Architecture · AI Architecture for B2B Sales

This time I would like to share (1) the first workflow I built, (2) the design rationale behind it, and (3) how the production implementation in Next.js inherited the workflow architecture originally designed in n8n.

The first building block of my AI architecture for B2B sales is not an AI email generator or chatbot. Instead, I started with something much less exciting—a Product & ICP Profiler.

---

I have been a hunter salesperson in the IT field for decades. Even after becoming one of the developers on my small team, I am still a hunter for business.

One challenge I see in corporate sales today is:

Why do buyers ignore so much outreach?

Several industry studies paint a fairly consistent picture:

• 73% of buyers ignore outreach because it is not relevant.
• 74% feel they know the product better than the salesperson contacting them.
• Yet only 38% of sales representatives say they feel well prepared before speaking with prospects.

When I put those numbers together, I started from the foundation:

The solution should help sales reps (and the AI copilot itself) become better prepared before any outreach.

Particularly because AI is a clever but naïve salesperson.

Understand your product and your strengths before deciding how to sell it. Right?

At the very least, don't let your client know your product better than you do during the outreach. That lesson has stayed with me since my early days in enterprise sales, and it still shapes how I design AI workflows today.

That thinking eventually became the first workflow in SignalLeading.

---

The following short demo shows how the Product & ICP Profiler analyses a product. Rather than asking AI to generate sales content immediately, the workflow first builds an understanding of the product, customer requirements, and the Ideal Customer Profile (ICP).

If you would like to explore the Product & ICP Profiler in more detail, including screenshots, workflow outputs, and industry statistics, you can find it here: Product & ICP Profiler.

This foundation is then reused by later workflow modules.

---

As mentioned, I wanted to make sure the AI copilot understood what it was selling before it tried to persuade anyone.

Just as when we onboard a new sales representative, we don't immediately ask them to write outreach emails.

We first help them understand the product:

→ What problems does it solve?

→ Who benefits the most?

→ What pain points does it address?

Only after that foundation is established does the second stage begin identifying the Ideal Customer Profile (ICP) and the business pains that could be most relevant to our solution capabilities.

---

The n8n workflow gave me a clear structure for the proof of concept. It is now implemented in Next.js as a production application. In Next.js, the same design is executed stage by stage, as shown below in app/api/signalleading/route.ts.

// app/api/signalleading/route.ts

import { s01ProductCatReqt } from '@/lib/qualifier/s01_product_cat_reqt';
import { s02IcpPainSolved } from '@/lib/qualifier/s02_icp_painsolved';

export async function POST(req: Request) {
  const { stage, userInput, history } = await req.json();

  switch (stage) {
    case 'stage1-ProductCatReqt':
      // ...
      return Response.json({
        nextStage: 'stage2-IcpPainSolved',
        result: s1,
      });

    case 'stage2-IcpPainSolved':
      // ...
      return Response.json({
        nextStage: 'stage3-ClientProfile',
        result: s2,
      });

    // The following stages are covered later in this series:
    // - Client profiling
    // - ICP matching
    // - Client signals
    // - AI verdict
    // - Door opener generation

    case 'stage3-ClientProfile':
    // ...

    case 'stage4-IsClientIcp':
    // ...

    default:
      return Response.json({ error: 'Invalid stage' }, { status: 400 });
  }
}

Why n8n → Next.js?

As I shared in my previous articles,The n8n Workflow Behind SignalLeading and Build an Automated n8n Workflow That Integrates Your Daily Applications in Minutes, I see n8n as an excellent environment for experimenting with workflow logic before committing to application code.

---

If you are interested in GTM engineering, AI workflow design, or simply want to exchange ideas, I would love to connect.

You can also tell me about your product. I will show you the results generated by the Product & ICP Profiler.

---

In the next article, Why I Never Just Ask AI to “Write an Outreach Email” in One Shot, I will explain why I split the workflow into multiple AI reasoning stages, how AI hallucinations affect B2B sales, and the thinking behind SignalLeading's design.

---

1. 73% of buyers ignore irrelevant outreach — Salesforce — State of the Connected Customer
2. 74% of buyers know the product better than the salesperson — Forrester — The Death of a (B2B) Salesman
3. 38% of sales reps feel well prepared before speaking with prospects — Salesforce — State of Sales

#AIArchitecture#B2BSales#GTMEngineering#n8n#NextJS#WorkflowAutomation#AIEngineering#SignalLeading#BuildInPublic