M8TRX.AI lands 50 fully-researched leads in your approvals queue every hour, 24/7 — a ~50× efficiency increase over a human researcher.

Typical range: $17–$27 per 50 leads. Cache hits and dossier complexity move the actual number around within the band. Numbers come from a real sweep on this account.

01 They sound like a person wrote them

We banned the words that make an email feel auto-generated — things like “leverage”, “seamless”, “streamline”, “unlock”, “cutting-edge”. We also removed the long dash (—) that AI tools lean on, and we mix short, punchy sentences with longer ones the way people actually write.

02 A sharper first line

The old emails often opened with a setup sentence. Now the very first line says something specific about them — a number, a likely problem, a real observation — so the reader knows in five seconds why this email is worth their time. No “I’m reaching out”, no “I noticed”.

03 It sells the result, not the technology

Instead of listing what we do, the email names a problem the prospect probably hasn’t put a number on yet (“boards resist the special assessment”) and what changes for them. One idea per email, not everything at once.

04 A low-pressure ask

Every cold email ends with one simple, easy-to-answer question — like “Is this on your radar this quarter?” — never a request for a meeting. We don’t push for a calendar slot in a first email; that reads as presumptuous and hurts replies. If someone shows interest, the follow-up is where your booking link gets shared and the meeting gets set.

05 Replies and follow-ups got the same upgrade

When a prospect writes back, the follow-up uses the same human voice. And if they push back — “too expensive”, “not right now”, “we’re too small”, “just send info” — the reply now handles it like a good salesperson would: acknowledge it, reframe it, and offer a smaller next step instead of pushing.

A Colorado Realty — residential property manager, ~800 units

The after-version leads with their real number (800 properties), uses short punchy sentences, and ends with a low-pressure interest question — no meeting ask in the cold email.

B RealManage — HOA management, dozens of associations

This one teaches a consequence they feel (boards resisting assessments over inconsistent pricing), then ends with a simple interest question. If they bite, the follow-up is where your booking link gets shared.

• You still approve every email. Nothing leaves without your click.
• Still one email per prospect, with a clear opt-out (“reply ‘no thanks’”). No one gets trapped in a sequence.
• The agent never invents facts about a prospect. It only uses what the research actually found.

Our first pass made the emails read less like marketing. But line up a real batch and read them together, and they still felt same-y. Same shape every time: a list of services, a repetitive “one X, one Y, one Z” line, the same closing question. Each one was fine on its own. Together they read like a machine wrote one email fifteen times.

So we fixed it. Here’s what changed, in plain terms.

What changed

• Every email opens and closes its own way. Each prospect gets its own opening angle and its own closing question. No two read alike.
• We banned the tells. No service lists. No repetitive “one X, one Y” phrasing. No formulaic “is this something you’re working on?” close. No vague guesses about the prospect.
• A second editor checks every draft. Its only job: does this still sound like a template? If it does, it rewrites it, keeping every fact exactly as researched. Nothing invented.
• Research comes first. The agent finishes its deep research on the whole batch before it writes a word, so every email can open with something real and specific. (On the dashboard you’ll see drafts sit at zero while research fills up. That’s the rule working, not a stall.)
• Real name, real person. It addresses the actual decision-maker by name when the research finds one, and it never breaks character. No “cold email” tells, no hint of automation.

Before: a draft listed what we do and closed like every other email in the queue. Now: it opens with something true about that company, makes one point, and ends with a low-pressure question in its own words, sent to the right person by name. It reads like a sharp rep took five minutes to write you a note. That’s the bar we held it to, and we expect it to show up where it counts: more replies.

Same rails as always. You approve every email, one per prospect, with a clear opt-out, and nothing ever gets invented.

None of this is a setting someone has to switch on, or could get wrong on one account. It lives in the core of the product. So a new customer’s agent writes to the same standard from day one, with nothing to configure. First customer or thousandth, the quality is identical. And it covers the whole conversation now: the first cold email and the follow-up replies both. We built the quality in once, and it ships the same way every time.

Every arrow except Find → Send can also be triggered by an inbound — the agent picks up the conversation wherever it lands. The whole flow is gated by your approval; nothing leaves without it.

01 Fill in Settings

Click Settings (left menu). Fill in:

• Your name — appears in the warm signoff at the bottom of each email (e.g. Regards,
Emma). Use your first name for a casual signoff, or your full name for a more formal one — either works.
• Email signature — your full name, title, phone — appears below every send.
• Company website — mentioned in the body of every email so recipients can take a look.
• Booking link — where prospects book a call with you. Two ways:
   — Preferred — Google Calendar appointment schedule: open calendar.google.com → Create → Appointment schedule, define your availability, and copy the public booking page URL (looks like https://calendar.app.google/…). Prospects pick a time themselves; you get the confirmation in your inbox and the agent stamps the booking automatically.
   — Or a permanent Meet room: open meet.google.com → New meeting → Create a meeting for later → copy the link (https://meet.google.com/abc-defg-hij). The agent inserts the link in the follow-up and proposes two time slots; you mark the booking manually once a time is agreed.
• Mailing address — the law requires a real postal address on every marketing email; it’s added automatically.
• Email account — Gmail, Outlook, or another provider. This is the inbox the agent sends and receives from. After saving, click ⚡ Test connection below the From-address field to confirm the credentials authenticate (sends no mail).

02 Create your agent

Go to Outreach and click + Create your own agent. Type one paragraph about what you sell, and one about who you want to reach. The agent designs its own plan in a minute or two and starts working straight away.

You can create more than one agent — for example, one for each product or audience.

01 Open Approvals

New drafts appear here as the agent finds and researches each prospect. The page refreshes itself every few seconds — you don’t need to reload.

02 Review each draft

Each draft shows you:

• Two subject lines — click one, or type your own.
• The full email body — edit anything that doesn’t feel right.
• 🔎 View research — a small button on each card pops a modal with the prospect’s structured research pack (summary, services, recent news, notable clients, pitch angle, plus deep-pass press mentions, open roles, leadership if the deep research ran). Read this before approving when you want to sanity-check the personalization.
• Appended on send — the signature and legal footer that get added after the body. This is what the recipient will see at the end of the email. The same block is previewed once at the top of the Approvals list so you know exactly what gets appended.

A "Show N sent" bar at the bottom of the list reveals an audit trail of every approved email — click a tile to expand the full sent body.

03 Approve or decline

Click Approve & Send to send it — or Decline to skip. Either way, the agent moves on. Declined contacts are never approached again.

04 Check Replies

When someone replies, the agent reads it, sorts it (interested / question / not interested / unsubscribe), and drafts your follow-up. Open Replies, read it, and approve when you’re ready. The nav badge shows how many follow-ups are waiting.

• If the prospect named a time (and your Booking link is set in Settings), the draft is a booking confirmation — the proposed time is shown at the top of the card, your link is inline in the body, and the button reads “✓ Send & mark Booked”. Approving sends the email AND advances the prospect to Booked in one click.
• If they were positive but didn’t name a time, you get the normal follow-up draft — the agent proposes a couple of slots and asks them to pick one.
• If the conversation needs closing out manually (deal won outside this flow, lead went cold), each reply card has Mark as Booked / Mark as Lost buttons. Marking as Booked prompts for the meeting time so it’s stamped on the prospect.

• Closed conversations (Booked / Lost) collapse to a one-line tile with a status badge. Click to expand the full chain; ↑ Minimize collapses back.
• ✕ Archive on a closed tile hides it from the default queue. Hidden ones live behind a "Show N archived" bar at the bottom; ↺ Unarchive brings them back. Archive is a Replies-view filter only — the prospect’s funnel state is unchanged.

01 Weekly calendar

Every prospect you marked Booked appears here as a small tile in the day-column for its meeting. ← Prev / Today / Next → arrows step the visible week. Today’s column is highlighted. Bookings without a parseable meeting time fall into an "Unscheduled" strip at the bottom so they’re still visible.

Meeting times are LLM-parsed from the prospect’s reply. If they wrote "Friday at 10 AM" with no timezone, the parser uses the prospect’s location to infer one (Austin → CDT, London → BST, etc.). The displayed time is in YOUR browser’s local timezone, so you always see when YOU need to be on the call.

02 Click a tile to open the meeting

The booking-detail modal shows the full conversation in chronological order — the original outreach you sent, the prospect’s reply, and the booking confirmation that went back. A big ↗ Join Meet button is at the top; a secondary 🔎 Research button opens the prospect’s research pack on top of the meeting view.

03 After the meeting: Yes / Follow-up / No

Once a meeting’s scheduled time has passed, the tile shows three buttons:

• ✓ Yes — converted — meeting won, prospect becomes a customer. Ticks the Conversions KPI on the gallery hero. The tile shows a green CONVERTED badge.
• ⏱ Follow-up — decision deferred. Tile shows a gold FOLLOW-UP PENDING badge until you decide.
• ✕ No — meeting didn’t convert. Tile hides from the default Bookings view (record kept — reversible via the small Change link on any decided tile).

01 Two research passes per prospect

• A basic pass reads the prospect's website (homepage + contact page) and works out what they do, their notable clients, recent news, and a one-line angle for the email.
• A deep pass uses live web search to find named, recent signals beyond the website itself.

02 What the deep pass surfaces

• Press mentions — articles, awards, funding announcements (with the article URL).
• Open job posts — pulled from careers pages and ATS boards (Greenhouse, Lever, Ashby, Workable) and from LinkedIn / Indeed.
• Named leadership — recent hires, role changes, named partners.
• Tech / tooling hints — what stack they run (Shopify, Square, BentoBox, Salesforce, etc.) where it's publicly visible.
• Pain signals — Glassdoor reviews, public complaints, operational frictions mentioned in press.

Every entry carries the source URL the agent actually fetched — entries without a source are dropped before reaching the composer.

03 Guardrails that protect your reputation

• No inventing facts. The composer is told to only use names, dates, headlines, and clients that appear in the dossier. A thin dossier produces a shorter email, never a padded one with fabricated specifics.
• Dead-lead detection. Closures, bankruptcies, "ceased operations" — caught and quietly skipped, never drafted.
• Do-not-contact list. Anyone you decline, anyone who unsubscribes, and any business flagged as inactive is added to a permanent block-list. Survives even if you delete the prospect.
• Placeholder filter. Contact-form copy like youremail@contact.com or john.doe@example.com is filtered out — only real mailboxes ever get used.

04 Apollo — finding the right inbox

Research tells us what the prospect does; Apollo tells us who to email. Before drafting, the agent calls Apollo with the prospect's domain to look up verified decision-maker emails (founders, partners, heads of growth / sales / BD).

• Primary strategy: domain + title search — pulls up to a handful of senior people at that company, then asks Apollo to reveal their verified work emails.
• Fallback: if domain search returns nothing, the agent uses the named leader from deep research (e.g. "Jane Doe, Managing Partner") and asks Apollo to match that exact name at the domain.
• Cost model: Apollo charges per verified-email reveal, not per search. A miss is free; a hit costs ~1 credit per revealed inbox (your Apollo plan, billed by Apollo directly — M8TRX.AI never marks it up). Typical: 3–6 credits per successful prospect.
• If Apollo misses (no listing, no verified email): the agent falls back to whatever public info@ / contact@ address the website exposes — lower reply rate, but never a fabricated address.
• You choose the inbox. The Approvals card shows every candidate Apollo returned (name + title + verified email) so you can pick the best person before sending — or override with one you know yourself.

Optional: set Apollo API key in Settings to enable. Without a key the agent still works — it just skips Apollo and uses website emails only.

01 Fit score per prospect

Once a prospect has been researched, the agent rates it against your ICP (the targeting you defined when creating the agent) and writes a Fit N/10 chip onto the card. The score uses the basic dossier plus the deep-research signals (press, hiring, leadership, pain signals) so it reflects the strongest evidence we have.

• Green 8–10 — strong fit; matches your target description and multiple buying signals. Work these first.
• Amber 5–7 — marginal; fits the description but no specific buying signals visible, or vice versa.
• Red 1–4 — weak or auto-disqualified (competitive peer of yours, dead lead, research failure).

02 Where you see it

• Prospect grid (agent detail page) — the chip sits next to the funnel status (New / Emailed / Booked / etc.). The grid is sorted highest-fit first, then newest as a tiebreaker.
• Approvals — same chip on each draft card so you can prioritize high-fit drafts when reviewing.

Hover the chip for the why: the model’s one-line reason, the ICP signals the prospect Matched, and the ones it Missed. Useful when an unexpectedly low (or high) score lands and you want to know what the agent saw.

03 Auto-disqualifiers (no token cost)

Before the model is asked to score, the agent short-circuits to 1/10 for prospects already binned by an earlier phase — no LLM call burned on something we’ve decided about.

• Dead lead — closure / bankruptcy signals in the dossier.
• Competitive peer — sells/builds/integrates the same kind of product or service as your offering.
• Research failed — not enough dossier signal to score against.
• Research rejected — ICP filter (e.g. on an exclusion list).

04 Editing targeting invalidates old scores

If you change ⚙ Settings → 02 Targeting (the ICP, business types, exclusions, or offering description), the agent treats existing scores as stale and re-rates affected prospects on the next tick. Chips update without re-running the basic or deep research — only the scoring call is repeated.

01 Star anything you want to revisit

Every approval card and prospect card has a ☆ in its top corner. Click it to save the prospect to your Favorites drawer. Click again to remove. The star fills in (★) when saved.

Favorites are per-prospect, not per-draft — the star survives if the draft is approved, declined, or rewritten.

02 Open the drawer from the top bar

The favorites pill in the top bar shows your saved count. Click it to slide out the drawer. Each entry shows the prospect’s name, its agent, the subject line, and a one-line body preview — enough to remember what made you save it without opening anything.

03 Jump back to the approval

Each row has → Open in Approvals — closes the drawer, navigates to the Approvals view, scrolls to the matching card, and flashes it briefly so it’s easy to spot.

If the draft has already been sent, the Sent section auto-expands and the jump lands on its compact audit tile. If the approval is on a different agent than you’re currently viewing, the per-agent filter is cleared so the card is no longer hidden. If the draft was declined there’s nothing to jump to — you’ll get a toast explaining why; the favorite still tracks the prospect.

04 Remove a favorite

✕ Remove on the drawer row unstars the prospect. The matching card’s star also flips back to ☆ on the next render — no manual sync.

From the agent detail page

Click on an agent in the Outreach page to open it. At the top:

• ⚙ Settings — rename, edit the subtitle/market, edit the targeting (see below).
• ❚❚ Pause / ▶ Resume — pause stops the agent working until resumed. Existing drafts stay in Approvals.
• ☞ Run tick now — force a tick immediately instead of waiting for the scheduled interval.
• + Add a target company — a small inline form at the top of the detail page. Type a name + URL, click Add, and the agent immediately researches that company and drafts an outreach for it (lands in Approvals in ~1–2 min).

Editing targeting

In ⚙ Settings → 02 Targeting, three editable chip lists:

• Regions in scope — e.g. USA, UK. Click a chip to remove. Type a new region + Enter to add.
• Business types — the types the sourcer prospects from. Same add/remove pattern.
• Business types to AVOID — hard exclusions the sourcer will never propose (e.g. Cryptocurrency).

Changes take effect on the next tick — the sourcer reads these directly when prompting for new prospects.

Every agent runs the Cold Outreach workflow — the only workflow this product offers.

Deleting

The trash icon on the agent card soft-deletes the agent + its prospects. Recoverable for 48 hours via the Restore action, then permanently purged.

• Sends an email without your approval.
• Contacts someone who has unsubscribed or asked not to be contacted.
• Approaches the same business twice (declined contacts stay declined).
• Invents facts about a prospect — if the research is thin, the email is shorter rather than padded with guesses.

Outreach — your agents live here. The hero KPIs show Prospects worked (total ever seen), Verified & emailed (cumulative count of prospects we’ve emailed), Leads booked, Conversions, Agents live.

Approvals — every drafted email waiting for your decision, grouped by agent tabs at the top. Each card has a 🔎 Research button. A "Show N sent" bar at the bottom reveals the audit trail of approved emails.

Replies — inbound replies + the agent’s drafted follow-ups. Nav badge counts un-actioned ones. Booked / Lost replies render as compact one-line tiles (click to expand, ✕ Archive to hide).

Bookings — weekly calendar of confirmed meetings. Click a tile to open the conversation history + Join Meet button. Past meetings get Yes / Follow-up / No decision buttons.

Settings — your name, email signature, company website, booking link (Google Calendar appointment schedule or Meet room), mailing address (CAN-SPAM), and email account credentials.

Guide — this page.

The one rule

Every email stops for your approval — first contact, follow-up, anything. You are always the last step before it reaches a real person.

Typical range: $17–$27 per 50 leads. Cache hits and dossier complexity move the actual number around within the band. Numbers come from a real sweep on this account.

vs. doing this in-house

Same work, but with a US researcher at $40 / hour fully loaded (~$50K base + benefits, taxes, overhead). M8TRX.AI side includes the $200 / month subscription plus the per-50 AI usage from the table above. The advantage grows as your volume goes up — the subscription is fixed.

M8TRX.AI lands 50 fully-researched leads in your approvals queue every hour, 24/7 — a ~50× efficiency increase over a human researcher.

How to spend less

• Pick Medium research instead of Deep when creating an agent — skips the deep-search pass, saves ~$6 per 50 leads.
• Competitors, dead leads, and obvious non-fits are auto-skipped by the Fit score at no cost — on by default.

01 Pause means $0

A paused agent does absolutely nothing — no research, no drafting, no inbox checks, no AI calls of any kind. Pause any agent and its spend stops the instant the next tick would have fired. It’s your hard-stop, anytime.

02 Agents never source more on their own

An agent sources up to 50 leads the first time it runs — then it locks. It will not go looking for new prospects again unless you ask it to. When you do want more, the Source more leads button on the agent page lets you pick a batch of 10, 20, or 50 and shows you the approximate cost of each first — you confirm before a single token is spent.

03 Today’s spend is always on screen

The small pill in the top-right corner shows $X.XX today · cap $30 and refreshes every minute. It changes color as you approach the cap — neutral well below, amber around halfway, red as you near it, gold once the cap is hit. You never have to wonder what’s being spent today.

04 A hard daily ceiling of $30

If total Anthropic AI usage reaches $30 in a single day across all your agents, every running agent auto-pauses and shows a banner explaining why. This is the safety net for anything we haven’t anticipated — even an unexpected bug couldn’t push your bill past it. Resume manually whenever you’re ready; the counter resets at UTC midnight.

05 Stuck prospects don’t loop

If the agent can’t research or score a particular prospect (broken website, unusual page structure, etc.), it tries 3 times, then gives up on that one and moves on. No infinite retries, no quiet token bleed on lost causes.

06 A quiet inbox costs $0

The reply checker only spends tokens when there’s actually a new inbound message to read. An empty or quiet inbox costs nothing. Once your initial 50 leads are filled, an idle agent sitting in your account for a month costs essentially $0.