How MimikFlow was built: three months in n8n, one month of code, and a product that finds its own customers

MimikFlow is an AI agent for LinkedIn prospecting. You connect your LinkedIn account and describe who you want to reach; it finds prospects, writes a personal first message, follows up, answers replies and books meetings in your calendar.^[6] It is aimed at B2B consultants, coaches, agencies and freelancers.^[6]

It was built by Mo Alani, a Paris-based data scientist with more than ten years in AI who, by his own account, hates prospecting.^[1][5] Founder Trace, a French newsletter for bootstrapped founders, describes his background as Telecom ParisTech, then AI work at DigitalGenius and leading a generative-AI team at Bpifrance.^[4] This piece puts together his Indie Hackers launch post, his 28-minute Founder Trace interview from May, MimikFlow's own site and two Stripe-verified revenue pages. Unless a line says otherwise, the numbers are his own.

Before the SaaS: his own system, then a service

MimikFlow started as a tool for himself. In the interview he says he finds prospecting techniques fascinating but doing it by hand robotic: look up the person, put them in a CRM, follow up, track it. So he automated it, and at one point he had four or five bookings a week, more than he could handle.^[5]

People around him wanted the same thing, so at first he sold it as coaching and installed the system for them. Then he decided a subscription product would be simpler than setting it up again and again.^[5][4] He says he had built things nobody wanted before, and this time did it the other way round:

The build: n8n first, then a month of code

The real prototype was in n8n, the visual automation tool. He spent three months there building the AI message logic, the sequencing and the LinkedIn orchestration, and tested it with real outreach. Only then did he spend about one month of focused development turning it into a SaaS, which he calls "mostly plumbing".^[1] His advice to other builders is blunt:

He doesn't say how much of the code Claude Code wrote. He does say he is "bad at front-end" and can't design, and that ten years in AI mean he knows which tool to use for what.^[5]

Much of the engineering went into making the agent safe and believable rather than into features. He describes hard caps under LinkedIn's limits, random delays, and replies split into several short messages, each sent after a pause based on how long a person would take to type it.^[5] Users can run it fully automatic or in a validation mode where they approve every message first; there is also a sandbox to test the AI on a fake profile without sending anything.^[5]

His worst bug shows why that mattered. A field wasn't passed through, so the AI signed a user's messages with Mo's own first name, once or twice. The user caught it in validation mode before anything went out, and Mo then added "coherence" checks: a second AI pass that reviews each message for tone, name and context before it is sent.^[5][4]

Pricing: up, then into tiers

Two weeks after launch he wrote that he had priced too low at €97 a month, moved to €197, and saw no drop in conversion: "the people who care about reply rates aren't price-sensitive."^[1] By the May interview the offer was two plans: Pro at €97 for finding leads and messaging, and Premium at €147 with the AI Setter that runs the conversation. Premium sold best.^[4][5]

Today there are three plans: Discovery at €49 a month (it finds and writes, you reply), Pro at €97 with the AI Setter, and Agency at €167 for several LinkedIn accounts. The free trial went from 7 days to 14 days, still with no card.^[6][1] The constant through every change: the conversation-handling setter is what the main paid plan is built around.

Growth: the product is the sales team

His first customer came from a LinkedIn post to his own network, where he had been writing about AI for customer acquisition before he launched.^[5] He says he posts on LinkedIn every day.^[5] In May he estimated that about seven of his 14 paying customers came from MimikFlow prospecting for MimikFlow, and the rest from his LinkedIn posts.^[5] Founder Trace's writer is an example: MimikFlow messaged him, he thought he was talking to Mo, and only later learned it was the AI.^[4] The ad budget at the time was zero.^[4] Today the homepage still says the same system books Mo about five qualified meetings a week.^[6]

The pitch was a contrarian one in a crowded market: fewer, personal messages instead of templates sent by the thousand, and no Chrome extension or scraper. He wrote that half his early trial users came because a Chrome extension tool had burned them before.^[1]

He watched onboarding closely. His main number was trial-to-paid conversion, which Founder Trace reports at 20%; he also tracked how many sign-ups connected LinkedIn and filled in their profile.^[4] When people stalled at the LinkedIn step, he didn't chase them one by one. He fixed the screen, added automatic emails for each step left undone, added an AI that pre-fills the profile from the user's own, and wrote reassurance copy about passwords and control.^[5] He also asks people who don't convert what stopped them.^[4]

Since then the site has grown a set of growth assets you can see today: a 30% recurring affiliate program (via PromoteKit) that gives referred customers 10% off^[8]; comparison pages against Gojiberry, Waalaxy, La Growth Machine and lemlist^[9][6]; six free calculators and message generators^[10]; a 2026 "observatory" of LinkedIn prospecting benchmarks from 98,426 prospects across 258 accounts, signed by Mo^[11]; named customer results on the homepage^[6]; and new channels: WhatsApp, plus an MCP connector to run it from Claude or ChatGPT.^[6][12] He hasn't said publicly which of these drove the climb that started in July, and we didn't find a post where he explains it.

The numbers

Two weeks in, his launch post reported $500 MRR, 5 paying users and 12 in trial, and a 54% reply rate with 15% booked on users' campaigns.^[1] At the May interview it was 14 paying customers, about 100 users, €1,400 MRR, about €650 a month in costs and roughly 53% gross margin.^[4]

The Stripe-verified monthly revenue on Indie Hackers tells the rest:^[2]

On November 2, 2026, Indie Hackers showed $11,195.44 in the last 30 days^[2], and TrustMRR, reading the same business through a Stripe API key, showed $11,328 MRR from 84 active subscriptions and $35,973 all-time.^[3]

Timeline

1. Before launch

   Automates his own prospecting; sells the setup as coaching; three months of n8n prototyping.^[5][1]

2. Early April 2026

   Launches the SaaS, built in about a month; first customer from a LinkedIn post.^[5][1]

3. Apr 16

   Indie Hackers launch post: $500 MRR, 5 paying users, price raised from €97 to €197.^[1]

4. May 4

   Founder Trace interview: 14 paying customers, €1,400 MRR, 20% trial-to-paid.^[4]

5. May-June

   The low point: $824 and $971.65 in Stripe revenue.^[2]

6. July

   Revenue more than triples to $3,456.77.^[2]

7. September

   Benchmarks from 98,426 prospects go up on the site (data read Sep 11, updated Sep 30).^[11]

8. October

   $11,414.92 for the month.^[2]

9. Nov 2

   $11,328 MRR from 84 subscriptions, Stripe-verified on TrustMRR.^[3]

Lessons for builders

1. 01

   Prove the hard part where it's cheap

   The value was in the message logic and sequencing, so he proved that in n8n with real outreach first. The code month was plumbing on top of something that already worked.^[1]
2. 02

   Sell it by hand before you build the product

   People paid him to install the system before MimikFlow existed. That was his validation, not a survey.^[5][4]
3. 03

   If you can, make the product your first channel

   About half his first paying customers were prospected by MimikFlow itself. Every booked call was also a demo.^[5]
4. 04

   Compete where the others stop

   Lots of tools send messages; few handle the reply. He built the paid plan around the setter that runs the conversation, and that plan sold best.^[4][6]
5. 05

   Put rails around AI output before you scale it

   Validation mode caught the wrong-name bug before it reached a prospect; a second checking pass came after. Start new users in a mode where they approve every message.^[5]
6. 06

   Fix the onboarding screen, not the users

   He tracks each onboarding step and fixes drop-offs with UI changes, automatic emails and AI pre-fill, rather than chasing people one by one.^[5]
7. 07

   Don't read the first months as the verdict

   Stripe revenue fell from $1,873 in April to $824 in May, then grew every month to $11,415 in October.^[2]
8. 08

   Turn product data into marketing

   Our read: the 98,426-prospect benchmark report, free calculators and comparison pages are content only a product with real usage can make.^[11][10][9]