Which marketing channels should a startup start with?

1. Blog
2. Demand & Channels

By Saroj Jha · October 1, 2026 · 10 min read

Start where your first segment already looks for answers, and choose channels whose cost fits what a customer is worth. Then test one or two at a time against a pass mark you set before you start.

In my view, founders rarely struggle for lack of channel ideas. The more common problem is running six half-funded experiments at once, reading the results too early, and concluding that "marketing doesn't work for us". The problem is usually the order, not the channels.

This post sets out the order I would work through with a founder, in five steps. One example company runs through every step.

Running case · Example Co. (fictional)

The running case: Example Co. (fictional). Example Co. is a seed-stage EdTech startup with an AI reading coach: children in kindergarten to grade 3 read aloud to it, it flags mistakes, and adults get a weekly progress report. In the Customer Segmentation Playbook it chose its first segment: independent tutoring centers with 2 to 10 tutors that already pay for scheduling software. It charges $200 a month per center, and its goal for the next two quarters is 40 paying centers from a named list of 400. Now it has to choose the channels that will reach those owners. Example Co. and all its numbers are illustrative.

Step 1: Start from the segment, not the channel list

Most teams start with the channel they know, the one a competitor uses, or the one an investor mentioned. The better order runs from the customer outward: which segment you are selling to, how it buys, where it looks for answers, and only then which channels to test. If you have not chosen a segment yet, do that first; the Customer Segmentation Playbook walks through it.

There is no shortage of options. Gabriel Weinberg and Justin Mares list nineteen traction channels in Traction (Weinberg and Mares, 2015). For a B2B startup, I group them into seven:

• Owned: your website, SEO and AI search, content, email and newsletters.
• Earned: PR, podcasts, speaking, reviews, word of mouth.
• Paid: search ads, paid social, sponsorships, retargeting.
• Outbound: founder-led and sales-led outreach to named accounts.
• Product-led: free trials, freemium, referral loops built into the product.
• Partner: integrations, co-marketing, resellers, other companies' audiences.
• Community and events: online communities, meetups, conferences, webinars.

Buyers use many of these at once. McKinsey's 2026 survey of about 4,000 B2B decision makers found buyers use an average of ten channels across a purchase, and generative AI has moved into the top five for finding and evaluating suppliers (McKinsey, 2026). You cannot be everywhere, which is why the order matters.

Running case · Example Co. (fictional)

Example Co., Step 1. Before naming a single channel, the team writes its segment at the top of the plan: independent tutoring centers with 2 to 10 tutors, serving K–3 readers. The owner decides and pays, tutors use the product daily, and parents receive the weekly report. Every channel idea now has to pass one test: does it reach these owners?

Do this: write your segment and its buyer in one line at the top of your channel plan, and cross out every channel idea that does not reach that buyer.

Step 2: Find where your buyers already look

B2B buyers now do most of their research on their own, from several sources, before they talk to you:

• They prefer to buy without a salesperson. In Gartner's survey of about 650 B2B buyers in August and September 2025, 67% said they prefer a rep-free experience (Gartner, March 2026).
• They use several sources and AI. A separate Gartner survey, published in May 2026, found buyers used an average of seven information sources in a recent purchase, and 45% used generative AI, mainly to gather information on vendors and products (Gartner, May 2026).
• Many people are involved. Forrester's 2026 research, from nearly 18,000 buyers, found that on average 13 internal and 9 external people influence a decision (Forrester, 2026).
• They want to try before they buy. More than 60% of buyers in the same research purchased some form of trial.

The practical meaning: your owned channels (website, content, answers that AI tools can cite) and the places your buyers' peers talk matter more than they did, because that is where most of the decision is made. The GEO and AEO Playbook covers being findable in AI answers.

Running case · Example Co. (fictional)

Example Co., Step 2. The team calls back the four center owners from its segmentation interviews, plus six more (fictional). Most say they hear about new tools from other owners, in owner communities and associations, and through the scheduling software they already use. Few would think to search for an "AI reading coach". So the team knows where to show up, and that search may stay weak until owners know the category exists.

Do this: ask your last ten customers or prospects where they first heard about the tools they use today, and count the answers.

Step 3: Shortlist the channels that fit your deal size

Match the cost of a channel to what a customer is worth. A channel that needs a salesperson and three meetings cannot pay for a $50-a-month product, and a cheap self-serve funnel rarely closes a $60,000 deal with a buying group of fifteen. This is my working view, not a benchmark:

Then apply three filters to each candidate:

1. Is my segment already there? If buyers do not search for the category yet, search will not work until you create the demand. My post on demand creation versus demand capture covers the difference.
2. Can I afford a fair test? A channel you can fund for only two weeks will tell you nothing.
3. Do I have the skill or time? Founder-led content needs a founder who will write every week.

Keep the three or four channels that pass.

Running case · Example Co. (fictional)

Example Co., Step 3. At $2,400 a year per center, Example Co. sits in the sales-assisted band. It starts with seven ideas and runs each through the filters. Search is parked because owners do not search for the category yet. Trade shows are out because the cost per conversation is too high for a $2,400 contract. Podcasts and PR are out for now because no one has time to pitch every week. Four channels remain.

What Duolingo's and HubSpot's filings show

Two well-known companies describe very different routes to customers in their IPO filings. These are what the filings say, not an account of how either company made its choices internally.

Duolingo grew mainly through word of mouth. Its 2021 registration statement says: "All of this has allowed us to grow our business organically, primarily relying on word-of-mouth virality rather than paid user acquisition." It adds that its brand "drives organic traffic to our platform and minimizes our reliance on paid marketing", and that "As of March 31, 2021, approximately 5% of our monthly active users were paid subscribers of Duolingo Plus" (Duolingo S-1, 2021). That route suits a free consumer app; it is much harder for a paid product sold to a few hundred businesses.

HubSpot leaned on partners and its own content. Its 2014 registration statement says: "We focus on selling our platform to mid-market businesses, which we define as businesses that have between 10 and 2,000 employees." It reached them "through our proven inbound go-to-market approach and more than 1,900 marketing agency partners worldwide". Partners mattered: "Marketing agency partners and customers referred to us by our marketing agency partners represented approximately 42% of our customers as of June 30, 2014, and 33% of our revenue for the six months ended June 30, 2014." Its blog attracted "approximately 1.5 million visits each month" (HubSpot S-1, 2014).

For Example Co., HubSpot's pattern is the closer one: small businesses, reached through partners who already have their trust. Duolingo's lesson still applies on a smaller scale: give happy owners a reason to tell other owners.

Do this: write your shortlist with one line per channel: why your segment is there, what a fair test costs, and who will run it.

Step 4: Set a pass mark, then run short, fixed tests

Use the Bullseye method from Traction: brainstorm every channel, rank your top three, run small, cheap tests on those three, and focus on the one that works (Weinberg and Mares, 2015). I add three rules to make each test readable:

1. Set the pass mark before you start. Work out the most you can pay for a customer from your margin and payback target (the Unit Economics Calculator does this), then turn it into a cost per qualified meeting or trial you can accept.
2. Time-box it. Give each test a fixed budget and a fixed window long enough to reach a few qualified conversations, usually four to eight weeks for B2B.
3. Measure the next step, not the click. Count qualified meetings, trials that activate, or pipeline created. Clicks and impressions tell you almost nothing about fit.

Running case · Example Co. (fictional)

Example Co., Step 4. A center pays $2,400 a year. The team assumes an 80% gross margin and wants to earn back what it spends to win a center within 12 months, so it can pay at most $1,920 per new center. It aims to win more than 25% of qualified demos, so it can pay at most about $480 per qualified demo. Each shortlisted channel gets six weeks and a fixed budget. All these numbers are fictional or assumed.

Do this: write your pass mark as one number per channel, such as "cost per qualified demo at or below $X within six weeks", before you spend anything.

Step 5: Focus on what passes, and keep one small brand channel running

Run one or two channels in earnest, plus at most one small test. A seed-stage team that spreads a small budget across five channels ends up with five inconclusive results. Weinberg and Mares make the same case: find the one channel that works and focus on it until it stops growing.

Should some budget go to brand rather than demand? Research from the LinkedIn B2B Institute, by Les Binet and Peter Field, found that B2B marketing efficiency peaked at around 46% of budget on brand and 54% on activation (Binet and Field, 2019). The Institute's summary page rounds this to 50/50. The study is from 2019, rests on fewer than 50 B2B cases, and describes established advertisers rather than seed-stage startups, so treat it as direction, not a rule.

The reason brand matters in B2B is timing. John Dawes at the Ehrenberg-Bass Institute estimated that only about 5% of business buyers are in the market in any quarter, because companies change providers only every few years (Dawes, 2021). That figure is a derived rule of thumb, not a survey result, but the point holds: most of your future buyers are not buying today. For a seed-stage startup, I would put most of the budget on reaching buyers who are in the market now, and keep one low-cost, always-on brand channel, usually founder-led content, running from the start.

A channel is working when it meets your pass mark for two consecutive periods and the cost stays acceptable as you put more into it. When you add the next one, take it from your shortlist, not from whatever is fashionable.

Running case · Example Co. (fictional)

Example Co., Step 5. After six weeks (fictional results), association webinars bring qualified demos at $310 each, well under the $480 pass mark, so the team keeps them and adds budget. The scheduling-tool partner listing went live three weeks late, so that test is extended by four weeks. Paid social costs $900 per qualified demo and is stopped. One founder posts once a week in two owner communities, as the always-on brand channel.

Do this: put a review date on each test now, and write down which result means keep, extend or stop.

Which one channel should a B2B SaaS startup start with on no budget?

With no budget, a B2B SaaS startup should start with a founder-led channel, because it costs hours rather than dollars. In my view, choose founder-led outbound if you can name the companies you want and your price justifies a one-to-one conversation. Choose founder-led content if your buyers are many, smaller, and gather in one public place, such as LinkedIn or a professional community. If you are unsure, start with outbound: replies tell you within days whether your segment and message are right, while content takes months to build an audience.

How to pick between them:

• Outbound fits when you can list a few hundred target companies, the buyer's role is clear, and a first-year deal is worth more than roughly $1,000 (my working view, not a benchmark).
• Content fits when the price is low, buyers are hard to list by name, and a founder has a real point of view and will publish every week.

A weekly routine for outbound: research a fixed number of accounts you can sustain, write a short, specific first message to each, follow up twice, and log every reply. Rewrite the message every second week based on what people answer. In the US, the CAN-SPAM Act covers cold B2B email too: the FTC says "the law makes no exception for business-to-business email", each commercial message must include a valid physical postal address, and opt-out requests must be honored within 10 business days (FTC, CAN-SPAM Act compliance guide). This is orientation, not legal advice. Other countries have their own rules; check them before you send.

A weekly routine for content: publish one or two posts built from real customer conversations, spend time answering comments and other people's posts in the same space, and end posts with one clear next step.

What to measure (not likes or impressions):

• Outbound: positive reply rate, meetings with companies that fit your ideal customer profile, and meetings that reach a second step.
• Content: conversations started by your posts (messages, demo requests that mention a post) from people who fit your profile.

Set a pass mark before you start and run the test for four to eight weeks. The common mistake is switching after two weeks, or automating volume before the message works. At Series A, hand the working channel to a first marketer or salesperson once it passes the mark twice.

What is the fastest way to get B2B SaaS leads without paid ads?

The fastest non-paid source of B2B SaaS leads is usually people who already know you, then founder-led outbound, because both can produce conversations within days. Partners, communities, founder content and search are slower but compound over time. The timings below are my rough expectations, not benchmarks; they vary with deal size, segment and how well you know the market.

| Channel, fastest first | First steps | First signal (my view) | What to measure |

|---|---|---|---|

| Warm network and referrals | Ask customers, advisors and investors for two named introductions each | Days, if you have a network; volume is limited | Introductions made, meetings held |

| Founder-led outbound | Build a named account list, send short personal messages, follow up twice | Replies in days, meetings in one to three weeks | Positive reply rate, qualified meetings |

| Partners | Pick one partner who already serves your buyer, offer a joint webinar or listing | One to three months to set up and launch | Qualified leads per partner activity |

| Communities | Join where buyers ask questions; answer helpfully before you mention your product | Weeks to earn trust | Conversations that start from the community |

| Founder content | Post weekly from real customer conversations | Months to build an audience | Conversations and demo requests citing a post |

| Search (SEO and AI answers) | Write pages for questions buyers already ask | Often a year or more to build authority | Qualified sign-ups or demos from search |

How to use the ranking. Your network and referrals run out quickly, so pair a fast channel with one slower channel that compounds. At pre-seed, outbound plus warm introductions gives the quickest learning. At Seed, keep the channel that passed your test and start one compounding channel. At Series A–B, the compounding channels carry more of the load.

On conversion rates: no reliable universal figure exists across segments and price points, so I don't give one. Set your own pass mark: work out what you can afford to pay for a customer from your margin and payback target, then turn it into an acceptable cost, in hours or dollars, per qualified meeting.

The common mistake is starting all six at once. Each gets a few hours a week, none produces readable results, and the team concludes nothing works. Run one fast and one compounding channel, and judge both after four to eight weeks.

What changes by stage?

Pre-seed. Founder-led conversations and one owned channel. The goal is learning who buys and why, not volume.

Seed. Test two or three channels from your shortlist with clear pass marks. Commit to the one that works.

Series A–B. Scale the working channel, add a second deliberately, and start measuring incrementality so you know what each channel really adds. The incrementality testing guide explains how.

What mistakes should I avoid?

• Too many channels at once, each too thin to read.
• Judging a test too early, before enough qualified conversations have happened.
• Copying a competitor's channel mix without knowing whether it works for them.
• Copying a famous company's route without its conditions. Word-of-mouth growth like Duolingo's came with a free consumer app; a paid product for a few hundred businesses usually needs partners or outreach too.
• Measuring clicks instead of pipeline.
• Ignoring how buyers actually research, including peers and AI answers.

Use this on AAJ

• Which channel should we start with?: eight questions about how your buyers buy, and a starting channel recommendation with a test plan.
• Unit Economics Calculator: Step 4. Work out what you can afford to pay for a customer before you set a pass mark.
• Paid Media Budget Allocator: split a paid budget across channels once you know your cost targets.
• Customer Segmentation Playbook: Step 1. Choose the segment before the channel; it also introduces Example Co.
• SEO for startups: Step 3. If SEO makes your shortlist, when to start it and what to do first.
• Cold email for founders: Step 3. If founder-led outbound makes your shortlist, how to write the sequence and meet the sending rules.