MyDressr — Team Leader Strategy Dashboard

$1.81

Cost / Free user / mo (today)

Loss-making

$0.40

Cost / Free user / mo (new)

−78% vs today

$3,055

Net monthly burn at current MAU (today)

AI > Revenue

+$0

Net monthly burn at current MAU (new)

Path to profit

The 3 problems I see as team leader

Every tier loses money. Plus user costs $22.20/mo at Gemini 3 Flash Image but pays $9.99 → −$12.21 net. Golden costs $52.48 vs $29.99 price → −$22.49 net. This breaks before launch.
Single-vendor lock-in on Gemini. 5 of 8 cost drivers (try-on, daily routine, what-if, scan-and-add, chatbot) all flow through one Google API. Pricing change or geo block stops the product.
Six of eleven cost drivers in the model are forward-looking — daily routine, chatbot, Veo videos, affiliate checkout aren't live. The MVP is narrower than the pitch deck suggests, so we still have time to fix the cost basis before users arrive.

The 3 moves that fix it

Hybrid AI routing. Gemini Pro only for the hard hero shots. Cheaper engines (FASHN $0.04, Imagen 4 $0.04, Claude Haiku $1/$5, GPT-4o-mini, self-hosted IDM-VTON at $0.01) for the 80% of routine calls. Cuts variable cost ~60–75%.
Caching layer for try-ons. Same avatar × same garment = same image. Redis-backed cache at 50% hit rate is realistic and already half-implemented. Saves another ~30%.
Resolution tiers tied to subscription. Free = 512px previews ($0.045 / image). Plus = 1024px ($0.067). Golden = 2048px hero shots ($0.10). Aligns cost to willingness-to-pay.

Cost per user — today vs new approach (live at your MAU + paid %)

Today's bars use the prices already in your Excel model. The "new approach" bars apply hybrid routing, 50% cache, and 0.5K free-tier resolution.

$8.27B

Global virtual fitting room market (2026) → $30.4B by 2034 at 17.7% CAGR

$1.2B

AI digital wardrobe market (2024) → $5.53B by 2033 at 18.5% CAGR

3 of 4

Top MENA fashion e-commerce markets sit in our launch zone (KSA, UAE, Egypt)

Why GCC + Egypt first is the right call

Smartphone penetration: UAE 99%, KSA 97%, Qatar 95%, Egypt rising. Mobile-first product fits perfectly.
Fashion e-commerce maturity: Namshi, Ounass, Sivvi, Noon all profitable — users are used to paying for fashion online, but nobody has cracked AI styling natively in Arabic.
Modest fashion gap: Abaya, jalabiya, hijab styling, occasion dressing for weddings/Eid/Ramadan — none of the global wardrobe apps handle these well. This is our moat.
High disposable income on fashion: GCC household spend on apparel is 2–3× global average; Egypt has the volume (110M people, fashion-conscious younger demographic).
Government tailwinds: Saudi Vision 2030 and UAE AI Strategy 2031 favor local AI products. Falcon-H1 Arabic (TII Abu Dhabi, Jan 2026) and Jais 2 70B (G42 Inception) are sovereign Arabic LLMs we can use to differentiate.

Direct + indirect competitors

App	Region	Free	Paid	Try-on?
Whering	UK / Global	Yes	Optional supporter	Outfit remix only
Acloset	Korea / Global	100 items	$3.99 / $9.99	Color analysis
Cladwell	US	1 outfit/day	$7.99 / $49 (human)	No
Indyx / Stylebook	US	Trial	$3.99 one-time	No
Glance AI	Global	Limited	Subscription	Yes (image-based)
Namshi / Ounass	GCC	Free	Commerce	No AI styling
Modanisa	Turkey / MENA	Free	Commerce	No AI styling
MyDressr	GCC + Egypt	20 try-ons	$9.99 / $29.99	Yes — full avatar VTO + trips

No competitor combines avatar-based try-on, trip planning, daily outfit AI, and Arabic-native styling for modest fashion. That's the wedge.

Strong & weak points — honest read

Based on the May 2026 code audit (90 fixes shipped, 17 client-reported issues) and the AI cost model in your Desktop folder.

STRENGTHS

Avatar pipeline is technically advanced — 2-step Gemini (text-only body → face injection), BMI-aware body shape, OpenCV face crop, GFPGAN sharpening, base-avatar cache with MD5 key. None of the competitors have this depth.
Trip-based outfit planning with weather + calendar — unique angle vs Whering/Acloset.
Stripe + 3 tiers + admin panel + 136 Swagger endpoints — backend is production-shaped.
Real performance work done — banner 44MB → 1.4MB (96% smaller), Redis cache layer, lazy loading, no jQuery dupes.
Apple Sign-In with proper JWT/JWKS verification — most MENA apps still skip this.

WEAKNESSES

Unit economics broken at every tier. $1.81 cost per free user, Plus loses $12, Golden loses $22. Not launchable as-is.
Heavy single-vendor dependency on Gemini (5/8 cost drivers). Geo-restricted in some GCC corridors — needed proxy workaround.
No Arabic AI yet. ChatSession + ChatMessage tables exist but no Gemini chat wiring, no dialect handling.
Daily routine / What-If outfit engine + Veo video + affiliate checkout are still on the roadmap — pitch promises more than MVP delivers.
Subscription billing not connected to Stripe (still test keys); push notifications need FCM key; forgot-password depends on test SMTP.
No Hijab / abaya / jalabiya / modest-fashion specific training in any of the AI models we use.

OPPORTUNITIES

Modest-fashion AI styling: untouched category, big GCC TAM. Could be the "Whering for the GCC."
Sovereign Arabic LLMs (Falcon-H1, Jais 2) are free for non-commercial and cheap to fine-tune. Marketing angle: "made for Arabic."
Affiliate revenue layer — Namshi, Sivvi, Modanisa, SHEIN all have affiliate programs paying 5–12%. Already modeled in our DB.
B2B white-label for retailers (Ounass, Namshi). Same VTO engine, their catalog, their app.
Ramadan/Eid/Wedding-season "occasion stylist" — recurring revenue spike windows.

THREATS

Google could raise Gemini prices or rate-limit free-tier API key. We have no fallback today.
Namshi/Noon/Ounass could bolt AI try-on onto their existing user base in 6 months and out-distribute us.
SHEIN/Temu are testing generative VTO globally — if they launch in MENA, free users won't pay for ours.
Regulatory: KSA PDPL + UAE PDPL — face/body photos = biometric data, needs explicit consent + local storage. Not yet implemented.
If launch is delayed past Ramadan 2027, we miss the highest fashion-spend window of the year.

NEW features I'd push as team leader

Feature	Impact	Cost	Effort
Arabic AI Stylist (Jais 2 / Falcon-H1) Dialect-aware chat: Khaleeji, Egyptian, MSA	High	Low	2–3 wks
Occasion Mode: Wedding / Eid / Ramadan / Office Curated outfit packs tied to GCC calendar	High	Low	2 wks
Modest Fashion Library Abaya, hijab styling guide, jalabiya, sportswear-modest	High	Med	3 wks
Affiliate "Shop the Look" with Namshi + Sivvi + SHEIN Revenue stream that doesn't hit subscribers	High	Low	4 wks
Group / Family wardrobes Style siblings, kids, partners in one account	Med	Low	2 wks
WhatsApp bot for outfit-of-the-day GCC's #1 messaging channel, removes friction	High	Med	3 wks
Virtual closet sharing → social feed Friends like / save / remix each other's looks	Med	Low	3 wks
"Will it fit my body" sizing AI BMI + body shape × brand size charts	Med	Med	4 wks

MVP improvements (the cleanups)

Improvement	Why	Effort
Switch try-on default to FASHN volume tier ($0.04 vs $0.067)	~40% saving, same UX	1 wk
Free tier images at 512px not 1024px	33% saving, mobile screens don't need 1K	2 days
Aggressive Redis cache on avatar×garment	~50% hit rate easily; cache key already exists for base avatars	1 wk
Hybrid LLM router: GPT-4o-mini for chat, Gemini Pro only for hero	$0.75/$4.50 vs $1.25/$10 per 1M tokens	1 wk
Background tasks: ensure Celery worker + Beat actually run in prod	Issue #13 from Marianne — currently fails silently	1 day
Stripe Checkout flow + webhook (real keys)	Currently can't take a single payment	3 days
FCM push + SendGrid SMTP	Account recovery + retention	3 days
Add local consent flow for biometric (face) data	KSA / UAE PDPL compliance before launch	1 wk
Rate limit + abuse detection on AI endpoints	One bad actor could rack up $1K in Gemini in hours	3 days
App Store / Play Store ASO in Arabic	Half the addressable market searches in Arabic	1 wk

Cost per user — Now vs New Approach

"Now" = your existing Excel model with Gemini 3 Flash Image at 1024px and Gemini 2.5 Flash for all LLM. "New" = hybrid routing (Gemini Pro only for hero, GPT-4o-mini for chat, FASHN for VTO at volume, self-host IDM-VTON at scale), 50% caching, and 512px on free tier.

Per-user monthly cost (live)

Total monthly P&L at your slider settings

Breakdown table

Tier	Cost/user NOW	Cost/user NEW	Saving	Revenue	Net NOW	Net NEW

Cheaper / alternative AI options to Gemini

Your concern is the right one: Gemini quality is great, but cost climbs linearly with users. Below is the full menu — image generation, LLM, Arabic-specific, and self-hosted.

Image generation / Try-on per render

Model	$ / image	Use case
Gemini 3 Flash Image (1K) — today	$0.067	Default everywhere
Gemini 3 Flash Image (0.5K)	$0.045	Free tier mobile previews
FASHN AI (volume tier)	$0.04	Specialized VTO, recommended for try-on
Imagen 4	$0.04	Pure image gen, no VTO specialization
Gemini 2.5 Flash Image	$0.039	Older but still usable for non-hero shots
OpenAI gpt-image-1	$0.08	Fallback only — pricier
Self-hosted IDM-VTON on A100 (steady state)	$0.01	Only profitable above ~13K MAU
Replicate IDM-VTON (no self-host)	$0.025	Mid-scale bridge before self-hosting

LLM (chat, scan-and-add, daily routine) per 1M tokens

Model	$ in	$ out	Use case
Gemini 2.5 Flash-Lite	$0.10	$0.40	Classification — keep, cheapest
Gemini 2.5 Flash — today's chat	$0.30	$2.50	Standard chat
GPT-4o-mini	$0.75	$4.50	Better quality for similar money, easy A/B
Claude Haiku 4.5	$1.00	$5.00	Best reasoning at this tier; great Arabic
Gemini 2.5 Pro	$1.25	$10.00	Hero requests only
OpenAI GPT-5.4	$2.50	$10.00	Alternative flagship
Claude Sonnet 4.6	$3.00	$15.00	Premium fallback for hard reasoning

Arabic-specific options (the differentiator)

Model	From	License	Why it matters for MyDressr
Jais 2 (70B)	G42 Inception + MBZUAI (Abu Dhabi)	Open weights, commercial OK	Trained on 1.6T tokens with strong Arabic. Run on Azure or self-host. Marketing angle: "Built for the Gulf."
Falcon-H1 Arabic (3B/7B/34B)	TII Abu Dhabi (Jan 2026)	Open	Hybrid Mamba-Transformer, 3B is small enough to self-host cheaply.
AYA-23 / AYA-Expanse	Cohere for AI	Open	Strong on Arabic dialects including Egyptian and Khaleeji.
Claude Sonnet 4.6 (premium)	Anthropic	API	Best non-Arabic-native quality at Arabic; good fallback for nuanced styling advice.

Recommendation: use Falcon-H1 7B self-hosted for high-volume Arabic chat (near-zero variable cost above the GPU), Claude Haiku 4.5 as paid fallback for hard reasoning, Gemini Flash-Lite for classification.

Recommended routing (the "new approach")

Workload	Today	New	Saving
Avatar (hero)	Gemini 3 Flash Image 1K	Keep Gemini 3 Flash Image 1K	0% — keep quality
Try-on (volume)	Gemini 3 Flash Image 1K	FASHN AI volume tier + cache	~70%
Daily routine images	Gemini 3 Flash Image 1K	Gemini 2.5 Flash Image / Imagen 4	~42%
Free-tier previews	1024px	512px	33%
Chatbot (Arabic)	Gemini 2.5 Flash	Falcon-H1 7B self-host / Jais	~85%
Chatbot (English)	Gemini 2.5 Flash	GPT-4o-mini	~0% same tier, better quality
Scan-and-add	Gemini Flash-Lite	Keep Flash-Lite (already cheapest)	0%
Background removal	Gemini	Self-host rembg / U2Net	~100%

Build-your-own-model: when does it make sense?

Year-1 cost to build: ~$572K (one-time $140K + annual $432K) — 2 ML engineers, GPU cluster, 50K labeled garment pairs, MLOps stack.
Cost per image self-hosted: ~$0.01 vs Gemini $0.067 → savings $0.057/image.
Break-even: ~12,700 MAU (≈700K images / month).
Do NOT build below 50K MAU. Use Gemini + FASHN + cache. The own-model cost is wasted runway before you have enough volume to amortize.
Build at 50K → 200K MAU as a fine-tune on open VTO (IDM-VTON) using MENA-specific data (modest fashion, body diversity, skin tones).
Cutover at 200K MAU. Default try-on goes to your own model. Gemini stays as fallback for the long tail.

From 1K to 1M: the viral scaling plan

If the app pops — say a Saudi or Egyptian influencer puts us on TikTok — this is what each stage looks like and what we need ready before that happens.

Stage map

Stage	MAU	Focus	AI strategy	Infra	Team	Monthly AI cost (new)
0. Pilot	0–1K	Hardening, GCC closed beta	Pure Gemini + FASHN	1 EC2 t3.large + Redis + RDS MySQL	You + Marianne + 1 dev	$1.0K
1. Launch	1K–10K	App Store + ASO in Arabic; Ramadan window	+ Caching live, FASHN volume tier, GPT-4o-mini chat	+ CDN (CloudFront), S3, Sentry	+ Flutter dev, ASO/marketing	$10K
2. Growth	10K–50K	Influencer + affiliate (Namshi, Sivvi)	+ Arabic LLM (Falcon-H1 7B self-host)	+ ECS / Kubernetes, autoscale workers	+ ML/data engineer, support	$50K
3. Scale	50K–200K	UAE / Egypt expansion + B2B pilots	+ Fine-tune IDM-VTON on MENA data	+ Multi-region (Bahrain + Egypt edge), read replicas	+ Senior ML, DevOps, BD	$200K
4. Viral	200K–1M	MENA-wide rollout, Morocco / Jordan	+ Cutover to owned VTO model for default; Gemini = fallback	+ Owned GPU cluster (A100s) for inference	+ ML team of 3, CFO, country managers	$500K

Total monthly AI cost across MAU scale

Gross margin path (with the new approach)

Margin turns positive around 25K MAU once paid mix exceeds 12% and cache hit rate hits 50%.

Things to have ready before the viral moment

Cost controls in-app: daily AI budget per user, soft caps that show "you've used your daily 20 try-ons", queue-and-retry for hero shots.
Autoscaling Celery workers on AWS ECS or Fly.io so a 100× spike doesn't crash try-on queue.
Multi-key Gemini rotation (Google rate-limits per API key). 5–10 keys behind a router.
FASHN, OpenAI, Anthropic accounts pre-paid with $5K float each — vendor switch must be instant.
S3 + CloudFront for all media so MySQL doesn't get hammered by image URL generation.
App Store TestFlight + Play Console internal track live and tested 2 weeks before launch.
PDPL consent + face-image local-region storage (KSA region for Saudi users, Egypt region for Egyptian users).
Crisis communication plan — what we tell users if Gemini goes down for 4 hours.

The 90-day action plan I'd run as team leader

Three sprints, each two weeks of build + one week of test/review. Goal: get unit economics positive before public launch and ship Arabic + modest-fashion before Ramadan 2027.

Sprint 1 — Days 1–30

Fix the money leak

Add hybrid routing layer (Gemini Pro / FASHN / GPT-4o-mini)
Wire Redis cache for try-on (avatar × garment key)
Drop free tier to 512px
Stripe live keys + Checkout flow + webhook
FCM push + SendGrid SMTP wired
Rate limit on /api/avatar, /api/tryon, /api/chat
Per-user daily AI budget enforcement

Exit gate: Cost-per-Plus-user under $8/mo at 1K MAU.

Sprint 2 — Days 31–60

Build the MENA wedge

Arabic Stylist chat — Falcon-H1 7B self-hosted on a single A10 GPU
Modest Fashion library + tags (abaya, hijab, jalabiya, modest sport)
Occasion Mode: Wedding / Eid / Ramadan / Office presets
Affiliate links wired to Namshi + Sivvi + SHEIN
App Store / Play Store Arabic ASO + screenshots
KSA + UAE PDPL consent flow on signup + face capture

Exit gate: Closed beta with 200 GCC users, NPS > 30.

Sprint 3 — Days 61–90

Launch & learn

Public launch in KSA + UAE first, Egypt 2 weeks later
WhatsApp outfit-of-the-day bot (Twilio or Meta Cloud API)
Group / family wardrobes
Daily routine engine activated for Plus + Golden
Influencer campaign — 3 micro-influencers per market
Multi-key Gemini rotation + FASHN failover live in prod
Weekly cohort review of cost-per-MAU + retention

Exit gate: 5K paid MAU, monthly burn under $5K, ready to brief investors.

MyDressr Strategy Dashboard