How to build a lead generation system from scratch: the six steps I use for clients

What results can a lead generation system like this produce?

I'm David Walker-Dobson, a GTM engineer, and I run Tentt, a GTM engineering agency. I've been running it for over three years, building systems for finance firms, search funds, private equity, data and AI companies, marketing agencies and security companies. This guide is the written version of my video on how I build $100k lead generation systems as fast as possible, starting from scratch.

Two examples. For a UK marketing company, one campaign sent 369 emails and generated 10 opportunities, about one opportunity for every 37 emails, follow-ups included. Across five campaigns we sent 10,493 emails for 51 opportunities. For a brand-new search fund with no pipeline, no list and a CRM they had coded themselves with Claude, the system booked three meetings in its first week, and it still runs today.

Every tool either client used, you could sign up for this afternoon. Most people buy the same tools and get nothing. The rest of this guide is about the difference.

Why don't tools like Instantly, Apollo and ListKit make a system?

Most business owners start by buying Instantly or Smartlead to send, and ListKit or Apollo for data. They have data and a way to send, so it feels like a system. But AI has made setting up a sending tool and pulling a list an afternoon's work. If you can do it, everyone can, so it is not an advantage any more.

What makes the difference is the thinking first. In the marketing company's campaigns, generic inboxes (info@, hello@, office@) beat named owner inboxes: one opportunity per 100 sends against one per 266. We built the list by pulling local businesses (window installers, solar installers and similar) out of search results with Serper.dev, then finding the owner's email. Where we couldn't, most people would stop. We used AI agents to pull the generic inbox from the business's website instead. Hardly anyone prospects those inboxes, so the email gets read. Serper.dev is cheap and well known. The edge was one extra layer of thinking.

The search fund needed owners close to retirement, and age is not on a website. We used Clay and a scraper to find the year each owner finished university, estimated their age from it, stacked that with owner and founder job titles and family-business language on the website, and sorted by a succession score. Clay calls this kind of custom data GTM alpha: an answer to a question nobody else has asked.

Step 1: how do you understand the people you sell to and why they buy?

Most B2B purchases involve more than one person. Say I'm selling a GTM system into a 20 to 30 person B2B company. The founder signs because they care about pipeline. The sales reps use it and care about good leads. The ops lead who owns HubSpot can't say yes, but can absolutely say no, and they care about clean data.

Talk only to the founder about founder problems and the ops lead can kill the deal later. So write down who signs, who uses and who can block, and what each one cares about. For the search fund, the people insight was the owner's age. It mattered more than company size and shaped everything that came after it.

Step 2: how do you design a lead generation system with the end in mind?

Start with the number. How many meetings a month do you want? Work backwards to the replies, sends and leads that takes, and you know how big the system needs to be, and whether your market is large enough for it.

Then look at what you have to work with. Design around what you have or can afford, not a wish list:

• Budget for ads, and where the ads send people: a booking page, a webinar or a lead magnet. Each needs a different system behind it.
• People: SDRs, cold callers, or a founder who can sell.
• Organic content: LinkedIn, YouTube, SEO and AI search.
• Assets: webinars, guides, case studies, an offer people want.
• The CRM you're on.

Most people who click an ad never fill in a form. Website visitor identification such as RB2B or Vector tells you who they were, so they can be scored and routed to outbound follow-up or retargeting instead of disappearing. On the outbound side, run your full market list at volume, then layer signals (hiring, funding, new leaders, site visits) on top: hot signals go to an SDR, warm ones get priority email and LinkedIn.

Then the offer, the most important part. Is it strong enough to sell to a complete stranger? If yes, push it and ask for the call. If not, use outbound to push content, webinars, assets and free trials, let people engage, then follow up. That path is longer, but a weak offer with strong assets can still win.

Finally, check the five parts every system needs: a trigger (something tells you to act), data (who they are and how to reach them), a filter (are they a fit), an action (email, call or LinkedIn) and a record (everything lands in the CRM). Design it right to left, starting from the meetings.

Step 3: how should you pick your lead generation tools?

Pick tools last, but pick carefully. Tools aren't the edge, but a bad pick can slow or break a good design. Choose one tool for each job:

Orchestration is the layer most people miss, and it is the engineering in GTM engineering. Without it you have a pile of tools that don't talk to each other. With it, when a signal fires the lead is found, enriched, verified, scored, sent the right message and logged in the CRM, without anyone copying and pasting. See our GTM engineering tech stack guide for more on each layer.

Step 4: how do you build the outbound part of the system?

Start with a list. There are two kinds. A dynamic list is signal based and changes every day: hiring, funding, new leaders, site visits, or someone posting about a problem you solve. A static list is built once from filters (industry, size, location, job title) and worked through. Everyone has the same databases now, so sharpen static lists with data nobody else has, like the succession score, or build them from places your buyers gather: directories, Facebook and LinkedIn groups, Slack communities, webinar and event attendees.

Every list goes through the same steps before anything is sent:

1. Qualify each company with AI using true or false questions (see AI lead qualification).
2. Find the right people at each company.
3. Find and verify contact details with a waterfall: provider two only if one missed, provider three only if two missed, then verify. You pay for hits, your bounce rate stays low and your sending domains stay healthy (see waterfall enrichment).
4. Remove existing customers and open deals.
5. Split by signal, so each group gets its own message.

For copy, lead with the signal or the relevance, make it about them, keep it short enough to read on a phone, and make the ask an easy yes. Don't ask a stranger for 30 minutes. If they're hiring two SDRs, offer a free list of warm prospects for the new reps to call. For the search fund we also built a copywriting skill that knows their offer, voice and buyer, so every message starts from the same rules.

None of this matters if you land in spam. Use separate sending domains, never your main domain. Set up SPF, DKIM and DMARC, keep a few inboxes per domain, warm them up for two to three weeks, keep volume per inbox low, and check bounce and spam rates every week. I buy domains from Porkbun and inboxes from ScaledMail (the one I use most), Zapmail or Cheap Inboxes. More in our cold email deliverability guide.

Cold calling usually fails before anyone picks up: no mobile number, or a number flagged as spam. Run a phone waterfall (Prospeo has the best mobile coverage I've seen in Europe, LeadMagic in the US), then use a parallel dialer such as Trellus or Salesfinity. Open with the signal and ask permission: "I saw you're hiring an SDR, so I'll be quick. Can I take 30 seconds to tell you why I called?"

On LinkedIn, connect first and pitch later, timed alongside your emails so the person sees one story. lemlist runs email and LinkedIn in one sequence. HeyReach and Aimfox suit running many LinkedIn accounts. For video, Tella records clean, high-quality videos for prospects.

Step 5: how do you build the inbound part of the system?

Inbound starts in three ways, and each needs a different next step:

• Form fill: enrich and qualify, then a good fit gets a call from an SDR.
• Calendar booking: enrich and qualify before the call, not after, so you arrive with research and can cancel a bad-fit call.
• Asset opt-in: enrich and qualify, then a nurture sequence or an SDR call depending on fit.

Step 6: how do you track what's working with CRM attribution?

Every reply, call and meeting goes back to the CRM. This is why knowing HubSpot or Attio well matters more than people think. Attribution tells you which signal, channel, message and rep booked each meeting, so you double down on what works and cut what doesn't.

Picture a signal that reached 300 people and booked zero meetings. Without attribution you'd keep paying for it forever, because everyone says you need signals, when the meetings were really coming from volume outreach or one rep. Or two reps making about the same number of calls, where one books twice as many meetings. What are they saying differently? You can't coach what you can't see.

Get a lead generation system built