Fractional CMO – Bootstrapped EdTech

Deliverables at a Glance

What was actually produced

Phase 1

Building the strategic foundation, not skipping to tactics

The instinct on a pre-revenue product is usually to start writing ads. That was deliberately resisted. Before paid acquisition was scaled, the question was: what does this product actually compete against, and who specifically is it for — because getting either wrong makes every downstream dollar less efficient.

Mapped 6 real competitors not just by feature list but by business model, target level, and — critically — where their reviews showed real users being disappointed (billing trust failures at one competitor, no speaking practice at another, broken search at a third). That second layer mattered more than the feature comparison: it showed where trust breaks down in this category, which became a standing design requirement later (billing transparency, easy cancellation) rather than a nice-to-have.

Plotted the market on a positioning map across content depth, human connection, community, and level focus — and found a real, unclaimed intersection: nobody combined a named instructor, a curriculum built entirely on song lyrics, and a paid community, aimed at learners who are past beginner and want cultural fluency rather than a passing grade. That intersection, not a generic "we're different" claim, became the spine of every later positioning decision.

Rather than accepting "songs help you learn English" as a marketing platitude, checked the actual peer-reviewed research — and found the real picture was more useful than the platitude: passive song-listening alone does not reliably teach vocabulary; the benefit only shows up with structured explanation around the song. That caveat became a useful strategic distinction, not a problem — it directly validates a teacher-led explanation model over the pure lyric-drill mechanic every competitor uses, and gave a legitimate, citable reason to reject a competitor's own unverified "3x retention" marketing claim rather than accidentally repeat it as fact.

Phase 2

Segmentation and journey mapping — the "why now," not a generic persona

Instead of one broad ICP, mapped 5 ICP segments — with 4 prioritized for launch — by asking a sharper question than "who might want this": what specific moment in someone's life makes them go looking for something like this, right now? Each segment was built as a full Jobs-to-be-Done — the push that creates dissatisfaction, the pull toward this specific solution, the anxiety that could stop them, and the habit that keeps them stuck with the status quo.

Then mapped the full 5-stage journey — awareness through advocacy — separately for each segment, because the same funnel doesn't serve a professional who wants podcast-style passive content the same way it serves a retiree who needs reassurance and a human touch at signup. This is what surfaced that a €5 impulse-purchase visitor needs an entirely different checkout experience than a subscription visitor weighing it against existing tutoring spend — a distinction that shaped the checkout-flow build directly.

Phase 3

Turning strategy into a buildable funnel, and defining success before spending a euro

The original plan called for 5 segments running across 2 permanent funnel tracks — too much for a pre-launch product with one lesson and one instructor to actually build and maintain. The alternative built instead: 3 funnel variants tied explicitly to content maturity — direct-to-lesson now, lead-magnet once the library grows, subscription-first once there's real brand trust — with the leanest one recommended for launch and the other two deliberately deferred, not abandoned.

While designing that funnel, a subtler risk surfaced: a single song only resonates with the slice of any segment that already likes that genre or artist. Running one "hero lesson" against a whole broad segment would look, in the data, like a pricing or positioning failure — when it would actually be a genre mismatch invisible in the metrics. Left uncaught, that confusion would likely have pointed any fix at price or messaging instead of the real lever: genre diversity in the content batch, and narrower per-song ad targeting.

Also defined the exact KPI to watch at every funnel stage before the first ad ran — CTR by segment/creative, page-to-checkout-start rate, checkout completion rate, month-1 churn by segment — specifically so that when something underperformed later, there was already a framework to isolate where it broke, instead of debating it after the fact with no pre-agreed baseline.

Phase 4

The business model correction — the most strategic call in the whole engagement

Before any paid channel was scaled, the pricing and packaging work surfaced a hard structural fact: a single €5, one-time course could not realistically support paid acquisition under the CPC and conversion scenarios available at the time. This wasn't a conclusion reached after a failed campaign — it was flagged from the earliest pricing analysis, before spend scaled up.

The response wasn't "charge more" in the abstract. It was a specific packaging redirection: benchmark subscription pricing against a direct competitor's proven $15–20/mo willingness-to-pay, add bundled multi-lesson packages to raise average order value, and build a community layer specifically because in a subscription model, lifetime value is a function of retention — not top-of-funnel volume.

Phase 5

A new site and landing-page structure, built from the research — including the copy

Designed the site's full architecture from scratch on top of the work above, and caught structural gaps missing from the original page plan (a Pricing page, an FAQ, a Referral flow, on-site Search) — each tied back to a specific risk or JTBD finding rather than added arbitrarily.

Built a new, 9-section, conversion-ordered course landing page template, and the ordering itself was a deliberate decision: curriculum and instructor credibility were placed before pricing, to get the visitor to decide "this is worth it" before a number ever appears, with objections and genre-alternatives placed right before the final ask, since that's empirically where doubt surfaces in the journey mapped in Phase 2.

Then wrote full page copy for the first published lesson against that template — real, shippable creative, not a brief for someone else to write from: an instructor-voiced personal hook tied to a well-known song, a concrete "what you'll learn" breakdown instead of a vague promise, and a closing that returns to the same emotional hook that opened the page. Homepage copy was written the same way, translating the segmentation work directly into "who this is for" self-identification sections rather than one generic pitch.

Delivered a full visual brand system alongside it so the site, ads, and landing pages would all read as one product — and reverse-engineered the platform's real technical constraints (not assumptions) into a checkout flow built to minimize friction within what it genuinely allows.

One non-obvious judgment call worth naming: the homepage has to work simultaneously for a French-speaking learner who is exactly the "plateaued after years of classes" profile and an English-fluent professional who wants long-form cultural content — two audiences with different comfort levels in English, landing on the exact same page. That tension mattered more than it first looked: keyword research showed French-language search demand for this category running roughly 5x higher than English, and French speakers searching in French, not English — meaning the language decision wasn't just a UX nuance, it was a demand-sizing one. That was surfaced explicitly as a decision to make deliberately, before copy was finalized.

Phase 6

No customers, no testimonials — proving the method and the founder instead

A pre-revenue product with zero testimonials and zero usage data has a real credibility problem: every page in Phase 5 assumes some form of trust signal, and none existed yet. Fabricating testimonials or leaving the site looking empty were both rejected outright.

The solution was two-pronged. First, prove the method, not the product: the same peer-reviewed research from Phase 1 became a citable, on-page credibility asset — "backed by research showing songs help you learn faster than speech alone" is a defensible claim with zero customers required. Second, prove the person: instructor credentials, teaching background, and a real personal story were built into a dedicated About page and a condensed version on every landing page, since the founder himself is the entire trust signal a faceless competitor app can't offer.

Beyond that, built a 10-tactic social proof roadmap, explicitly sequenced by what's honestly available at each stage: a beta-tester program usable from day one with zero paying customers, running in parallel with teacher endorsements and press outreach; Trustpilot, video testimonials, and usage numbers deliberately held back for later, because a small, real usage number ("12 learners") would read as proof of low traction rather than proof of demand if shown too early. That's a judgment call that's easy to get backwards under pressure to "just show something."

Phase 7

Paid Search execution, under a hard hour cap

Ran the Google Search Ads campaign against this foundation: full EN and FR keyword research (1,000+ and 2,000+ terms respectively), filtered into core / reserve / wrong-intent / too-expensive / no-data buckets rather than launched wholesale, with ongoing negative-keyword management as real search-term data came in.

Directly applying the business-model math from Phase 4, imposed a hard CPC stop-loss on the Search campaign as a guardrail — bids weren't allowed to drift up to whatever the auction would pay, because the margin on a single lesson couldn't absorb it.

The campaign's average CTR landed around 12%, peaking at 14% in some ad groups — several times above typical Search benchmarks, and evidence that the keyword filtering and message-match discipline (ad copy echoing the exact landing page, not generic claims) was working. Generic, high learning-intent keyword groups ran especially hot — as high as 12.45% individually. That number alone would have been enough to justify scaling spend in most accounts. It wasn't used that way here — see Phase 8 for why.

Phase 8

The CAC reality check — and the fix, not just the rejection

Mid-engagement, budget was increased and a new weekly target set — €200 in spend producing 65 new subscriptions, implying roughly a €3 customer acquisition cost. The immediate response, communicated before any deeper analysis, was that raising budget alone doesn't fix unit economics — cheap, broad, untargeted traffic doesn't convert at a higher rate just because more of it is bought. That was the starting position; the model below is what backed it up.

Rather than pushing back on gut feel, the response was to build something falsifiable: separate "what we pay per click" from "what fraction of clicks become a paying subscriber," and build the CAC chain explicitly instead of arguing about the target as an abstraction. Rather than relying on one data source — which would have been easy to dismiss as one person's read on the market — triangulated three independent angles deliberately, so the finding would align expectations with market reality rather than read as a personal opinion: the account's own real Keyword Planner data, three separately-sourced industry benchmarks, and a self-built three-scenario forecast model against the product's own economics from Phase 4.

In money terms: hitting 65 subscriptions/week at the CAC the data actually supported would have taken far more than the €200/week budgeted for it. Scaling on the original assumption would have meant running every single-course sale at roughly −90% margin. Modeled against a 3:1 LTV:CAC floor and the product's real margins, the fix wasn't a bigger budget — it was leaning harder on the subscription/bundle/retention model from Phase 4, plus blending paid with organic so true cost-per-customer isn't carried by paid spend alone.

Phase 9

Auditing a separate paid campaign — from a weak signal to a specific, provable cause

A separate paid video campaign — launched outside the current scope of work — was underperforming. Rather than accepting the surface metrics (cheap clicks, some impressions) at face value, the audit worked backward from the critical signal — zero recorded conversions — and checked one layer at a time: audience settings, placement-level data, creative execution, and finally the analytics themselves.

That last check turned a routine audit into the most useful finding: cross-referencing the ad platform's reported clicks against the site's own independent analytics (GA4, PostHog) showed far fewer recorded visits than reported clicks. The discrepancy was consistent with a measurement-visibility problem, and the cookie-consent configuration emerged as a material candidate for investigation — analytics only starts recording after consent is given, which could mean a meaningful share of visitors were never being tracked at all.

Alongside it, the placement-level data was independently useful on its own terms: based on manual placement classification across 48 individually audited ad placements, 56% of spend landed on content clearly outside the intended audience — direct, specific evidence rather than a vague sense that something was off.

The deliverable was a sequenced fix plan (settings first, then targeting, then creative, then analytics/consent setup, then a controlled relaunch with a smaller test budget to isolate signal) — prioritized so the next test would actually be conclusive, not another ambiguous result.

What the Client Said

Public Upwork feedback, unedited

"Dimitri provided valuable research and strong marketing expertise during the early stages of our project. Ultimately, the combination of our small size, early stage, and limited budget proved incompatible with a long-term collaboration, but I appreciate what Dimitri did to get the first campaign off the ground and wish him all the best." — Upwork Client