How to Set Up a Cold Email Campaign: Every Setting, the Knowledge Base and Best Practice

A campaign in PitchButler is more than an email and a list. It is a set of instructions for an AI sales rep: who to look for, what it is allowed to say, when to follow up, when to hand a conversation to you and when to book the meeting itself. The better those instructions, the better every email and every reply.

This guide covers every input a campaign takes, in the order you meet it, with the setting we recommend for each. Bookmark it and keep it open the first time you build one.

Before you start: the five things that must be true

A campaign can be perfectly written and still send nothing. Before a lead is contacted, PitchButler checks a short list of gates in this order: the campaign is not paused, the kill switch has not tripped, a sending account exists, that account’s domain passes authentication, and there is room under the mailbox’s daily limit, the campaign’s leads per day and the send window. Get these right first:

1. Your company profile is filled in. In Settings (or step one of onboarding) you describe your company, industry, target audience and outreach goals. This is shared context for every AI step across all your campaigns.
2. At least one mailbox is connected. Connect your Google Workspace mailbox during onboarding or on the Sending page. We strongly recommend dedicated sending inboxes on a separate domain, so your main domain is never at risk. More in our deliverability guide.
3. SPF, DKIM and DMARC pass. A mailbox whose domain explicitly fails any of the three is taken out of rotation, and sending is held until it is fixed. Sending, then Sending domains, shows each record with a Check now button and the exact DNS records to add. There is an override to send without authentication; leave it off. See the SPF, DKIM and DMARC guide.
4. The mailbox is warming up. New mailboxes start at 5 emails a day and climb automatically to 50 over about two weeks. A mailbox that gets no replies at all stops climbing until one arrives, which protects your reputation. On a brand-new domain, give it three to four weeks before you judge results. See the warmup guide.
5. You have credits. Research, writing and reply handling run on monthly credits. If they run out, outreach pauses; replies keep flowing.

Optional but worth it: voice calibration. Under Settings, Your writing voice, you can let the AI read around a hundred of your own sent emails (read-only) and build a tone profile. Every AI-written email then sounds like you. Only the distilled style profile is stored, never your emails.

Step 1: Campaign basics

Go to Campaigns, then New campaign. The wizard has four steps: Basics, Lead Lists, AI Config and Review.

• Campaign name. Only you see it. Name it after the audience and offer, for example “Nordic SaaS CFOs, Q4 audit offer”, so your reports stay readable when you have ten campaigns.
• Timezone. Send times follow this zone’s clock. Pick the zone where most of your prospects are, not where you are. On top of this, PitchButler holds each send to 07:00 to 18:00 in the lead’s own local time when it knows it.
• Outreach language. The language for AI-written replies and follow-ups. Your own templates are always sent exactly as you wrote them, and if a lead answers in another language the AI mirrors the lead. Match this to your list: one language per campaign is cleaner than a mixed one.
• Leads to process per day. How many new leads are researched and get their first email each day. The default is 50, but start with 5 to 10 while you check the first emails, then scale up. The mailboxes’ daily limits always apply on top, and the wizard shows your real capacity and warns you if you ask for more than your mailboxes can send. See how many cold emails per day.
• Send window and send days. Default 07:00 to 17:00, Monday to Friday. Keep business hours for real campaigns. The 24/7 mode toggle is for testing only.
• Sending accounts. Use all connected accounts means mailboxes you connect later join automatically. Only the accounts I select freezes the campaign to exactly those. Leads rotate across mailboxes weighted by each one’s daily limit, so warmed-up inboxes carry most of the volume, and each lead keeps the same mailbox for the whole conversation. The list shows each mailbox’s warmup day, health and SPF/DKIM/DMARC status. Fix any red chip before you launch.

Step 2: Lead lists

Pick one or more lists you have imported, or click Import CSV. This step is optional: PitchButler can also find companies and people for you based on your ideal customer profile.

The CSV needs at minimum an email column, or just a column of companies. We recommend email, company_name, contact_name and domain. On import, keep these three options on:

• Remove duplicates, which catches exact and fuzzy matches.
• Auto-format columns, which maps any language or header format to the standard fields.
• Skip leads that already exist in database, so nobody gets two first emails.

Every extra column you import becomes a template variable. A column called recent_funding can be used as {recent_funding} in your copy. Leads on your do-not-contact list are never assigned, no matter which list they come from.

Best practice: a smaller, tighter list beats a big one. Verified addresses, one clear segment per campaign, and one or two people per company. See how to reduce your bounce rate.

Step 3: AI config, the part that decides quality

At minimum, fill in your value proposition and your sender name. Everything below makes the emails more specific.

The structured knowledge base

This is the most important section of the whole campaign. It holds the facts the AI is allowed to use when a prospect replies. With any field filled in, the AI is held to knowledge base only: it may not invent product details, prices or claims. Anything that is not in it, the AI will not say.

Start with Draft from my website. PitchButler reads your homepage plus your about and pricing pages and proposes all the fields. It only fills blank fields and saves nothing until you click Save. Then edit each one:

• Product description. Two or three sentences on what you sell and to whom, in plain words. Skip the slogans.
• Key features, one per line. Concrete capabilities, not adjectives.
• Pain points we solve, one per line. Write them the way your customer would say them out loud. The AI uses these to connect your offer to what it finds when it researches each lead.
• Pricing. This is a real decision. If you fill it in, the AI quotes exactly this and nothing more. If you leave it empty, the AI never discusses price and offers a meeting instead. Leave it empty if your price depends on scope; fill it in if a clear price is a selling point.
• Ideal customer profile. Who buys and which role. This also steers whom discovery looks for at each company.
• Case studies. A customer and the result they got. Numbers beat adjectives: “cut onboarding from 3 weeks to 4 days” beats “improved efficiency”. Two or three strong ones are enough.
• Competitors. Each competitor and your advantage over them. Prospects who already use someone else will ask.
• Objection handlers. The objection and how you answer it. Write the five you hear most: “we already have a tool”, “no budget right now”, “send me more info”, “not my area”, “how is this different”. This is where good reply handling comes from.
• Forbidden claims, one per line. Things the AI must never say, such as guarantees, customer names you cannot share, or features that are still on the roadmap.
• Tone guide. For example: “Direct, friendly, no jargon. Short sentences. Never pushy.”

There is also a free-text Knowledge Base box. It is only used when every structured field is empty, so once you use the structured fields, put everything there instead.

Guardrails and handoff

• AI Allowed Actions / AI Forbidden Actions. If you leave these empty, sensible defaults apply: no pricing unless your knowledge base has a pricing field, no emails longer than eight sentences, no emojis, and never mentioning that it is an AI. Add your own lines only for things specific to your business.
• Handoff Triggers, one per line. When should a human take over? Good triggers: “asks for a contract or proposal”, “mentions a procurement process”, “asks about security or compliance”, “wants to talk to the founder”.
• Disqualification Criteria. When a lead is clearly not a fit, the AI sends a polite closing message first and only then closes the lead. Be specific: “under 10 employees” or “only B2C” works better than “not a good fit”.
• Email Signature. Available as {signature} in every template. Keep it short: name, title, company, one link.

Meetings: let the AI book the call

You have two options, and you can combine them:

• Propose times automatically from my availability. First set your weekly windows under Meetings, Your availability. The AI can then propose concrete times, and it can never commit a time you did not authorize there. Choose whether meetings are confirmed automatically or after you approve, the meeting length (30 minutes by default) and how soon the earliest proposed time may be. The prospect confirms with one click.
• Booking link. Your own calendar link (must start with https). The AI only offers it to leads who are clearly interested and have replied at least twice, so it never feels like a hard sell on the first reply.

When both are on, proposed times win. Best practice: turn on proposed times. A reply that says “Tuesday 10:00 or Wednesday 14:00?” converts far better than a link.

Generic inboxes

What should happen with addresses like info@ or contact@? Send normally, Ask for the right contact (the AI asks who handles your topic and routes the pitch to that person) or Skip them entirely. Asking for the right contact is usually the best of the three when your list has many of them.

Ideal customer profile

Industries, geographies, employee count, annual revenue, technographics, role titles and exclusions. CSV imports flag rows that do not match and show you why, and the AI uses the profile when it scores replies and writes follow-ups.

Role titles matter most. They carry the most weight when PitchButler decides which person to contact at each company. List the titles of the person who owns the problem, not just the most senior person: “Head of Finance, CFO, Finance Director” rather than “CEO”. Our ideal customer profile guide walks through building one.

Your initial outreach email

This is the one email you must write. Both the subject line and the body are required.

With AI writes each opener from your email on (the default), the AI rewrites the opening for each lead using what it found in its research. Your text is the factual boundary: the AI can never invent offers, links or claims that are not in it. If it cannot write a good opener, or you switch to Verbatim, your email is sent exactly as written.

Before you save, use See 3 AI openers for your actual leads and Check spam risk. Read the openers as if you were the prospect.

Template variables use single braces: {first_name}, {company_name}, {title}, {website}, {city}, {signature}, plus AI research variables such as {personalization_hooks}, {suggested_angle}, {pain_points} and {company_summary}. Add a fallback with {first_name|default:there}. A malformed variable, for example one with a space in it, is rejected when you save and never sent.

What a good first email looks like:

• A short, plain subject line, lowercase is fine. See subject lines that get opened.
• An opening line about them, not you. See opening lines.
• One problem, one sentence of proof, one question.
• Under 100 words, plain text, no attachments, at most one link.
• A low-friction ask, such as “Worth a look?” rather than “Book 30 minutes”. See the cold email call to action.

Follow-ups

Most replies come from follow-ups, not the first email. By default PitchButler sends up to 4 follow-ups, starting 5 days after the first email, with each gap one day longer than the last: 5, 6, 7 and 8 days, about four weeks in total. Every send gets a little natural timing variation and lands inside your send window.

You can write follow-ups yourself under Templated follow-ups, sent as replies in the same thread (up to six steps including the opener). After your last templated step, the AI writes the rest, guided by Follow-up instructions. A good instruction: “Each follow-up adds one new reason to reply: a case study, a relevant insight, a simpler ask. Never just bump the thread.” See the follow-up sequence guide.

Conversation playbook

Optional stages the AI follows when a lead replies, one stage per exchange, with the last one repeating. A simple, effective playbook has three stages:

1. Understand. Answer their question and ask one question about their situation.
2. Relate. Connect their answer to a relevant case study.
3. Book. Propose a short call with concrete times.

Qualify and alert only

With this on, the AI reads and scores every reply and alerts you about hot leads, but never replies on its own. Use it for your first campaign. Watch how the AI scores real replies for a week, then turn it off when you trust it.

Step 4: Review and launch

The review page summarizes every step and shows what will happen after the first email. Check three things before you click Create & Launch: at least one sendable mailbox is selected, leads are assigned, and leads per day is low for the first week.

If you rely on discovery rather than an imported list, leads activate automatically as PitchButler finds their email addresses, so the first sends can take a little while.

Settings you only find on the edit page

Open the campaign and click Edit for these:

• Research pacing. How many leads are researched in parallel (3 by default), the minimum time between new leads, and how many days of ready-to-send leads to keep researched ahead (2 by default). The page shows the resulting steady state in leads per day. The defaults are right for most campaigns.
• Second contact at silent accounts. If a good-fit company stays silent, PitchButler can contact a second person there. The default is two people per company; choose “Off” for one person only.
• Recipient local hours. On by default. Keep it on.
• Referral asks. The copy used when the AI asks a generic inbox for the right contact, how many reminders it sends, and what happens if nobody answers.
• Custom AI prompts. An advanced override for the qualifier and responder. Most teams never need it: a good knowledge base does the same job more safely.

Reply handling: what happens when someone answers

• Approval mode (Settings, off by default). The AI drafts every reply but waits for you to approve it. A good choice for the first weeks, alongside or instead of Qualify and alert only.
• Hot leads. When a reply scores high enough, or matches one of your handoff triggers, the lead is handed to you. Set where alerts go under Settings, Where your alerts go: an alert email, a Slack or Discord webhook, and the daily briefing. If nobody picks up a hot lead, the escalation ladder pages you again after one hour and four hours, and after 24 hours sends the prospect a polite holding reply so they are never left waiting.
• Opt-outs are permanent. When someone asks not to be contacted, they are blocked for good across every campaign. There is no undo, and that is deliberate. See is cold email legal under GDPR.
• Kill switch. If bounces go above 5 percent over seven days (after at least 50 sends) or spam complaints above 0.3 percent, sending pauses automatically to protect your domain. Fix the list before you resume.

A/B testing once you have volume

When the campaign is running, you can turn your email into variant A and add up to four challengers. New leads are split between the variants; leads already contacted are unaffected. A winner is only declared after 100 sends per variant with a statistically significant difference, and Autopilot can promote it automatically. Test one thing at a time. See how to A/B test cold emails.

The short version: our recommended first campaign

1. Company profile filled in, voice calibration done.
2. Dedicated sending mailboxes with SPF, DKIM and DMARC passing, warmed up.
3. One narrow segment, verified list, one or two people per company.
4. All nine knowledge base fields filled in, with real case studies and your five most common objections.
5. Role titles in the ICP set to the person who owns the problem.
6. A first email under 100 words with one soft question.
7. Four follow-ups with a new reason to reply in each.
8. Proposed meeting times on, availability set.
9. 5 to 10 leads per day, business hours, Qualify and alert only on for the first week.
10. Review the first replies, tighten the knowledge base where the AI hesitated, then scale.

The knowledge base is the step most people rush, and it is the one that decides whether replies sound like your best salesperson or like a polite stranger. Spend your first hour there. For the bigger picture, read cold email best practices or see how PitchButler works.