How much does AI consulting cost?

BlokAI's engagements range from $5,000 to $75,000 depending on scope. Simple automations and LLM integrations start around $5K–$15K. Custom AI agent systems and full data infrastructure projects run $30K–$75K. We scope every project before committing to a price.

What does BlokAI build?

BlokAI builds AI agents, LLM applications, RAG systems, automation pipelines, and data infrastructure for startups and SMBs. We take projects from architecture to production — not demos, not prototypes.

Is BlokAI based in Japan?

Yes. BlokAI is a solo studio run by Kent Tran, based in Tokyo, Japan. We work with clients globally in English.

What is RAG and why does my business need it?

RAG (Retrieval-Augmented Generation) lets AI systems search your own documents, databases, or knowledge bases before answering — instead of relying only on training data. It powers AI customer support, internal search tools, and sales enablement systems that are accurate and up-to-date.

How to Build an AI-Powered Lead Intelligence System (With Real Numbers)

"Anthropic has closed a $3.5 billion Series E funding round, reflecting strong enterprise demand for Claude and reinforcing Anthropic's position as a leading AI safety company." — TechCrunch, March 2025

This post describes a real project: an AI-powered lead intelligence system we built for a B2B professional services firm. I'm sharing it with the client's permission, with some identifying details changed.

The reason I want to write this in detail is that most AI case studies are either vague ("we improved efficiency by a significant amount") or unrealistically positive. I want to give you the actual numbers, the actual timeline, and the parts that didn't go as planned.

The Client and the Problem

A professional services firm — 40 staff, focused on the Japan and Southeast Asia market — generates most of its business through direct outreach to potential enterprise clients. Their business development process required significant manual research before each outreach: understanding the target company's situation, recent news, technology context, and fit for their services.

Before the project: each BDR (business development representative) spent 25-35 minutes researching each lead before writing an outreach email. With roughly 30-40 leads per BDR per week, this was 12-15 hours/week per person on research — nearly a third of their working week before they'd written a word of outreach.

The goal: automate the research phase so BDRs could focus on judgment, relationships, and actual outreach rather than information gathering.

The System We Built

The lead intelligence system is a multi-step AI agent that runs automatically when a new qualified lead is added to the CRM.

Step 1: Company intelligence gathering

The agent retrieves the company's website, LinkedIn page, and recent news mentions. It extracts: company size and growth stage, recent product launches or strategic announcements, leadership changes, technology signals (tech stack mentions, integration announcements), financial context if available (funding rounds, public filings).

Step 2: Contact intelligence

For the specific contact at the company, it gathers: LinkedIn profile summary, recent posts or public content, professional background, tenure at the company.

Step 3: Relevance scoring

The agent assesses relevance to the client's specific service offering against a defined criteria set: industry fit, company stage, geographic presence, typical pain points. It outputs a structured score with reasoning.

Step 4: Research brief generation

A structured one-page brief is generated covering: company overview, contact context, relevance assessment, suggested outreach angles, specific conversation hooks based on recent news or announcements.

Step 5: CRM update

The brief is written directly to a custom field in the CRM, attached to the lead record. The BDR opens the record and finds it already populated.

Technology stack:

Orchestration: LangGraph
Web research: Tavily search API + custom scrapers
Model: GPT-4o for research synthesis, GPT-4o mini for structured extraction tasks
CRM integration: HubSpot API
Hosting: AWS Lambda (event-triggered on CRM webhook)

The Numbers

Build time: 9 weeks (including 2 weeks of data quality issues with CRM data that required cleanup before the system could reliably identify the right contacts) Build cost: ¥4,200,000 (~$28,000 at project-time exchange rates) Infrastructure cost: ~¥35,000/month (~$230/month) for API costs, hosting, and monitoring at current usage levels Time before: 30 minutes average research per lead. 35 leads/BDR/week. 3 BDRs = 52.5 hours/week of research time across the team. Time after: ~3 minutes for a BDR to review the brief and personalise their outreach. 3 BDRs = 5.25 hours/week. Reduction of 47 hours/week. Financial impact: 47 hours/week × ¥5,500/hour average loaded BDR cost = ¥258,500/week = ~¥13.4 million/year in recaptured time. Payback period: The infrastructure investment paid back in roughly 10 days of operation. The build cost paid back in about 4 months. Outreach quality impact: The BDRs report that the quality of their outreach improved significantly because they're starting from a brief rather than their own imperfect recall of rushed research. In the 3 months since launch, outreach response rates improved from 6.2% to 9.1% — a 47% lift. We attribute roughly half of this to the research quality and half to the fact that BDRs now have more time and energy for crafting their actual outreach.

What Didn't Go as Planned

CRM data quality: The existing CRM had significant data quality issues — inconsistent company naming conventions, duplicate records, missing fields. We spent two weeks cleaning and standardising CRM data before the agent could reliably trigger and work correctly. This wasn't scoped upfront and added time and cost. LinkedIn rate limiting: We initially planned to scrape LinkedIn directly for contact intelligence. LinkedIn's aggressive rate limiting made this unreliable. We pivoted to using a third-party LinkedIn data API (with appropriate licensing), which added ~$150/month in ongoing cost but made the contact intelligence reliable. Brief quality on some company types: The system produces excellent briefs for mid-to-large companies with good web presence. For small, private companies with minimal online footprint, the briefs are shorter and less insightful. We flagged these explicitly in the output rather than pretending the brief was equally good — honesty about confidence level is important for user trust. Initial user resistance: One of the three BDRs initially didn't trust the system and kept doing manual research in parallel. After seeing that the system's briefs were accurate (verified against their own manual research) for 2-3 weeks, they switched to trusting the system fully.

The Takeaways

This project is a good example of where AI agent automation clearly wins: high-frequency, research-intensive workflow; clear and measurable time savings; immediate business value.

The lessons that generalise: budget for data quality cleanup, have a plan for third-party data sources, be honest about where the system underperforms, and build in a trust-building period with users.

The ROI math isn't always this clean. But for the right workflow, it's very clean indeed.

Want to implement AI in your business? Book a 30-minute architecture call at blok-ai.com.