Research-backed outreach with Claude: the method behind 160% of quota and 40 demos a month, run from one chat

📊What 40 demos a month looked like

As an SDR I sold enterprise investor-management software to private real estate sponsors. A generic email dies on contact in that market, so I never sent one. I ran at 160% of quota, month after month, which worked out to about 40 booked demos a month. The demos did not come from one channel. They came from the same research, used three ways.

The email half of this has numbers you can check. Eight campaigns on my own accounts, 950 leads, 34 replies: 3.6% of leads replied, and the best campaign got 8.6%.[14] They are counted per lead and set against the public benchmarks on the playbook page, so I will not repeat them here. What that page does not show is the procedure. This one is the procedure.

🧭What counts as a real reason to email

Intent-data vendors sell a score. A score tells you a company visited a category page. It does not tell you what to write. My definition is narrower and more useful: a real reason to email is a verifiable event, in the last 90 days, that changes what the reader's seat is worried about.

• Verifiable means you can point at where you saw it.
• Recent means the writing rules further down will not let you call anything older than 90 days recent, and neither should you.
• Seat means it has to land on the person you are writing to, not on their company in general. A new CFO worries about a different break than a new VP Sales, and the same funding round is a different email to each.

Six signals do most of the work. Here they are, with the play for each:

The ids in the second column are not decoration. They are the source ids SimpleSend's dossier hands Claude, and the ids Claude has to cite for every specific it writes. Step 4 shows what happens when it cannot.

⚙️Who does what: Claude, SimpleSend, you

This runs over MCP, the open standard Claude uses to call tools.[10] SimpleSend is a connector: Claude calls it the way it would call a calendar or a CRM, from claude.ai, Claude Desktop or Claude Code.[7] No tab-hopping. No CSV round trips until the export you actually want.

🔌Setup, once, about five minutes

You need a SimpleSend account (the 14-day trial works: no card, 375 units) and, for claude.ai or Claude Desktop, a Claude Pro, Max, Team or Enterprise plan.[8] Claude Code works with any plan that runs Claude Code, using a personal key from Profile → Claude → ⋯ → Get a key.[9]

https://simplesend.tech/api/mcp

Connecting sends you to SimpleSend to approve. The screen says what a connection can do, in these words:

• Read your lead lists, campaigns and drafts
• Create lists and campaigns, and research companies (research uses your SimpleSend units)
• Save the drafts Claude writes, and export campaigns

It cannot see your billing details, change your plan, or read your HubSpot or LinkedIn credentials.

The first thing to type: "what lead lists do I have in SimpleSend?". If Claude lists your lead lists, you are connected. The Profile card shows the connection and when it was last used, and Disconnect lives there too.

🎯Step 1: pick the hundred

Move 1 on the playbook page is the whole of Step 1: one persona, one nameable operational break, a hundred people or fewer, and one sentence that is true of everyone on the list and false of everyone off it. Write that sentence first. It becomes the bridge in email one and the filter for everything else.

Four ways the list gets to SimpleSend, cheapest first:

1. Paste it into Claude. Names, titles, companies, emails, LinkedIn URLs. Claude creates the list (up to 500 contacts) and it costs nothing, because the data is already yours.
2. Upload a CSV in the web app. Any extra column you already have (a note, a trigger, a mutual contact) becomes a fact Claude can cite, at 1 unit per contact for a notes column and nothing for a plain value.
3. Import a HubSpot list. Read-only OAuth from Profile; the list lands in the wizard deduplicated across your team. How the import works.
4. Search for them. From Claude, by title, seniority, function, industry, location or company, with the Lead Enrichment add-on (from $55/mo): 0.25 credit per new person shown, billed once ever, then 1 credit per work email and 10 per phone number actually found. Claude shows the estimate and asks before anything is billed. Find Leads.

One thing the list tells you before you draft: how many contacts have an email. Below about 80%, enrich or cut. And if enrichment found someone's email at a different company, the campaign leaves them out by default and says so. Resolve the flag or include them on purpose.

🔎Step 2: name the signals

A research target is a signal written as an instruction. It has two parts: a topic, what to find, and an angle, how a finding should change the email. The angle is where the judgment lives. "Company growth signals, use it to personalize" is not a target. "Open SDR or AE roles posted in the last 60 days; if they are hiring reps, name the role and lead with ramp time; if the postings are engineering-only, skip this" is.

SimpleSend runs up to three search strategies per target and keeps the first finding above 70% confidence. The drafter uses at most two findings per email. Findings are scarce on purpose. Two or three well-chosen targets beat five vague ones, and they cost less: planning is 5 units per target once per campaign, research is 5 units per target per company, cached for 30 days and shared by everyone at the same company. A target a web search cannot answer is neither searched nor billed.

SIGNALS AS RESEARCH TARGETS (topic, then the angle). Pick two or three.

1. The incumbent tool
   topic: which [CATEGORY] vendor or tool they use, from their website,
          careers page or job posts
   angle: if a vendor is named, name it and ask about the break it causes
          at their stage; if none is found, skip this finding entirely.

2. A new leader in the buying seat
   topic: a new [TITLE, e.g. CFO / VP Sales / Head of RevOps] hired or
          promoted in the last 90 days
   angle: if found, open on the appointment and offer the comparison a new
          leader runs in their first quarter; if the hire is in another
          function, skip it.

3. A raise, a close, an expansion
   topic: funding, a fund close, an acquisition or a new office announced
          in the last 90 days
   angle: if found, name the event and the operational break that arrives
          with that scale, and when it usually lands; older than 90 days,
          past tense and no "recently".

4. Open roles that imply the break
   topic: open [ROLE] roles posted in the last 60 days
   angle: if they are hiring [ROLE], reference the role and lead with
          [RAMP TIME / PRODUCTIVITY / COVERAGE]; if the postings are
          [OTHER FUNCTION] only, skip this finding.

5. Their own words
   topic: a LinkedIn post or interview by [FIRST NAME] in the last 30 days
          about [PAIN]
   angle: if found, reference the idea (not the post) and give the one
          thing that helps; never quote them back to themselves.

🧾Step 3: write the brief once

The brief is the same shape as the playbook's, mapped onto the fields the connector takes. What I sell, in my words. Who I sell to. Value propositions one per line, because the drafter rotates them across steps and the first one runs in email one. The tone word, which two or three pasted emails I actually sent will override, and should. The ask as a short verb phrase with a link, because the sentence that carries it becomes the link. Length short, which is 60 to 110 words unless my samples say otherwise. My signature.

Then the sequence: four steps, twelve days, and email one's CTA set to none. That is how "no ask in email one" stops being advice and becomes a setting. Gong's 304,174 cold emails found an interest close, asking whether the problem is worth a conversation, booked meetings at twice the rate of a calendar ask.[4]

THE BRIEF FOR CLAUDE + SIMPLESEND (one per segment)

description   What I sell, one concrete paragraph, in my own words. Name
              the operational break it removes and for whom. No adjectives
              that could describe a competitor.
icp           The seat, the company type, the stage. One line.
value_props   Three lines, one outcome each, specific enough to argue in
              a sentence. The first one runs in email one. Any number here
              must be one I can defend; it is the only number Claude may
              attribute to me.
tone          One word: casual / direct / warm. The samples override it.
cta           A short verb phrase that reads as a link:
              "grab the checklist", "see a sample", "book 15 minutes".
cta_link      Where that phrase points. Ungated if the phrase promises a gift.
length        short (60 to 110 words). Samples set it instead when pasted.
samples       Two or three emails I actually sent. Rhythm, vocabulary,
              contractions and directness are matched; sentences are not
              copied.
signature     Name / title / company, substituted for [Your name] at export.

sequence      intro (day 0, CTA none), value (+3 days), followup
              (+4 days), breakup (+5 days, no CTA)
research      Two or three targets from the signal set.
save_as_template: true

Save it as a template the first time (one flag on the campaign) and the next campaign is one line: "use my saved brief".

✍️Step 4: run the loop

Everything above is preparation. This is the part that used to take 58 hours. You paste one prompt, and Claude runs a loop: read the budget, read the list, create the campaign, then pull 3 to 5 contacts at a time, write every step of every sequence from that contact's dossier, save up to 10 at once, fix what the verifier flags, and go again until nothing is left. Then it shows you the two sequences it is least sure about and offers the export.

Draft a research-backed cold email campaign in SimpleSend for my lead list
"[LIST NAME]". Work through this in order and stop where I say stop.

Before anything else
1. get_account. Tell me my plan, units left and lead credits. Research costs
   5 units per target per company (cached 30 days) plus 5 per target to plan
   the campaign. If the list below will not fit, say so and stop.
2. list_lead_lists, then get_lead_list for "[LIST NAME]". Tell me how many
   contacts have an email and how many are flagged as a possible job change.
   If fewer than 80% have an email, stop and tell me.

The segment (true of everyone on this list, false of everyone off it)
[ONE SENTENCE]

Research targets (topic, then the angle as an if/then play)
1. [topic]. If found: [angle]. If not: [skip / fallback].
2. [topic]. If found: [angle]. If not: [skip / fallback].
3. [topic]. If found: [angle]. If not: [skip / fallback].

The brief
- description: [what I sell, one concrete paragraph, in my words]
- icp: [who I sell to]
- value_props (one per line):
  [value prop 1]
  [value prop 2]
  [value prop 3]
- tone: [casual / direct / warm]
- cta: "[verb phrase, e.g. grab the checklist]"   cta_link: [https://...]
- length: short
- samples: [paste 2 or 3 emails I actually sent]
- signature:
  [name]
  [title, company]

The sequence (four steps, twelve days)
- step 1, intro, day 0, cta mode "none": the signal, the bridge, one thing
  that is only useful to them, an interest close. No ask.
- step 2, value, 3 days later, inherit the CTA: a second signal, one insight.
- step 3, followup, 4 days later, inherit the CTA: one customer like them,
  one number from my brief, then a soft ask.
- step 4, breakup, 5 days later, no CTA: one line, door open.
Save the brief and sequence as a template named "[TEMPLATE NAME]".

Then
3. create_campaign named "[CAMPAIGN NAME]" on that list with the brief, the
   sequence and the targets. If research_planning is "incomplete", call
   plan_research. Show me the planning units, the per-company research cost
   and my remaining units, then continue without asking.
4. The loop: get_drafting_context five contacts at a time (pass
   include_rules: false after the first call). For each contact write every
   step as JSON that follows writing_rules and output_schema exactly. Every
   sentence that states a specific about the recipient goes in claims with
   its source id; if you cannot cite it, say the general thing or cut it.
   Open on the signal, never on what their company sells. save_drafts, fix
   every step that comes back with unresolved sentences and resave it, and
   repeat until remaining_contacts is 0. Tell me when a contact is deferred
   and call again for it.
5. When it is done, show me the two sequences you are least sure about, in
   full, and say why. Then stop and wait for me before exporting.

The rules Claude is handed

The connector gives Claude a writing rulebook with every batch. These are the ones that are the method:

• The opener is one sentence, at most 22 words, naming one concrete thing about them. "Saw the Series A news, congrats" is an opener. "You're selling robots to 3PLs" is not. An opener that could go to a different company unchanged is a failure.
• Never tell the recipient what their company sells or who it sells to.
• Cite or hedge. A sentence with a specific in it (an event, a number, a date, a vendor, a hire) has to carry the source id it came from. No source, no specific.
• Nothing older than 90 days is "just" or "recently".
• No dated fact and no research answer: open on the person's role with a pointed question, and keep the whole email under 70 words.
• Each step opens on a different fact and argues a different value prop. Follow-ups never "circle back". The breakup has no pitch and no CTA.

What the verifier does when a sentence breaks them

Three reasons come back: bad_source_id, claim_without_source and uncited_specific. A resave replaces the step. You never argue with it. You cite, hedge or cut.

Speed and limits: paid plans get 180 tool calls per five minutes, 3,000 a day, and 200 uncached companies a day, so a hundred-contact campaign fits in one sitting. Your part is reading, not typing.

📤Step 5: review, export, send

Read two full sequences yourself, including one you would have written differently. Have Claude run the QA pass below. Then export. SimpleSend never sends email, on purpose: your domain reputation stays in your hands, and the sequencer you already run does the sending.

Review the drafts in campaign "[CAMPAIGN NAME]" against my rules before I
export. Use list_campaign_contacts, then get_drafts. For every contact and
step, check:
1. Step 1 makes no ask. "Worth a chat?" counts as an ask. Flag it.
2. Every step is under 120 words and argues one idea.
3. The opener names one verifiable, dated thing about them, not what their
   company sells and not a compliment.
4. Each follow-up carries a new signal; nothing "circles back" or "bumps".
5. Every specific claim has a source; list them. Anything the dossier
   cannot back gets hedged or cut.
Give me one table: contact, step, problem, proposed fix. Then apply the fixes
with save_drafts (a resave replaces the step) and tell me exactly what
changed. Stop before exporting.

Two export shapes. CSV: one row per contact with the drafting status and, per step, subject, body and the QA verdict, so the columns map straight onto Smartlead, Instantly or HubSpot sequence fields. JSON: the same rows paged by contact with a hand-off block that names the per-step variables and asks Claude to confirm the mapping with you before anyone is enrolled. A step without a saved draft exports empty, so a half-drafted contact cannot slip through as a blank email.

export_campaign "[CAMPAIGN NAME]" as json. Walk me through the handoff
block: which variable carries each step's subject and body, and the
confirmation rule. Confirm the field mapping with me before I enroll anyone.
Then list every contact whose status is not "done" so I can exclude them,
and give me the CSV version too, with links written as plain text.

How SimpleSend works, from CSV to CRM

The infrastructure floor has not moved: SPF, DKIM and DMARC on every sending domain, warmed secondary domains, twenty to forty sends per mailbox per day, complaint rate under 0.10%.[5][6] It is Move 5 on the playbook page, with the pre-send checklist, and the sending infrastructure guide walks through the setup.

☎️Step 6: open the other two doors

Email caps out. On the playbook page the best campaign got 8.6% of leads to reply, and the average across eight was 3.6%. The demos number at the top of this page came from the same signals used three ways: as email one, as the call opener, and as the LinkedIn note. Same research, three doors.

The call block

Talk tracks are the same dossier turned into a card a rep glances at before dialing: why call now (the signal, dated), the hooks, a permission-based opener, discovery questions ordered easiest to deepest, talking points, the objections you actually hear with a spoken response for each, and the ask. They are drafted in the web app on the Talk Tracks channel (no draft units; research plus one card per contact), and the Chrome extension shows the card inside HubSpot and Salesforce Lightning the moment a record opens, with Nooks, Orum, Kixie and PhoneBurner as opt-in hosts. From Claude, before a call block, ask for the cards.

Before my call block: list_campaigns with channel "talktrack" and find the
one for "[LIST NAME]". For the next 20 contacts, get_drafts and give me one
card each, four lines: why call now (the signal, with its date), the opener,
the first discovery question, and the objection I am most likely to hear
with the response. If there is no talk-track campaign for this list, say so;
I will draft one in the web app.

The LinkedIn layer

From Claude, two ways. Templated: a connection note under 300 characters with no link and no pitch, follow-ups after acceptance with merge tags, paced by SimpleSend (25 invites a day by default, 80 at most; quiet hours 22:00 to 07:00; three days between follow-ups; acceptance polled for 14 days). Drafted per person: SimpleSend writes the note and the messages from the same research (draft units per step plus 2 units of person research per contact), and you send when the previews look right. Either way Claude shows the preview and the pacing, nothing goes out until you say go, and sends spend your We-Link credits, not units. It needs your own LinkedIn account connected in the web app and the LinkedIn Outreach add-on ($50/mo) or the Outbound Bundle ($300/mo). How LinkedIn outreach works.

Set up the LinkedIn layer for the same list. First get_linkedin_account: if
my seat is not connected or my plan does not include LinkedIn, tell me what
to do in the web app and stop.

Otherwise create_campaign with channel "linkedin" on "[LIST NAME]" using my
saved brief "[TEMPLATE NAME]" as template_id, with a two-step sequence: a
connect note (under 300 characters, no link, no pitch, the same signal as
email one where the research supports it) and one follow-up 3 days after
acceptance with the interest close. Tell me the units it will cost per
contact, then start_drafting. When get_campaign reports it is done, show me
the notes for the first five people and the pacing you propose (25 invites
a day, quiet hours 22:00 to 07:00). Do not send anything until I say go.

🧮What it costs

Two calculators, because the two questions I get are "what does a campaign cost me" and "how many people does forty demos take".

Defaults: 3.6% replied per lead is the playbook's eight-campaign figure;[14] 15.3% and 29.6% of replies booked are Gong and 30MPC's comparison group and top decile across 85 million cold emails.[3] Instantly's average reply rate is 3.43%, for scale.[1]

📋Before you hit send

The prompts above only work as written with SimpleSend connected, because they name its tools. The segment test, the signal set, the brief and the QA rubric work in any tool; they just take 58 hours to fill by hand. This one is the last thing I run every time.

BEFORE YOU HIT SEND: A CLAUDE-DRAFTED CAMPAIGN

Connection
[ ] SimpleSend shows as connected in Claude (connector settings, or /mcp in
    Claude Code)
[ ] get_account: enough units for the list (5 per target per company, plus
    5 per target to plan)

List
[ ] One persona, one break, 100 people or fewer; the segment sentence is
    written
[ ] 80% or more have an email; the rest enriched or cut
[ ] Job-change flags resolved, or included on purpose

Signals
[ ] Two or three research targets, each a topic and an if/then angle
[ ] Nothing older than 90 days will be called recent (the rules enforce it;
    the angles assume it)

Brief
[ ] Two or three real sent emails pasted as samples; signature set
[ ] The CTA is a verb phrase with a link; step 1's CTA is "none"; step 4 is
    a breakup with no CTA
[ ] Any number in the brief is one you can defend

Drafts
[ ] remaining_contacts is 0 and every unresolved list is empty
[ ] You read two full sequences yourself, including one you would have
    written differently
[ ] Export shape matches your sequencer; the mapping was confirmed before
    anyone was enrolled

Sending (the playbook page, Move 5)
[ ] SPF, DKIM and DMARC on every sending domain; warmed secondary domains;
    20 to 40 sends per mailbox per day; complaint rate under 0.10%

🧯When it doesn't work

Run this on your own list

Everything above is the method. The fastest way to judge it is to point it at a hundred people you already know you want, and read what comes back.

1. [1]Instantly, Cold Email Benchmark Report 2026 (a year of sends across 700k+ businesses)
2. [2]Smartlead, The State of Cold Email 2026 (850M+ sends)
3. [3]Gong Labs × 30MPC, The Ultimate Cold Email Data Report (85M+ cold emails)
4. [4]Gong Labs, The surprising cold email CTA that books more meetings (304,174 emails)
5. [5]Google, Email sender guidelines (bulk sender requirements)
6. [6]Yahoo, Sender best practices (bulk sender requirements)
7. [7]Anthropic, Third-party connectors with remote MCP (vendor documentation)
8. [8]Anthropic, About integrations using remote MCP (vendor documentation)
9. [9]Claude Code, Connect Claude Code to tools via MCP (vendor documentation)
10. [10]Model Context Protocol, the specification
11. [11]SimpleSend, Use SimpleSend from Claude (our own documentation)
12. [12]SimpleSend, Security and administration for the Claude connector (our own documentation)
13. [13]SimpleSend, Privacy policy: AI assistants and the Model Context Protocol (our own documentation)
14. [14]SimpleSend, What still gets replies in 2026: 8 campaigns, 950 leads, 5 moves (our own data)

The quota and demo numbers are one person's months in one niche, and the campaign figures are my own accounts over my own lists. Benchmark figures move quarter to quarter and depend on what each dataset divides by. Treat all of it as direction, not promises.