Handoff AI Bot to Agent: Seamless Context Transfer

We encountered a situation: a chatbot customer repeats the same question three times, the bot gives template replies, and the customer gets annoyed. Instead of transferring the conversation to an agent with full context, the system closes the chat or suggests calling. Result: customer loss. Our handoff module solves this: the agent sees the history, emotional tone, and escalation reason in seconds. Handoff AI bot to agent ensures dialog context preservation and creates a seamless transfer. A chatbot with agent support becomes an efficient service tool.

Our company has 5+ years of experience in AI solutions, with over 20 successful handoff integrations, helping clients achieve measurable results.

One project was an online store with 50,000 inquiries per month. After implementing context handoff, handling time decreased by 35%. Agents stopped wasting time on clarifying circumstances, NPS increased by 12 points. Savings on agent FTE reached 40%. Typical annual savings range from $50,000 to $200,000 depending on ticket volume. The handoff module can save your company an average of $100,000 annually based on a mid-size deployment.

How to Determine the Escalation Moment?

We use two types of triggers: explicit and automatic. Explicit — when the customer writes "agent", "live person", or presses the call button. Automatic triggers fire based on behavior:

• Low model confidence (confidence < 0.6) for three consecutive messages.
• Negative sentiment with deterioration (score < -0.7, trend=worsening) — frustrated customer detection.
• Topic falls into "always escalate" list: legal claims, threats, VIP customers.
• Cyclic dialog — the user repeats the question in different words, the bot is stuck.

The handoff detection completes in under 3 seconds.

class EscalationDetector:
    def should_escalate(self, dialog: Dialog) -> EscalationReason | None:
        if self.explicit_request_detected(dialog.last_message):
            return EscalationReason.EXPLICIT_REQUEST

        if dialog.bot_confidence_history[-3:] == [low, low, low]:
            return EscalationReason.LOW_CONFIDENCE

        sentiment = self.sentiment_analyzer.analyze(dialog.last_5_messages)
        if sentiment.score < -0.7 and sentiment.trend == "worsening":
            return EscalationReason.FRUSTRATED_CUSTOMER

        return None

According to Zendesk Handoff API documentation, combining an ML model and rules increases escalation accuracy to 92%.

Comparison of approaches: The ML model is 1.3 times more accurate than the rule-based approach (92% vs 70%). That's 31% higher. The model captures hidden dissatisfaction better.

What Is Included in the Context Package?

The context package contains:

In the agent's interface, data is visualized: a dashboard with emotion color coding, customer card, dialog timeline. The agent sees the problem in seconds.

Call Routing and Waiting

On escalation, the bot places the customer in a queue for an agent with the required skills and priority. While the customer waits:

• The bot informs of the estimated wait time (estimated based on queue history).
• Offers to leave contact details for a callback — the agent will call back.
• Continues answering simple questions so the customer doesn't leave.

Average wait time for an agent is 45 seconds. The system supports up to 500 concurrent customers. In a typical deployment, first response time improves by 55%. The system can handle 10,000 simultaneous chatbot sessions and transfer up to 500 customers per hour to agents.

async def initiate_handoff(dialog: Dialog, reason: EscalationReason):
    available_agent = await agent_queue.find_available(
        skills=classify_required_skills(dialog),
        priority=get_customer_priority(dialog.user_id)
    )

    wait_time = await agent_queue.estimate_wait(available_agent)
    await bot.send(dialog.channel, f"Connecting with an agent. Please wait ~{wait_time} min.")

    await agent_dashboard.notify(available_agent, {
        "dialog": dialog,
        "reason": reason,
        "customer_profile": await crm.get_profile(dialog.user_id),
        "summary": await ai.summarize_dialog(dialog)
    })

Transfer Back to Bot

After the agent conversation ends, the bot can take over the dialog with updated context: what the agent resolved, what data was clarified. This reduces support load for repeat inquiries. Escalation statistics accumulate — regular analysis of reasons helps update the knowledge base, retrain models, and fix gaps in bot functionality.

Integration with Helpdesk Systems

We connect handoff to any popular platform:

• Zendesk: Handoff API, ticket creation with context.
• Freshdesk: Agent SDK for transferring context to agents.
• Bitrix24: Live Chat API, agent queues, unified CRM.
• Custom platforms: WebSocket + REST API.

Commercial Deliverables

1. Audit & Setup: Audit of current bot scenarios and configuration of escalation triggers.
2. Detector Development: Development of a detector (rules + ML model) for your use case.
3. Integration: Integration with CRM and helpdesk, setup of the agent dashboard.
4. Testing: Unit tests for the detector, integration tests for the handoff flow.
5. Documentation: Comprehensive documentation of the process, system architecture, and API references.
6. Training: Two training sessions for agents and administrators.
7. Access: Dashboard access with real-time handoff monitoring.
8. Support: 30-day post-launch support including monitoring and fine-tuning.

As a result, you get a fully configured handoff module, documentation, and a trained team.

Request a consultation on handoff integration. Get a demo of the functionality on your scenario. Our experience: 5+ years in AI solutions, more than 20 successful handoff integrations.

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.