The lead research workflow below is the one we actually run, not a tidied-up version of it. It is not revolutionary. It is reliable, which matters considerably more when you have numbers to hit.

Everyone says quality over quantity and almost nobody explains how to get there. So here is the process, including the boring parts that make the difference.

Step 1: Apollo, the foundation

Every campaign starts in Apollo, and the filters decide everything downstream.

• Industry keywords that actually narrow the set, not broad categories

• Job titles and seniority that match who really decides

• Company size that fits what is being sold

• Location parameters that make sense for the campaign

The threshold is 1,000 contacts minimum before going any further. Below that the targeting is probably too narrow, or there is not enough volume for the campaign to be worth running at all.

Step 2: Clay, the cleanup

The Apollo export goes straight into Clay to be made ready for enrichment. The checklist is short and none of it is optional.

• Deduplication, and there are always more duplicates than expected

• Email verification, because bounces damage sender reputation

• Company name normalisation, because Apollo gets creative with naming

Then filter for safe-to-send addresses only. It does not matter how good a lead looks on paper. If the address is risky, it is out. Deliverability is worth more than any single contact on the list.

Step 3: Enrichment

This is where most of the return lives, and what you enrich depends entirely on what the campaign needs.

For one client, GPT researches company websites and summarises their mission, flagging any past or upcoming events they mention. That research goes to Claude with a prompt that suggests a relevant event idea, or a way to improve an event they already have planned using the client’s product.

The output is oddly specific. Creative enough to hold attention, practical enough that the prospect can picture actually doing it.

The gap between generic and specific is the gap between “I saw you work in tech” and “I noticed your organisation focuses on youth education, and you mentioned your upcoming annual gala. Here is a way to build an interactive donor experience around student success stories.” One prospect replied asking how we knew they were struggling with exactly that.

Personalisation that survives scale

The point is producing relevant, specific suggestions for hundreds of prospects without brainstorming each one by hand. The model connects a company’s mission to a plausible idea. The strategy and the judgement stay with a person.

It is closest to having a researcher who has read every prospect, never tires, and reliably produces ideas odd enough to be memorable and grounded enough to work.

What one full run looks like

The stages above are easier to judge with the attrition made visible. Rough proportions, not a specific campaign:

• Apollo returns the raw set against your filters.

• Deduplication and normalisation in Clay remove a slice of it, usually a small one.

• Verification removes a much larger slice, because anything not safe to send is dropped rather than risked.

• Enrichment runs on what survives, and anything where the research comes back empty gets held rather than sent generic.

• What reaches the sequence is meaningfully smaller than what you searched for.

A list can lose a third or more of itself between search and send. That looks like waste and it is the opposite. The contacts removed are the ones most likely to bounce, land on the wrong person, or receive something generic enough to damage the sender reputation everyone else on the list depends on.

This is also why the 1,000 minimum exists. Losing that proportion is normal, so a search returning 600 leaves a campaign too small to learn anything from.

What we learned

Volume matters, but not the way people assume. Starting at 1,000 or more contacts is not about sending to all of them. It is about having enough good options left after cleanup and enrichment have removed the rest.

Cleanup is not optional, and most people rush it. Sending to 800 verified contacts beats 1,200 unverified ones on both deliverability and response rate.

Enrichment should match the message. Do not enrich because you can. Work out what the campaign actually needs and enrich for that, or you are generating expensive noise.

Specific personalisation reads differently. Using someone’s company name is not personalisation. Showing that you understand their situation is.

The bottom line

The breakthrough is personalisation that survives scale: genuinely relevant suggestions for hundreds of prospects, with no manual research per contact.

In an inbox already full of generic outreach, a systematic route to something actually relevant is close to the only way through.