Skip to content
Research and enrichment

Read the company first, then earn the sentence.

The Research agent crawls the lead's own website and analyses what it finds. The Strategist turns that into an angle for this one prospect. Neither of them describes a company that was never read.

No payment method. No recurring charge. Your own SMTP, your own data.

Source
The lead's own website
Research model
Claude Haiku 4.5
Angle and analysis
Claude Sonnet 5
No crawl
No profile
Research - extracted profileCrawled
IndustryIndustrial equipment distribution
Size signalTeam page lists 20-50 people
Tech signalShopify Plus, HubSpot forms
Recent changeNew depot announced on the site
Strategist angle

Lead on the depot expansion, keep the ask to a 15-minute call, avoid discount framing.

Illustrative preview - research output for one lead
Why the first line falls flat

Personalisation is not a merge field with a town name in it.

Every prospect can tell the difference between a message written for them and a message written for a list they happen to be on. The tell is not the greeting. It is whether the sender knows what the business actually does.

  1. The same opener, sent to a thousand people

    A template with a company name slotted into it still reads as a template. The recipient has had four versions of it this month, and the pattern is the thing they recognise.

  2. Research that costs twenty minutes a lead

    Reading a site properly works. It also puts a hard ceiling on volume, and it is the first task dropped in the week the pipeline actually needs filling.

  3. A guess written down as a fact

    Assuming a stack, a size or a growth story, then building the whole message on it. When the assumption is wrong the email is not merely generic. It is wrong in a way the reader remembers.

  4. Lessons that never leave one campaign

    What worked last quarter lives in somebody's memory or in a spreadsheet nobody reopens. The next campaign starts from nothing for no good reason.

How a lead becomes a profile

Four steps between a domain and a sentence worth reading.

Research reads. The Strategist decides. The knowledge base remembers. Nothing is written until there is something real to write about.

  1. 01

    Crawl the site itself

    Research fetches the lead's own website with Cheerio. When a page renders in the browser rather than in the HTML, Playwright and Chromium render it first, and an optional Bright Data Web Unlocker fallback is available for sites that refuse a plain request.

  2. 02

    Analyse what was read

    Claude Haiku 4.5 turns the crawled pages into a company profile: industry, technology stack, pain signals and growth signals. The profile is built from what the site says, not from what a company of that type is expected to say.

  3. 03

    Fail closed on the gaps

    If the first-party crawl does not succeed, no company profile is generated. An empty result is more useful than a plausible one, because a plausible one gets quoted in an email as though it were checked.

  4. 04

    Choose the angle

    The Strategist, on Claude Sonnet 5, picks the approach, the tone and the type of call to action for this prospect, and queries the knowledge base for what has worked in earlier campaigns.

Inside a profile

Four things the crawl can defend, and nothing past them.

Industry, technology stack, pain signals and growth signals. That is the shape of the profile, and each part of it traces back to a page that was actually fetched from the lead's own domain.

  • Cheerio reads the HTML; Playwright and Chromium handle sites that render in the browser
  • An optional Bright Data Web Unlocker fallback covers sites that resist a plain fetch
  • Research runs on Claude Haiku 4.5; the Strategist and the Analyst run on Claude Sonnet 5
  • No successful first-party crawl means no profile at all
Research - extracted profileCrawled
IndustryIndustrial equipment distribution
Size signalTeam page lists 20-50 people
Tech signalShopify Plus, HubSpot forms
Recent changeNew depot announced on the site
Strategist angle

Lead on the depot expansion, keep the ask to a 15-minute call, avoid discount framing.

Illustrative preview - research output for one lead
In the operator's hands

Enrichment you can run on one lead, not only on a queue.

The pipeline enriches leads as they move through it, and the same research is available on demand from the interface when a single lead needs it now.

  • Research domain runs enrichment for one lead, on request, from the interface
  • Field suggestions, rewriting and an alternatives panel sit beside the text they change
  • A campaign wizard drafts the setup and you edit what it proposes
  • AI queries are limited per organisation and per user, so the cost stays bounded
Scout - discovery results5 of 5 shown
BusinessScoreStatus
Hydraulics supplierManchester84New
Industrial coatingsLeeds78New
CNC machining shopSheffield71Queued
Packaging converterBristol64Queued
Cold storage operatorHull52Held
Illustrative preview - lead pool after a discovery run
What changes

The difference is that the email starts from something that was read.

A profile with a source

Industry, stack, pain and growth signals come from the prospect's own pages. If those pages were never fetched, there is no profile to quote.

Volume without the twenty minutes

The reading still happens for every lead. It happens at the pace of a queue rather than the pace of somebody's afternoon.

An angle chosen per prospect

Approach, tone and call to action are decided for this company, instead of one campaign voice applied evenly across the whole list.

A knowledge base that grows

Your own documents and your own results feed the store the Strategist queries. It is fuller in month three than it was on day one.

Weekly analysis instead of anecdote

Once a week the Analyst reviews performance across campaigns and writes what it finds back into the knowledge base for the next campaign to use.

A gate the model cannot open

Research improves what gets proposed. It does not decide what leaves: a person still approves every draft before it is sent.

The awkward cases

What happens when the site does not cooperate

Enrichment is mostly a set of edge cases. These are the ones that come up in practice and how the pipeline behaves in each.

  • A rendered site

    Everything sits behind JavaScript

    A plain fetch returns an empty shell, so Playwright and Chromium render the page first and the crawl reads what the browser produced.

  • A resistant site

    The request is refused

    An optional Bright Data Web Unlocker fallback covers the sites a direct fetch cannot reach. It is there for the hard cases, not for every crawl.

  • No site

    There is nothing to read

    No first-party crawl means no profile. The lead stays in the pool without an invented description attached to its record.

  • A wrong read

    The model draws the wrong conclusion

    It can happen. The profile is visible before anything is written, the draft is visible before anything is sent, and a person approves both.

  • A first campaign

    The knowledge base is still empty

    The Strategist picks the angle from the profile alone. History is what the base adds later, once the Analyst has a week of results to write down.

  • Several clients

    One agency, separate knowledge

    Knowledge can be filtered by project, and every record is scoped to your organisation, so one client's material is not pulled into another organisation's work.

The specifics

What the enrichment layer actually is

The short version, without adjectives, for the person who wants to know what runs and on what.

Crawler
Cheerio over first-party HTML
Rendering fallback
Playwright and Chromium
Hard sites
Bright Data Web Unlocker, optional
Research model
Claude Haiku 4.5
Strategist and Analyst
Claude Sonnet 5
Profile rule
Fail-closed, no crawl no profile
Knowledge uploads
PDF, DOCX, images, txt
Embeddings
Voyage AI voyage-3-lite
Vector size
512 dimensions
Semantic search
JSONB and cosine_similarity SQL
Cross-campaign analysis
Weekly
AI query limits
Per organisation and per user

What the research part of your process stops needing to be.

How each task is handled today compared with MyLeadTeam
TodayIn MyLeadTeam
An assistant reading websites and pasting notes into a spreadsheet column.The crawl and the analysis run per lead, and the profile sits on the lead record.
A vendor's guess about a company, bought by the row and never checked.A profile built from the company's own website, or no profile at all.
One campaign voice applied evenly to every prospect on the list.An approach, a tone and a call to action chosen per prospect by the Strategist.
A folder of case studies and briefs nobody reopens before writing.PDF, DOCX, image and text uploads, chunked, embedded and searched semantically.
A retrospective that happens when somebody has time and decides nothing.A weekly cross-campaign analysis written back into the knowledge base the Strategist queries.
FAQ

Questions people ask about this

It means reading the prospect's own website before writing to them. The Research agent crawls the lead's site and analyses it with Claude Haiku 4.5, producing a profile of the industry, the technology stack, the pain signals and the growth signals it can find there. The Strategist then uses that profile to choose an approach, a tone and a type of call to action for that one company. The enrichment comes from the site itself rather than from a bought contact dossier.

Get started

Write to a company you have actually read.

Let the crawl do the reading, let the Strategist pick the angle, and keep the final word with the person who signs the message.

No payment method or recurring charge during early access.