Mladen Grozev

2 Shopify apps exits | Co-founder @gethookdai (7-figure ARR) | Talking about building SaaS products and AI

[Mladen Grozev @MladenGrozev]

From a developer to a founder of an eCommerce development agency.

From zero to SIX figures per year in revenue with no upfront investment in my first business ✨

Here are ten lessons I wish I knew when I started.

>> Thread <<

[Mladen Grozev @MladenGrozev]

Most ad-policy checks start after Meta rejects the creative.

Run this preflight before upload:

1. Pull the current Advertising Standards from Meta’s Transparency Center.
2. Map each line of copy to the policies that apply.
3. Put the risky phrase beside Meta’s exact policy text.
4. Rewrite the line without changing the selling intent.
5. Give the ad one verdict: Cleared for launch, Fix before launch, or Grounded.

Run it across the batch and hand the report to the media buyer.

The media buyer should see the risky line, the policy citation, and the proposed fix in one view.

[Mladen Grozev @MladenGrozev]

Stickman ads remove several failure points from AI video.

No faces to keep consistent.
No wardrobe.
No character design.
The model only needs to preserve simple motion and action.

A workable script structure:

1. Give the character one clear problem.
2. Turn the problem into a visible villain.
3. Use simple, expressive movement.
4. Delay the resolution so the viewer has a reason to watch the ending.

The source tested the format across health, ecom, legal and finance. That is enough to justify a test, not a universal claim.

[Mladen Grozev @MladenGrozev]

@MichLieben 6750

[Mladen Grozev @MladenGrozev]

There is no useful fixed ratio between new angles and iterations. Use two phases.

Phase 1: test three angles in one batch with the same format and offer.

A humidifier test compared stainless steel, easiest to clean and doctor designed. Stainless steel won. Doctor designed kept losing.

Phase 2: hold the angle and vary hooks, formats and images.

New imagery kept ROAS steady while spend rose 80%. A larger shift to AI voiceover storytelling for parents increased spend another 4.4x at the same ROAS.

Use the first phase to find direction. Use the second to test the depth of the winning belief.

[Mladen Grozev @MladenGrozev]

Once an account has 50 to 100+ active ads, creative review needs a fixed schema.

Label every asset by:

• asset type
• messaging angle
• hook tactic
• funnel stage

Then calculate win rate by category, split kill and scale calls by TOF, MOF and BOF, and rank iteration ideas by the spend behind the source ad.

The weekly review should end with ranked iteration priorities from ads with real spend behind them.

[Mladen Grozev @MladenGrozev]

Turning a winning TikTok into a usable ad brief takes more than saving the link.

The workflow:

• Search by keyword and date range.
• Pull engagement, captions and thumbnails.
• Have a model break down the hook.
• Mine comments for buyer questions and angles.
• Generate the brief from your own template and brand bible.

The final brief should specify the hook, shots, captions and the buyer objection it came from.

If the creative team still has to reverse-engineer the video, the research step is unfinished.

[Mladen Grozev @MladenGrozev]

@KodyNordquist An 80% same-day CVR drop across brands points to a shared failure mode.

I’d segment by placement and device, check landing-page events against backend orders, then cut spend after the checks locate the break.

[Mladen Grozev @MladenGrozev]

@noctadn Stickmen may win because viewers parse the scene fast and focus on the claim. I’d test the same script across stickman and UGC, then compare hold rate and purchase CVR to isolate format.

[Mladen Grozev @MladenGrozev]

@maverickecom At $2 per asset, teams spend more on review and media than production. Tag each ad by hook and audience, or your team cannot tell which choice drove the result.

[Mladen Grozev @MladenGrozev]

@mikefutia A reproducible brief gives this workflow leverage. Lock the audience and visual rules before generation, then tag each asset with its hypothesis and result so Claude can build the next batch from evidence.

[Mladen Grozev @MladenGrozev]

@Ads_gamma A signup gap can come from message mismatch, page speed, form friction, or bad traffic. I split the funnel by event rate and segment, then change the owner tied to the first broken step.

[Mladen Grozev @MladenGrozev]

@GibbyBerryhill Before adding another dashboard, define contribution margin per order from landed COGS, fulfillment, payment fees, discounts and returns.

Then reconcile it to the books weekly. A prettier ROAS view won’t fix bad cost inputs.

[Mladen Grozev @MladenGrozev]

@arceyul Persistent brand memory is the interesting part. The guardrail is versioning it: approved claims, banned phrases and current visual rules need a dated source of truth, otherwise the agent scales yesterday’s mistakes very efficiently.

[Mladen Grozev @MladenGrozev]

@mikefutia 40 outputs is useful only if they span distinct hypotheses. I’d force variation across angle, hook, format and awareness level, then dedupe visually before launch or you’re mostly buying the illusion of testing volume.

[Mladen Grozev @MladenGrozev]

A useful content research workflow:

1. Enter a product, niche or competitor.
2. Pull Reddit, YouTube, TikTok and Reels.
3. Rank posts by real engagement.
4. Extract hooks, pain points and customer wording.
5. Turn the result into three shoot-ready ideas.

On “magnesium for sleep,” the source workflow found a 5.1M-view hook and a repeated complaint: “it works for a few weeks, then stops.”

Keep the source, view count and customer quote attached to every brief. Otherwise the research cannot be audited.

[Mladen Grozev @MladenGrozev]

Offer-focused image ads need to answer three questions in seconds:

What is the product?
What is the offer?
Why should I want it?

A practical layout: product centered, offer in a high-contrast badge, then 3-6 benefits. Give Google multiple images, headlines and descriptions to test.

[Mladen Grozev @MladenGrozev]

A useful low-cost ad workflow from this week:

1. Write the script in Claude.
2. Build the voice in ElevenLabs.
3. Generate the character and clips in Higgsfield or G Flow.
4. Assemble and trim in CapCut.

The source produced a stick-figure ad in 15 minutes for about $0.50.

[Mladen Grozev @MladenGrozev]

Raw view count is a weak research filter when one creator has 10M followers and another has 10k.

Track creators, hashtags, keywords and niches in one feed. Rank each video against that creator’s normal baseline.

Then tag the hook, angle, format and proof before saving it by theme.

[Mladen Grozev @MladenGrozev]

A clean voice-of-customer workflow:

1. Paste a Shopify product URL.
2. Pull 50, 100 or 500+ reviews.
3. Export rating, body, author and date to CSV.
4. Ask an LLM what customers love and hate.
5. Turn recurring language into ad angles, hooks and headlines.

[Mladen Grozev @MladenGrozev]

One AI UGC workflow took a single script and one face, then produced a dozen versions across languages, tones and backgrounds.

That is the useful production shift: hold the message constant while changing delivery variables.

Review the variants as a controlled batch, not twelve unrelated ads.