B2B SaaS outreach: how to do cold outbound in 2026

What makes B2B SaaS outreach different

In most industries, cold outbound competes with other salespeople. In B2B SaaS it competes with people who run outbound themselves. Your prospect is a founder, an engineer, or an operator at a software company. They have sent cold email. They recognize the templates, the merge tokens, and the manufactured familiarity, because they have used them.

AI writing tools made this worse. A researched-sounding sequence now costs nothing to produce, so a SaaS buyer's inbox is full of messages that mention their company and say nothing about it. The result is a filter those buyers now run by default: could this exact message have been sent to anyone else? If yes, it gets deleted, usually without being consciously read.

That filter is the constraint the rest of this page works around. Nothing below is a trick for getting past it. The only reliable way through is a message that could not have been sent to anyone else. That is expensive to produce, which is exactly why it works.

Targeting before copy: ICP, list quality, bounce hygiene

Most outbound advice starts with the email. Most failed campaigns die earlier than that, in the list. Work in this order, and do not skip ahead.

• Define an ICP narrow enough to change the message. Stage, team shape, and the trigger that makes your problem current for them this quarter. The test: reading the ICP tells you what the first line of the email should mention. If it does not, it is a market description, not an ICP.
• Verify every address before it enters a campaign. Database exports and scraped lists decay fast. Keep bounces under 2%; above that, stale addresses damage sender reputation, and every later send pays for it. The thresholds and the studies behind them are on the cold outbound benchmarks page.
• Fix deliverability before touching copy. A reply rate means nothing if the emails sit in spam. If replies are under 1%, work the diagnostic order on the benchmarks page: bounce rate, then a placement test, then and only then the copy. Most sub-1% campaigns are a list or deliverability problem, not a copy problem.

For SaaS specifically, small and specific beats large and plausible. A hundred companies that match the trigger will out-reply a thousand that match the industry code, and the smaller list is the only one a founder has time to research properly anyway.

How to write a cold email subject line

Short, specific, and plain. Two to four words that name the thing the email is about: "your onboarding flow", "the checkout bug", "QA without a QA team". The subject line has one job, which is to not look like a campaign. Internal email has plain subjects, so plain subjects get opened.

Clever subjects, questions with the prospect's first name in them, and fake "Re:" threads are all campaign signals now, and your prospect sends campaigns for a living. Also skip optimizing for opens entirely: open tracking has been unreliable for years, which is covered on the benchmarks page. Write the subject for the person who will already open it, as the label above your first line in the inbox preview.

How to start a cold email

With the observation, not with yourself. The first line is the whole message in miniature: it should contain something you could only write after looking at their product, their docs, their changelog, or their job posts. "You are hiring three backend engineers and no QA" passes the test. "I came across your company and was impressed" does not. Neither does "congrats on the raise", which AI sequences now generate automatically off funding announcements.

Skip the introduction of yourself and your product. The prospect will read your signature and your domain if the first line earns it. "I hope this finds you well" costs a line and announces a template.

How long should a cold email be

Under 100 words is the usual advice, and it is roughly right, but length is a symptom rather than a rule. The email should be exactly as long as the specific thing you have to say: the observation, why it matters to them, one ask. If that takes 60 words, stop at 60.

What kills replies is not word count, it is padding: the paragraph about your company, the second feature, the list of logos. Every sentence that could be deleted without losing the specific point should be deleted. And keep it to one ask. Two asks usually gets neither answered.

How to write a follow-up that is not just a bump

"Just bumping this to the top of your inbox" adds nothing and reads as automation, because it usually is. A follow-up earns its send by adding something: a new observation, a sharper version of the ask, a piece of work you did since the first email. If the first email showed you looked at their product, the follow-up can show you looked deeper.

Two or three follow-ups per thread is enough on email. Space them out: a few days after the first send, then about a week. Send them in the same thread, so the prospect sees the original message above your follow-up instead of a context-free nudge.

How many touches before you stop

Across email and LinkedIn together, four to six touches over two to three weeks, then stop. This is a judgment call, not a measured law. The public datasets do not agree on an optimal touch count, and the ones that exist come from volume senders, so treat any precise answer you read with suspicion.

The reasoning is simpler than the data. By the sixth unanswered touch, the reply you are most likely to get is an annoyed one, and the prospect you are most likely to convert is one you have not contacted yet. Silence after a good sequence is information: it usually means the list is wrong or the message did not show enough work. The fix is upstream, not an eighth touch.

What a proof-of-work message looks like

Everything above raises the floor. It does not solve the core problem, which is that well-written, researched-sounding copy is now exactly what AI sequences imitate. The way out is to put something in the message that cannot be templated: work.

A proof-of-work message carries evidence that someone spent real effort on this specific prospect. For a SaaS company, the strongest version is your own product already working on the prospect's business: their site, their data, their use case, before the first call. The message stops claiming a demo would be worth their time and includes the demo. We call this demo-led outbound.

This is the one section of this page about our product. driftwood (driftwood.sh) is an AI SDR that runs demo-led outbound: for each prospect, the agent does the research, builds a working demo of your product for their business, and drafts the message. You approve every send, and messages go out from your own email and LinkedIn accounts. At our first customer, on the same lead list the founders had been working themselves, replies went from under 1% to over 14% in week one at Autosana (YC S25), including founders who had ignored more than four months of prior outreach. That is one company and one week, observed once, not promised; the full writeup is in the Autosana case study.

You can also run proof-of-work outreach by hand: audit a prospect's public signup flow and send what you found, or build a small working example against their API. It works. The cost is time, and the arithmetic of that time is the founder-led sales page.

Realistic funnel expectations

Plan with public numbers, not with a vendor's case studies. The short version, with named sources and links on the cold outbound benchmarks page: at average quality and at scale, cold email replies now measure under 1% (Belkins' 2026 study of 7.5 million sends measured 0.45%). Small, carefully targeted, heavily personalized campaigns still measure between 5% and 18% in the public studies. On LinkedIn, connection accepts average roughly 19-30%, and message replies run about double cold email replies in the same datasets.

So a realistic founder-run plan looks like this: 50 to 100 well-researched sends a week, a reply rate in the mid single digits if the list and the message are good, and one or two meetings a week from it. The full arithmetic, including where the hours run out, is the funnel table on the founder-led sales page. If your numbers sit far below these, start with the diagnostic order on the benchmarks page, not with the copy.

Common questions

These are the questions founders ask about the channel itself. Product questions about driftwood are on the FAQ.