Evidence before claims

Outbound that does the research first.

Megdot researches the account, finds the decision-maker, verifies the address and writes from evidence — then grades its own draft before it reaches your queue.

Connects to what you already run

GmailGoogle WorkspaceFirecrawlParallelBouncerCloudflare Workers AIAnthropicScrapeGraphCSV importIMAP repliesGmailGoogle WorkspaceFirecrawlParallelBouncerCloudflare Workers AIAnthropicScrapeGraphCSV importIMAP replies
The difference

One of these gets deleted.
You already know which.

Volume tools already exist, and the inbox has priced them in. Megdot is built on the opposite bet — that a message is only worth sending if something specific and checkable is true about the person receiving it.

Sequence templateStep 1 of 7
Quick question about {{company}}

Hi {{first_name}},

I came across {{company}} and was really impressed by your growth in the {{industry}} space.

We help {{industry}} leaders cut costs by up to 40%. Would you be open to 15 minutes on Thursday?

  • "impressive growth"Unearned. Nothing was checked.
  • "{{industry}} leaders"The merge field is the personalization.
  • "15 minutes Thursday?"An ask with no reason attached.
Nothing in it is false. Nothing in it is checked, either.
MegdotJudge 87 / 100
Second hub, same dock team?

Priya —

Two hubs in a quarter is a lot of new dock doors. Most ops teams we work with hit a scheduling wall at the second site, not the first — the first one gets everyone’s attention.

If that’s already showing up in your load planning, I can send the two things that fixed it fastest for a fleet about your size. Worth a look?

  • Two hubs opened this quartercareers.northwindfreight.com · 2d ago
  • Scheduling load lands on ops firstDerived · why-now, conf. 0.83
  • No claim without a sourceThe writer only ever sees evidence that carries one.
Every sentence traces back to something on the record.
FIG 0.1The same prospect, written two ways. Highlighted spans on the right are the phrases that trace directly to an evidence record.
The pipeline

Twelve stages, and you can open every one.

A prospect moves through the same chain whether Megdot is driving or you are. Each stage records what it did, what it used, what it cost and why — so a bad email is a question with an answer rather than a mystery.

Find

Turn a messy list into real people with real addresses.

Learn

Gather what is checkably true, and score what it means.

Write

Reason first, write second, then grade it independently.

Send

Deliver inside your limits, and learn from what comes back.

01
Imported
CSV, list, or ICP discovery
02
Resolved
Row → company, deduped
03
Person found
Decision-maker + rationale
04
Email found
Public, provider, inferred
05
Verified
Checked before anything sends
06
Researched
Evidence, source, timestamp
07
Scored
ICP, buyer fit, timing
08
Written
From a plan, not a template
09
Judged
A second model grades it
10
Reviewed
You, or your policy
11
Sent
Rotated, inside your window
12
Replied
Classified, fed back
FIG 0.2Stage names are the product’s own. Colour marks the four phases: find, learn, write, send.
Operating modes

One core. Three amounts of autonomy.

The same research, the same writer, the same judge. What changes is how much Megdot may do without asking — a per-mission setting you can move either way, at any time.

Manual

You drive

Run any stage on demand, for any prospect. The same intelligence core, exposed one step at a time — right for a short list you care about individually.

  • Per-prospect stage runs
  • Draft workbench
  • Nothing happens unprompted
Copilot
Most start here

It proposes, you decide

Everything is researched, written and judged in advance. What reaches you is a queue of decisions with the evidence attached — approve, edit, or reject with a reason the writer learns from.

  • Review queue with full context
  • Inline editing before send
  • Rejections become signal
Auto

It runs the desk

Megdot works the mission end to end inside the policy you set — score floors, quality floors, which email statuses may ever be sent to, how many a day, in which window.

  • Policy-bounded sending
  • Daily caps and send windows
  • High-value holds for approval
Uncertainty, on purpose

An address it guessed is never shown as one it checked.

Most tools return one word: found. That word hides an enormous amount, and the bounce rate it produces is charged to your domain, not theirs. Megdot keeps the six states apart the whole way through, and your policy decides which may ever be sent to.

VerifiedChecked by an independent verifier
Provider foundA data provider asserts it
InferredPattern-guessed from the domain
RiskyCatch-all, role account, or disposable
InvalidKnown bad — never sent to
UnknownNothing established yet
What the writer is handed
claimOpened two regional hubs this quarter
sourcecareers.northwindfreight.com/openings
seen2 days ago
qualityprimary
conf.0.91

No evidence, no email. A prospect Megdot could not learn anything checkable about is skipped and reported to you — not padded with a compliment about their “impressive growth”.

FIG 0.3One evidence record, verbatim. The writer never receives a free-text impression of a company — only records shaped like this.
The platform

Everything the desk actually needs.

Company resolution

Messy rows become real companies — domain, size, industry, stack — deduped so one account is never worked twice.

Deep research

Crawl and search the open web for hiring, funding, expansion, leadership and product signals, each filed with its source.

Verification, not guessing

Inferred addresses stay labelled inferred. Policy decides whether they may ever be sent to. The default is no.

Reply intelligence

Replies are ingested, classified and routed. Positive, referral, objection and unsubscribe are different events.

Deliverability guardrails

Per-mailbox limits, warmup state, rotation, and a suppression list checked before every single send.

Learning loop

Which subject style, which playbook, which angle. Copy performance is attributed back to the choices that made it.

Straight answers

The questions that actually matter.

Mostly about your domain, and whether the machine can be trusted with it.

That is the failure mode the product is built around. Addresses are verified before anything sends, and inferred ones are labelled inferred — your policy decides whether they may ever be used, and the default is no. Each mailbox has its own daily cap and warmup ramp, sending rotates across connected mailboxes, and the suppression list is checked before every individual send. You also set the window: which hours, which timezone, weekdays only or not.

The writer is never handed a free-text impression of a company. It receives evidence records, each carrying a claim, a source URL, a timestamp, a quality grade and a confidence — and it writes from those. A prospect Megdot could not learn anything checkable about is skipped and reported to you, rather than padded with a compliment about their impressive growth. A second, independent model then grades the draft before a human sees it.

A sequencer starts from the email and fills in blanks. Megdot starts from the account: resolve the company, find the decision-maker, verify the address, research what is actually going on, score whether it is worth writing to at all — and only then write. Most of the work happens before a draft exists, and a large part of the value is the prospects it tells you not to email.

No. Three modes over one intelligence core. Manual runs a stage at a time, on demand. Copilot researches, writes and judges everything in advance and hands you a review queue with the evidence attached. Auto works the mission inside the policy you set — score floors, quality floors, daily caps, which email statuses may ever be sent to. It is a per-mission setting and you can move it in either direction at any time.

Model spend is tracked per prospect, per mission and per day, and shown to you. Work is routed by task across cheap, mid and frontier models rather than sending everything to the most expensive one — scoring does not need the model that writes the email. Every record carries the cost that produced it, so the unit economics are visible rather than inferred at the end of the month.

In your own workspace. Signing up provisions a private workspace that only you belong to; workspace membership is re-checked on every single request, so nothing leaks between accounts. Connecting a Gmail mailbox is a separate consent from signing in, and it is the only thing that grants sending access.

Start with one list and one mission.

Import a CSV, describe who you sell to, and watch the first ten records get researched. Every claim it makes will have a link next to it.

Your own workspace in seconds · No card required