Getting the Most Out of SimpleSend

Everything you need to write briefs that produce great cold emails — from first principles to the advanced levers power users pull. SimpleSend turns a CSV of leads into a CSV of personalized cold email sequences, one complete multi-step sequence per contact, written from scratch. It does not send anything. The output is a ready-to-import file you drop into SmartLead, Instantly, Apollo, HubSpot, or a Gmail mail merge.

What is SimpleSend?

The key difference from a template-and-token approach: every subject line and every email body is drafted uniquely per contact. There are no {{first_name}} slots filled in at the end — the AI drafts each email knowing who the recipient is, what their company does, and what recent signal (if any) it found about them.

Per-contact pipeline

For each contact in your CSV, SimpleSend runs the full sequence through this chain:

1. 0Fit check (pre-flight)If your brief has an ICP, one cheap Haiku call scores the CSV row against it first. A confident non-match marks the contact skipped — no research, no drafting, no unit spend. The check fails open on thin data, so borderline rows still get generated.
2. 1EnrichFetches and summarizes the company homepage. Parses the LinkedIn profile for LinkedIn DM campaigns. Results are cached 30 days.
3. 2ResearchRuns your lookFor entries — targeted searches that mine job postings, LinkedIn activity, news, and more for contact-specific signals.
4. 3Draft each stepThe model you selected writes each email. Each step gets a unique template, a unique value prop from your list, and a unique finding (if found).
5. 4Fact-check & lintAI verifies that no claims were invented. Deterministic rules check length, forbidden phrases, CTA link presence, and formatting.
6. 5Polish & exportIf any issues are found, a final editing pass fixes them without inventing new facts. The result is exported to CSV.

How it works — the 5-step flow

Once you upload your CSV, SimpleSend walks you through five screens:

The brief

The brief is the single most important thing you control. Every field maps directly into the AI's prompt — knowing what each field does lets you write briefs that actually move output quality.

Company description & ICP

Description — one paragraph. Goes into the research planner as industry context and into the drafting model as your company description. The planner uses it to decide where to look for contact signals; if your description mentions "SaaS revenue operations," the planner knows to check job boards for RevOps hires and LinkedIn for GTM content.

ICP — 2 lines max. Goes into the research planner as recipient context and into the drafting model to ground the "light research flavor" sentence when no specific finding is available. Include signals the planner can search on: stack names, company size/stage, hiring patterns, GTM motion.

Value props ⭐

The most leverage-heavy field in the entire brief. Up to 3 value props, one per line. SimpleSend assigns one prop per email step in round-robin order: Step 1 gets prop #1, Step 2 gets prop #2, Step 3 gets prop #3, Step 4 wraps back to #1. Each email is locked to its assigned prop and told not to reuse props from prior steps. Breakup steps skip prop assignment entirely.

If you give only one value prop for a four-step sequence, all four emails must pitch the same angle. The AI is told to "find a fresh angle," but with nothing new to work with, drafts become repetitive and thin. Three distinct, specific props give each email in the sequence a unique hook.

Writing high-leverage value props

Think in terms of specific outcomes, not category features. The drafter is given one prop and told to build the entire email around it — so the prop needs to be a complete story seed, not a label.

One outcome per line. The parser strips numbered prefixes like 1. or 2), so those are fine to include. The system splits on line breaks, bullets, and semicolons, so each line becomes exactly one prop — don't combine multiple outcomes on the same line.

Tone & length

Tone — choose one of casual, formal, witty, direct, or warm. Tone shapes the opener style and phrasing cadence. Some formatting rules override tone regardless of your choice — sentence case is always enforced, and em dashes are never allowed. Casual means conversational but not all-lowercase. Direct means no preamble, get to the point immediately.

Length — three options:

CTA & CTA link

The CTA is the action you want the reader to take, e.g. "15-min intro call," "grab 20 minutes," "see a 3-min demo." If you provide a ctaLink, the CTA text carries the link through drafting and QA. If you leave the link blank, it renders as plain text. Always include a link — it makes the ask one click instead of requiring the recipient to reply and coordinate.

CTA placement

A new brief control decides where the CTA lands in the body. It applies to every step in the sequence.

Per-step CTA overrides

Each step in the sequence builder has its own CTA mode: inherit (default — use the campaign CTA), custom (this step gets its own text and link — a blank link means no link, not the campaign link), or none (this step carries no CTA at all). Breakup steps and LinkedIn connection-request notes always resolve to no CTA regardless of what you pick.

Voice samples

Optional, but recommended. Paste 1–3 real cold emails you've sent before. The drafting model uses these to mirror your cadence, opener style, sign-off, and idiosyncrasies.

• Paste real sent emails — subject line, body, and signature included
• Include emails of similar length and step type to what you're generating (short intros if you want short intros)
• Include your actual voice tics: hedge words, how you sign off, your typical opener
• Don't paste email templates with merge tags — paste real, sent emails
• Don't paste emails that are much longer or shorter than your target length — the model picks up on cadence

Research intelligence: lookFor ⭐

This is what makes SimpleSend emails feel hand-written rather than AI-generated. Every lookFor entry is a research directive the system executes per contact before drafting. Each entry has two parts:

For each contact, the system runs up to 3 search strategies per entry and picks the first one that returns a finding with ≥70% confidence. That finding gets delivered to the drafter as an observation plus phrasing hint. The drafter uses 0–2 findings per email by design — findings are scarce on purpose.

Conditional play-rules in the angle

The angle field now supports conditional play-rules — "if the finding shows X, pitch A; if it shows Y, pitch B." The research planner captures these branches, and the drafter is told which branch the actual finding supports so it activates the matching pitch. This replaces the older "one angle for any finding" pattern.

Writing high-quality lookFor entries

The single most common mistake: a vague topic with no angle. The topic alone doesn't do much — the angle is what shapes the phrasing. Every lookFor entry should have both.

Recommended lookFor entries by use case

Each finding gets a confidence score and a recency label (very_recent, recent, stale, unknown). The drafter is told to downgrade stale findings. You don't control this directly, but you can bias it by writing topics that target recent signals (job postings from the last 60 days, announcements from the last 90 days).

User templates ⭐

Paste your own email body copy into the brief and the drafter will keep the fixed text close to verbatim, filling in only the parts you flag with [square-bracket directives]. This is the most direct lever you have over the exact wording — great for founders and senior AEs whose voice is the pitch.

Hey [first name] — [comment on their recent sales hires in one sentence].

It seems like ramp-time is a priority right now. We help RevOps teams
cut new-rep ramp from 90 days to 45 by [reference the closest value prop
in one specific sentence].

Worth a 15-min chat this week?

How directives work

• Fixed text is preserved almost verbatim — the drafter is instructed not to rewrite it.
• [bracket directives] are instructions the drafter replaces with personalized, research-grounded prose. Each directive typically expands to 1–2 sentences.
• Slot — each template is tagged intro / value / followup / breakup and only rotates into sequence steps of that kind.
• Per-step pin — open a step in the sequence builder and pick a specific template to lock that step to it, overriding rotation.

Sequence design

How to structure your step sequence for maximum impact — and what each step type actually instructs the AI to do.

Recommended sequence shapes

Standard 4-step (most common): Day 1 intro (prop #1) → Day 3 value (prop #2) → Day 7 followup (prop #3) → Day 14 breakup.

3-step lightweight: Day 1 intro (prop #1) → Day 4 followup (prop #2) → Day 10 breakup.

The focus field

Each step has an optional focus text field. Whatever you put there is passed to the drafter as a literal instruction. Use it to override default template behavior for specific steps:

type: "followup"
focus: "Open with the job posting finding if confidence is high. If no finding, lead with the CFO surprise-at-board-meetings angle."

Focus is a power-user lever. When you're tuning high-stakes campaigns, use it to inject step-specific logic the default type rules don't cover.

Template library

SimpleSend has 13 built-in email templates. You don't choose them directly — the AI assigns one per step in round-robin order so your sequence varies. Here's what's available and when each shines.

Choosing a model

You pick one model per campaign. The choice trades output quality against unit cost. All other pipeline steps (enrichment, research, fact-check, polish) always use Haiku regardless of your choice.

LinkedIn DMs

LinkedIn DM campaigns use a different pipeline. When your CSV contains linkedin_url instead of email, SimpleSend switches to the LinkedIn path automatically. Key differences:

• No subject line — LinkedIn DMs don't have one.
• No email signature.
• No markdown links — [text](url) format doesn't render on LinkedIn; plain text only.
• No polish loop — the DM path skips the fact-check/polish step.
• No template lock — DMs use different prompts entirely.
• LinkedIn profile parsing replaces homepage enrichment.
• Sequence opens with a connection request note — pre-connection, ≤300 characters, no CTA. Post-connection DMs follow.
• InMail is a separate step type — use it when you're paying for InMail credits and want a subject line plus a longer, email-style body that lands without needing a connection. Don't stack both a connect step and an InMail step in the same sequence — pick one entry point.

Hard rules baked into every draft

These rules are non-negotiable — enforced at the drafting model level and re-checked deterministically after drafting. You can't override them via brief fields, and they'll be corrected in the polish pass if the drafter violates them.

• No fabricated facts about the recipient. If no finding is available for a claim, the AI falls back to generic value props. Claims that fail fact-check are deleted and rewritten.
• No "I hope this finds you well" or "I came across your profile." These openers are on the forbidden list and will be linted out.
• No em dashes or en dashes — ever. Stripped via regex post-hoc as well. Use commas or rewrite.
• Subject in sentence case, 5–8 words. Never Title Case, never all-lowercase. Proper nouns are capitalized; nothing else is.
• One named third-party person max per email. If a finding names someone (e.g., a new hire), they can't be named again in a later step.
• Company name copied exactly from the contact row. Misspellings are auto-corrected using edit-distance matching before export.
• One value prop per email. The drafter does not repeat the same prop used in a prior step as the main pitch.
• Each email's opener uses a different angle from prior steps in the sequence.
• Persona-aware emphasis. The recipient's title maps to a persona bucket (decision maker / finance / IR / ops / technical / marketing / sales) and biases which angle the drafter emphasizes — no new facts, just where the weight lands.
• Off-ICP contacts are skipped. The pre-flight fit check drops confident non-matches with no research or drafting spend. Borderline rows still get generated.

Quick-reference checklist

Use this before running any campaign. If you can check every box, you're set up for strong output.

Brief checklist

• Description is one concrete paragraph — deliverable + category words
• ICP includes searchable signals (stack, stage, hiring patterns)
• 3 distinct value props — specific outcomes with numbers or named pains
• Prop #1 is your strongest hook (it runs in the intro)
• CTA link included — low-friction destination
• CTA placement chosen (or left on Drafter decides)
• User templates added if you want the drafter to riff on your exact copy — directives are concrete, not vague
• 1–3 real sent emails pasted as samples

lookFor checklist

• 2–4 entries (not more)
• Every entry has both a topic AND an angle
• Topics target concrete observables (not "growth signals")
• Topics reference recent time windows ("last 60 days")
• Angles specify which value prop to activate for each finding
• Where useful, angles are written as conditional play-rules (if X → pitch A; if Y → pitch B)

Sequence checklist

• Ends with a breakup step
• No consecutive intro steps
• 3 or 4 steps matches your 3 value props
• Length setting matches target recipient seniority
• Sample run reviewed before full batch
• LinkedIn: chose connection note vs. InMail intentionally — don't stack both as entry points

CSV checklist

• No blank first_name values
• company_website included where available
• title column present for ICP matching
• No duplicate rows
• Email or LinkedIn URL — not both in the same campaign

Common mistakes

The issues most often seen in campaigns that underperform — and how to fix each one.

1. One value prop for a 4-step sequence

What happens: all four emails argue the same angle. The AI tries to find fresh approaches but runs out of ideas — drafts become thin and repetitive.

Fix: write 3 distinct, specific outcome-based props before generating.

2. lookFor entries with no angle

What happens: the research planner finds a signal but doesn't know how to use it. Findings end up in emails as vague, non-specific references ("I noticed your company has been growing").

Fix: every lookFor entry needs an explicit angle — "if you find X, pitch Y by saying Z."

3. Pasting email templates as samples (not real emails)

What happens: the model mirrors the template style — generic, token-heavy, formal structure. Defeats the purpose of voice mirroring.

Fix: paste real, sent emails with your actual sign-off, opener habits, and phrasing quirks.

4. Medium length + xshort-type value props

What happens: the model is asked to write 120–200 words but your value props are one-liners. It pads with generic content to hit the word count.

Fix: match prop specificity to target length. For medium emails, write props with context: who benefits, what outcome, what the before state was.

5. Not running a sample before the full batch

What happens: a brief issue that would be obvious in one email — a vague description, a weak prop, a tone mismatch — gets multiplied across hundreds of contacts.

Fix: always use the Sample step. Review 2–3 outputs across different company types in your list before generating the full campaign.

6. Blank or incorrect company names in the CSV

What happens: the AI is given a blank or misspelled company name and uses it verbatim. SimpleSend auto-corrects close misspellings, but a blank company field produces "Hi [name], I noticed your company has been..." with no company name.

Fix: review and clean your CSV at the Review step. Prioritize the company and first_name columns.

7. Vague directives in user templates

What happens: a template with [intro line] or [personalize this] produces generic filler because the drafter fills the directive literally as written. Templates are prompts inside a prompt — vague ask, vague output.

Fix: write directives with concrete instructions and a length hint, e.g. [comment on their new CRO hire in one sentence, tie to fresh-stack evaluation] or [reference the value prop closest to their stated priority in one sentence].