AI Agent for B2B Sales: Lead Qualification and Personalization

Every day, SDR managers spend hours qualifying cold leads, answering the same questions about budget, authority, need, and timeline. Scripted chatbots fail when faced with unexpected answers or multi-step objections. We built an AI agent powered by an LLM that handles the first line of communication, qualifies leads using the BANT methodology, and passes warm contacts to managers with full dialog context. Our experience shows such an agent cuts first response time from hours to minutes and boosts pipeline by 20–30%.

What the AI Agent Solves: Common Sales Problems

• Slow first response. Leads wait hours for a reply—enough time to turn to competitors. The agent responds within 2–3 minutes, 24/7.
• Poor qualification of incoming traffic. Managers spend 60% of their time on dead-end leads without budget or authority. The agent filters them out early.
• Impersonal outreach. Mass campaigns yield <1% conversion. The agent generates emails tailored to company data: industry, size, recent news.

Compare the AI agent to a traditional chatbot:

How the AI Agent Qualifies Leads

The agent starts a conversation by asking natural questions embedded in the flow. It doesn't fire all questions at once; it gradually uncovers budget, authority, need, and timeline. Based on answers, the lead score is updated in real time. When the score exceeds 70, the agent suggests a demo call and hands off to the manager with full context. If the score is below 30, the lead goes into a nurture sequence. This reduces SDR workload and improves conversion.

Step-by-Step Setup of an AI Agent for Qualification

1. Define qualification criteria. Choose a methodology: BANT, GPCT, or your own. Specify what data is needed for scoring.
2. Collect historical dialogues. You need at least 50–100 successful and unsuccessful conversations for few-shot learning.
3. Develop the prompt and schema. A Pydantic model to validate fields (budget, authority, need, timeline). Example below.
4. Integrate with CRM. REST API to write leads and update statuses.
5. A/B test. Compare agent conversion with SDR on a control group.

How Our Agent Works: Code and Stack

We use gpt-4o and Pydantic models for response validation. The prompt (in Russian) sets the agent as a B2B SaaS salesperson following BANT. When the score exceeds 70, the agent calls the schedule_demo function—transferring the lead to the CRM with full context.

from openai import OpenAI
from pydantic import BaseModel
from typing import Optional
import json

client = OpenAI()

class LeadQualification(BaseModel):
    lead_score: int           # 0-100
    budget_fit: bool
    authority_confirmed: bool
    need_identified: bool
    timeline_clear: bool
    next_action: str          # "schedule_demo", "nurture", "disqualify"
    disqualify_reason: Optional[str]
    key_pain_points: list[str]
    notes: str

QUALIFICATION_SYSTEM_PROMPT = """You are a B2B SaaS sales manager.
Your task is to qualify incoming leads using the BANT methodology:
- Budget: is there a budget for the solution?
- Authority: are you talking to a decision-maker or influencer?
- Need: is there a real business need?
- Timeline: when do they plan to implement?

Conduct a natural conversation. Don't ask all questions at once—weave them into the dialogue.
Record answers and update the qualification as the conversation progresses.
On positive qualification (score > 70) - suggest a demo call.
On score < 30 - politely end and add to nurture sequence."""

def sales_agent_response(session_id: str, user_message: str, conversation_history: list) -> dict:
    conversation_history.append({"role": "user", "content": user_message})

    # Generate agent response
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[
            {"role": "system", "content": QUALIFICATION_SYSTEM_PROMPT},
            *conversation_history,
        ],
        tools=[
            {
                "type": "function",
                "function": {
                    "name": "update_lead_qualification",
                    "description": "Update lead qualification based on new information",
                    "parameters": LeadQualification.model_json_schema(),
                }
            },
            {
                "type": "function",
                "function": {
                    "name": "schedule_demo",
                    "description": "Offer a time slot for a demo",
                    "parameters": {
                        "type": "object",
                        "properties": {
                            "lead_name": {"type": "string"},
                            "lead_email": {"type": "string"},
                        },
                        "required": ["lead_name", "lead_email"],
                    }
                }
            }
        ],
    )

    # ... process response
    return {"response": response.choices[0].message.content}

Personalized Outreach

For email generation we use the same LLM but with a separate prompt. For each email, the system fetches enrichment data from our API (industry, employee count, recent news).

Sequence template:

def generate_personalized_outreach(lead_data: dict, sequence_step: int) -> str:
    """Generates a personalized email based on company data"""
    # Enrich company data via API (Dadata, LinkedIn)
    company_info = company_enrichment_api.get(lead_data["company_domain"])

    sequence_contexts = {
        1: "First contact - introduction and value proposition",
        2: "Follow-up after 3 days - specific pain point based on company data",
        3: "Follow-up after a week - social proof (case study from the industry)",
        4: "Final email - direct meeting offer",
    }

    email_content = client.chat.completions.create(
        model="gpt-4o",
        messages=[{
            "role": "system",
            "content": f"""Write a personalized sales email (step {sequence_step}).
Context: {sequence_contexts[sequence_step]}
Tone: professional but not formal. No generic phrases.
Length: 100-150 words."""
        }, {
            "role": "user",
            "content": f"""Lead data:
Name: {lead_data['first_name']}
Title: {lead_data['title']}
Company: {lead_data['company']}
Industry: {company_info.get('industry')}
Company size: {company_info.get('employee_count')} employees
Recent news: {company_info.get('recent_news', 'no data')}"""
        }],
        temperature=0.7,
    )

    return email_content.choices[0].message.content

Why an AI Agent Beats a Scripted Bot

A traditional bot follows rigid scripts—any deviation breaks the flow. An LLM-based agent understands context, rephrases questions, and adapts tone to the conversation partner. Results from our practice: first response time cut 130x (from 6.5 hours to 3 minutes), qualified leads up 34% because the agent pushes through ambiguous cases.

"Thanks to the agent, our SDRs stopped wasting time on cold screening and focused on high-level consultations." — Client's head of sales.

Case Study: B2B SaaS Lead Generation

One of our clients sells an ERP system for mid-market businesses. Inbound flow: 200+ leads per month; a team of 4 SDRs couldn't cope. We deployed the agent to handle initial conversation and qualification.

Agent handles: first response to inbound request, qualification dialogue (3–7 messages), scoring, passing hot leads to SDR with full context.

Results in the first 3 months:

• Time-to-first-response: 6.5 h → 3 min
• Qualified leads passed to SDR: +34%
• Demo-scheduled conversion: 18% (agent) vs 22% (SDR) — marginally lower
• SDR focus: shifted from L1 qualification to working already warm leads
• Revenue pipeline: +28% quarter-over-quarter (additional revenue of several million rubles)

Limitation: the agent does not handle final negotiations or work with C-level corporate clients—only qualification and handoff.

What's Included in the Work

Total: 7–10 weeks. Cost is calculated individually for your business. Get a consultation to estimate the scope.

Objection Handling

OBJECTION_HANDLERS = {
    "price": "I understand the cost concern. Let's look at ROI—our clients typically recoup the investment within {payback_period} months thanks to {key_benefit}. Would you like me to show a calculation for your scale?",
    "timing": "I understand this might not be the best moment. When would be a good time to revisit the conversation? We can set a reminder for {suggested_date}.",
    "competitor": "I hear you're considering {competitor}. We work with several companies that switched from them to us—I can share their experience. What's most important to you when choosing?",
    "no_need": "Interesting to hear—most of our clients felt the same way until they discovered {pain_point}. How are you currently solving {relevant_challenge}?",
}

Our Experience and Guarantees

Our team has been developing AI agents for over 5 years. We have 25+ successful lead generation projects. We guarantee uninterrupted operation of the agent and provide post-launch support. Want to try the AI agent on your data? Contact us for a demo.

LLM Development: Fine-Tuning, RAG, Agents, and Production Deployment

Using GPT‑4 or Claude 3.5 Sonnet through a public API is not a solution — it's just a tool. When the requirement is to "make it like ChatGPT, but on our data," there is a real engineering challenge behind it: from prompt engineering to training a 70B model on your own infrastructure. End-to-end LLM solution development is a complex stack, and we have been doing it for over 5 years. During this time, we have completed over 20 projects in generative AI: from RAG systems for legal departments to custom support agents. Where exactly your task falls depends on data, latency requirements, budget, and how critical confidentiality is.

A typical situation: the client has already tried ChatGPT, but results are unstable — sometimes accurate, sometimes hallucinating. Or they need integration into a corporate portal while complying with security policies. Let's break down each layer of the stack in detail — from RAG to production deployment.

Why Do RAG Systems Break and How to Fix It?

RAG (Retrieval-Augmented Generation) looks simple: find relevant documents, put them in context, get an answer. In practice, it fails in several places.

Chunking without overlap. Classic mistake: chunk_size=512, overlap=0. If the answer lies across two chunks, retrieval won't find either with sufficient confidence. Solution: overlap 15–25% of chunk_size, or better yet, sentence-aware splitting with spaCy or NLTK instead of naive character splitting.

Poor embedder. text-embedding-ada-002 is good for general use, but on legal or medical texts, specialized models like E5-large-v2, BGE-M3, or fine-tuned sentence-transformers on domain data outperform it. Recall@5 differences can be 15–25%.

No re-ranking. Vector search optimizes for speed, not relevance. A cross-encoder re-ranker (ms-marco-MiniLM-L-6-v2, bge-reranker-large) after initial retrieval improves top-3 accuracy with acceptable latency (+50–150ms). This is often more impactful than improving the embedding model.

Hybrid search. Dense vectors alone work poorly on exact queries: names, SKUs, codes. BM25 (sparse) finds exact matches but misses semantics. Hybrid via RRF (Reciprocal Rank Fusion) is the optimal compromise. Qdrant, Weaviate, and pgvector 0.7+ support hybrid search natively.

When to Fine-Tune Instead of Prompt Engineering?

Prompt engineering solves ~70% of LLM adaptation tasks for a domain. The remaining 30% require fine-tuning. Three indicators: the model ignores a specific output format even with detailed prompting; the task requires deep knowledge of specialized vocabulary (medicine, law); you need to significantly reduce token costs by replacing a large model with a smaller specialized one.

LoRA and QLoRA are the standard for SFT. LoRA adds trainable low-rank matrices to attention layers. A typical configuration for Llama-3 8B: r=64, lora_alpha=128, target_modules=["q_proj","v_proj","k_proj","o_proj"] yields ~0.8% trainable parameters, training on one A100 40GB. QLoRA adds 4-bit quantization (NF4) and allows fine-tuning 70B models on two A100 40GB, though speed drops by half compared to bf16.

DPO instead of RLHF. Direct Preference Optimization requires only (chosen, rejected) pairs, not scalar reward signals. DPOTrainer from the trl library (Hugging Face) implements it in a few dozen lines.

Common mistake. A dataset of 500 examples, 5 epochs, validation loss 0.8 — seems fine. But on test, the model degrades on general instructions. Cause: catastrophic forgetting. Solution: add 10–20% general instruction-following examples (Alpaca, FLAN) to the training set to preserve original capabilities.

How to Choose a Base Model: 8B or 70B?

For most tasks, fine-tuning an 8B model is sufficient. 70B is needed when deep reasoning is required or the 8B baseline does not reach the required quality even after fine-tuning. Inference cost for Llama-3 8B via vLLM on A100 is efficient; the exact cost depends on volume.

What Does PagedAttention Bring to Production?

vLLM is the first choice for serving open-source models. PagedAttention is the key technical innovation: KV-cache is managed like virtual memory in an OS, without fragmentation. This yields 2–4x higher throughput compared to naive HuggingFace Transformers inference. The vLLM documentation confirms that continuous batching and PagedAttention are the standard for high-load LLM services.

Typical numbers on A100 80GB for Llama-3 8B (bf16): 400–600 req/s, P50 latency 200–400ms, P99 latency 600–900ms at concurrency 64. For 70B on two A100 with tensor parallelism: 80–120 req/s, P99 latency 1.5–2.5s. AWQ or GPTQ quantization reduces memory consumption by 2x with quality loss within 1–3%.

Multi-Agent Systems

Agents are LLMs with access to tools: search, code execution, API calls, database interaction. Common patterns:

• ReAct (Reason + Act): the model reasons → chooses a tool → observes the result → reasons again. LangChain and LlamaIndex implement it out of the box.
• Multi-agent orchestration: multiple specialized agents with a coordinator on top. Example: coordinator → researcher (search + summarization) → coder (code generation and execution) → critic (verification). Tools: AutoGen (Microsoft), CrewAI, custom implementation on LangGraph.

In production, agent systems are non-deterministic. Essential: guardrails, step limits, logging of each step, human-in-the-loop for critical actions.

How We Work: Stages, Timeline, Deliverables

What Is Included

We deliver:

• Technical documentation (model card, configs, deployment instructions)
• Access to infrastructure (code repository, trained weights)
• 1 month of post-deployment support (consultations, bug fixes)
• Customer team training (2–3 sessions on system operation)

Timeline: basic RAG prototype — 1–2 weeks. Fine-tuning with customer data — 3–6 weeks (including data preparation). Production system with monitoring and retraining — 2–4 months. Cost is calculated individually based on data volume, model complexity, and infrastructure requirements.

We guarantee the quality of the final model with performance benchmarks and ongoing monitoring. Our engineers have hands‑on experience with dozens of production LLM systems.

Want to evaluate your project? Leave a request — we will prepare a preliminary summary within 1–2 business days. Or get a consultation on choosing the approach: RAG, fine-tuning, or hybrid — we will tell you what works best for you. Contact us to discuss your LLM development needs. Schedule a free consultation today.