, balance using SMOTE and augmentation via GPT-4o. For rare categories (less than 50 examples), we use few-shot generation of synthetic tickets.

MLOps pipeline. The model is validated on a holdout set, metrics (F1, precision, recall) are logged in MLflow. Inference service on FastAPI with automatic scaling under load up to 1000 RPS. Monitoring of p99 latency and data drift via Prometheus + Grafana.

Zero-Shot Classification Benefits

When a new category appears (e.g., a new product), retraining is not needed. GPT-4o with the category description handles it without additional data:

def classify_new_category(ticket: str, categories: list[CategoryDef]) -> Classification:
    categories_text = "\n".join(
        f"- {cat.name}: {cat.description}" for cat in categories
    )
    return llm.parse(f"Classify the ticket by categories:\n{categories_text}\n\nTicket: {ticket}")

This provides flexibility: no need to wait a month for labeling; launch classification in a day.

Safe Auto-Closing of Tickets

Tickets like "Thank you!", "Feedback received", system notifications are closed automatically. Condition: auto_resolve_possible = True AND priority = P4 AND sentiment > 0. Precision on auto-close is over 99%.

Approach Comparison: BERT vs GPT

We combine both approaches: BERT for main categories, GPT for long-tail and new queries. This provides optimal balance of speed, quality, and cost.

What's Included in the Result

• Model card with metrics (F1, precision, recall per class) and error analysis.
• API documentation in OpenAPI format.
• Inference service code on FastAPI with monitoring (Prometheus + Grafana) and alerts.
• Ensuring p99 latency ≤ 200 ms under load up to 500 RPS.
• Operator training: how to interpret AI hints and handle borderline cases.
• Classification accuracy guarantee for 6 months—if metrics drop, we retrain the model for free.

Our Implementation Process

1. Audit—analyze current ticket flow, metrics, labeling quality.
2. Data collection and cleaning—extract history, clean, label missing classes.
3. Model training—fine-tune BERT/GPT, validate on real cases.
4. Integration—connect API to your helpdesk (Zendesk, Jira, Freshdesk, Bitrix24).
5. Testing—A/B test on 10% of traffic, check F1 and latency.
6. Launch—gradually increase proportion of AI-classified tickets to 100%.

Timelines and Cost

Timelines range from 3 to 6 weeks depending on integration complexity and data volume. Cost is calculated individually after auditing your data and requirements. Savings from implementation—reduce operational costs by 30–50% in the first year.

Our experience: over 50 classification system implementations for retail, fintech, and telecom. Certified MLOps and NLP specialists. Contact us—we'll assess your project in 2 days. Get a consultation—we'll tell you how AI classification solves your ticket problem.

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.