The Day We Built the 'Perfect' AI Lead Gen Tool (And Why We Killed It)

After a couple of weeks of vibe coding, testing and debugging, we had finally nailed it. Paste in a list of prospects, click a button, and watch as AI generated cold emails so personalised they'd think you went to university together.

The promise: Our AI would scrape company websites, analyse LinkedIn profiles, find recent news, and craft openers that felt like catching up with an old friend. Every SMB's dream - personalisation at scale.

Four weeks and roughly 200 debugging sessions later, we killed the entire project.

Here's why.

Week 1: The Honeymoon Phase

The prototype worked perfectly on our test data. We scraped our client's website, pulled our client's LinkedIn profiles, and generated emails that were a little odd but we thought we could tweak with more prompt amendments.

"Tom, I see you just posted about API rate limits on LinkedIn - funny timing given our recent push into AI tooling. Would love to chat about how you're handling the technical challenges."

Good, right?

Week 2: Meet Reality, She's Not Friendly

First test. Pulled 50 prospects from their CRM.

• 31 websites returned 403 forbidden errors

• 12 had Cloudflare protection we couldn't bypass

• 5 were behind login walls

• 2 actually scraped successfully

"It's fine," we said. "Let's lean more on LinkedIn data."

Then LinkedIn's rate limiting kicked in. Hard.

After 10 profile views, we hit a wall. The Enterprise Sales Navigator API? Starts at £8,000 per year. Our SMB clients just don't have the budget for that.

Week 3: The Pivot Dance

"What if we add in news APIs for personalisation instead?"

Great idea, except small businesses rarely make the news. We searched for 100 SMB prospects:

• 3 had recent news articles

• 2 were about disolving the business

• 1 was the founder's wedding announcement

We tried everything:

• Rotating proxy servers (expensive and sketchy)

• Browser automation tools (LinkedIn detected them immediately)

• "AI-powered" data providers (just reselling the same stale databases)

Each solution created two new problems. Our simple tool was getting to the point where it would require a DevOps engineer to maintain.

What Actually Works (Based on Real Client Data)

After killing our Frankenstein creation, we surveyed 50 B2B SMB businesses. Here's what actually moves the needle:

1. Cold Email with Decent Lists

• Average response rate: 1-3%

• Tools like Apollo.io provide good enough data

• Basic personalisation (name, company, industry) is sufficient

• Consistency beats perfection

2. AI for List Building

• Finding companies that match ICP criteria

• Enriching data with employee counts, tech stack

• This is where AI SMASHES it - pattern matching at scale

3. Manual LinkedIn Outreach (or VA-Assisted)

• 20-30 connection requests daily

• Personal messages for accepted connections

• Slower but higher quality conversations

4. Partner Channels

• Still the highest ROI channel

• One good partner equals 100 cold emails

• AI can't replace relationships

The AI Tools That Actually Earn Their Keep

Instead of chasing lead gen magic, we help clients implement AI tools that enhance existing processes:

Meeting Intelligence (ROI: 45 minutes saved daily) We recently implemented Fireflies.ai for a recruitment client. Result:

• No more manual note-taking

• Auto-extracted action items

• Follow-up emails drafted automatically

• Client feedback: "It's like having a PA in every meeting"

Proposal Generation (ROI: 2-3 hours per proposal) Built a simple tool using GPT-4:

• Input: Meeting transcript + company data

• Output: First draft proposal with pricing

• Still needs human review, but cuts creation time by 70%

Email Reply Management (ROI: 30% more leads converted) Categorises inbound emails:

• Hot leads → Alert sales immediately

• Questions → Draft technical responses

• Not interested → Polite follow-up in 3 months

• Spam → Archive

Content Repurposing (ROI: 5x content output) One blog post becomes:

• 5 LinkedIn posts

• 10 tweets

• 3 email newsletters

• 1 slide deck

• All maintaining the original voice

The Pattern We Kept Missing

Every successful AI implementation had three things in common:

1. Enhanced existing workflows rather than replacing them

2. Saved time on repetitive tasks humans already did

3. Required minimal behavior change from users

Our failed lead gen tool? It required companies to completely change their outreach process, maintain complex infrastructure, and trust a black box with their first impression.

The Uncomfortable Truth

Here's what we tell clients now:

"AI won't magically fill your pipeline. It won't replace relationship building. It won't turn bad salespeople into good ones.

But it will give your good salespeople more time to sell. It will help your marketing team create more content. It will ensure no hot lead sits in your inbox for three days."

Less sexy? Sure. Actually valuable? Absolutely.

People Also Ask

Q: Can't enterprise scrapers solve the technical challenges? A: Yes, tools like Bright Data or ScraperAPI work better. But they cost £300-500/month and still require technical maintenance. For that price, SMBs could hire a part-time SDR.

Q: What about competitors offering AI personalisation? A: Most are hitting the same walls. The successful ones use pre-scraped databases (often outdated) or focus on enterprise clients who can afford the infrastructure.

Q: Is personalisation dead? A: No, but the bar is lower than we think. Mentioning someone's actual job title and company pain point beats generic spray-and-pray. You don't need to know everything about their careers, homes and hobbies.

Q: What's the minimum viable AI stack for B2B SMBs? A: Meeting recorder (£20/month) + Email AI assistant (£30/month) + Content repurposing tool (£50/month). Total: £100/month for measurable time savings.

Q: Should we stop innovating in lead generation? A: Innovate in lead quality, not lead gymnastics. Better targeting, clearer value props, and stronger follow-up beats clever personalisation tricks.

Six Months Later

Our abandoned "perfect personalisation" tool sits in a private GitHub repo, a monument to solving the wrong problem with the right technology.

The irony? We've had three inbound leads ask if we can build them an AI tool for "hyper-personalised outreach at scale."

We politely decline and send them this post instead.

At RicochetB2B, we've learned to ask a better question: "What repetitive task is eating your team's time?" That's where AI shines. Everything else is just expensive theatre.

Want to explore AI tools that actually work? Let's talk about your current workflows, not your automation fantasies