Growth hacking

The term comes from a 2010 post by Sean Ellis, 'Find a Growth Hacker for Your Startup'. He described a person whose true north is growth and who relies on data, creativity and curiosity to find it. The word then came to describe a whole way of working.

The core idea is simple: instead of one big marketing plan set for the year, you run a string of small experiments. Each experiment has a hypothesis, a measure and a time limit. If it works, you scale it. If not, you drop it without regret.

Growth hacking got a bad name from the dubious tricks filed under it: fake accounts, mass sends, aggressive automation. Those are abuses of the method, not the method itself.

The loop has four steps. Find ideas by looking at the numbers and at customers. Rank them by expected impact, confidence and effort. Test them on a small volume and for a duration fixed in advance. Analyze, then decide.

Many teams rely on a reference funnel to know where to look, such as the AARRR model proposed by investor Dave McClure: acquisition, activation, retention, referral, revenue. Each step has its metric, and each experiment targets the weakest step.

Discipline matters more than creativity. An experiment without a written hypothesis or a measure defined in advance only produces an impression, and impressions contradict each other from one meeting to the next.

Turning content into a source of prospects: publish a useful analysis, then contact the people who reacted with a message that extends the topic. In MimikFlow's 2026 LinkedIn Prospecting Observatory (published in French), people who had reacted to a post replied 74.6% of the time, against 35.6% for strangers found through search.

Testing two first messages on two halves of the same target, then keeping the better one. Testing two targets with the same message to see which one deserves the effort. Offering a free template or tool in exchange for a work email.

None of these examples is spectacular. In B2B, growth hacking mostly looks like a stack of small measured gains that end up changing the result.

Another field for experiments that often gets ignored: prospects you already contacted. People who accepted your invitation and never replied, or who said 'later' six months ago, form a base that costs nothing to re-approach with a new angle. Testing what brings them back sometimes pays more than a new list.

Before you start, set the main metric, the volume needed and the duration. For a first LinkedIn message the metric can be the reply rate, on at least a hundred sends per version, with a week of waiting: in the 2026 Observatory, 84.8% of replies arrived within seven days of the first message.

Compare comparable populations. A test that sends version A to profile visitors and version B to strangers does not measure the messages, it measures the sources.

Keep a log of experiments, including failed ones. Knowing that an idea did not work six months ago saves you from retesting it by accident, and helps newcomers avoid starting from scratch.

LinkedIn frames what you can do: invitation volume, message volume, the use of automation tools. Tricks that promise hundreds of invitations a day expose your account to a restriction, and a restricted account produces no growth at all.

The other limit is human. A trick that lifts reply rates but annoys prospects ends up costing you in reputation. In B2B, where the same decision makers meet at the same events, that cost arrives quickly.

MimikFlow lets you test without taking that risk. Volumes stay within cautious limits set per account, and you can compare two versions of the first message within one campaign to keep the one that gets replies. The Test the AI workspace also lets you try a setting without sending a single message.