Swift Innovation, Prepared for Cliff Smith
Hair Visualization Project:
Where I Landed
Where I Landed
The full brief from Charles to Cliff. Architecture, feature breakdown, cost model, compliance review, and competitive scan. Some of it is exciting and some of it is going to be annoying to read. You are getting both.
<45s
Four renders in under 45 seconds
$220K–$400K
Estimated range to launched product (Rev F)
5
Direct / near-direct competitors now in market
Day 0
Consent architecture must be built first
01
What I understood you to want
Technology for women of color, hair first. Something like a Snapchat filter, but instead of a novelty, a real preview of what she would look like in a given style or after using a given product. Maybe makeup later. Maybe outfits. The core idea: show her how good she could look before she spends four to eight hours and a few hundred dollars finding out.
If I got any of that wrong, stop reading and correct me, because everything downstream is built on it.
02
What I would actually build
"She uploads a selfie, picks a style, and gets four renders of herself in it within about 45 seconds. She compares them, saves the ones she likes, shares one to get opinions, and sends her favorite to a stylist as a consultation reference."
Two things make it different from the filters that already exist:
Differentiator 01: Face Lock
Her face is never regenerated
The system replaces only the hair region and puts her actual face back, pixel for pixel. Then it automatically checks that her skin tone did not shift, and throws the image away and retries if it did. Every generic AI hair app quietly lightens and "corrects" faces. Users notice. I have read the reviews. Women are saying, in plain language, that the app changed their face and the result did not look like them. Building the thing that does not do that is a real position.
Differentiator 02: Feasibility Layer
The app tells her whether she can actually get the style
Not just what it looks like, but whether it works on her hair as it is today, whether it needs extensions, whether it needs a few months of growth, or whether it needs processing and what that costs her hair. That is what a good stylist tells her in the first two minutes of a consultation, and as far as I can find, nobody has built it. This may matter more than the render itself.
03
Three things I would push back on
The Snapchat framing is wrong for this market, and that is good news
Live AR filters handle braids, locs, and twists worst, because those are three-dimensional and high-detail. Still-image generation renders them far better. So the right build is faster, cheaper, and better suited to exactly the styles that matter here. We give up "instant" and get "actually looks like braids."
Going shorter has to wait
Adding hair to a photo is straightforward. Going from waist-length braids to a tapered cut means the computer has to invent a hairline, forehead, and ears it cannot see, and hairlines are personal in a way that makes getting it wrong genuinely harmful. This is Phase 3, not launch. Which is painful, because the big chop is where this product would help the most.
The product half was under-built. I have corrected it.
You said style and product. The original spec handled product barely at all. Showing what a curl cream or a color actually does on her texture is easier to build than styles, more honest as a claim, and probably where the brand money is. Product-effect rendering (E-16) is now a first-class engine mode in Rev D.
04
The market reality
This space has funded competitors now
When we talked, I assumed nobody was serving this market. That was true a few years ago. It is not true today. Industry press called textured hair tech a gold rush, and a fifth entrant — GloVerse, backed by Grammy-nominated GloRilla — launched in October 2026.
Myavana
AI hair analysis for Black women, founded 2012. $5.9M raised led by Ulta's venture arm with Amazon. Integrated into Ulta e-commerce. 14-year head start on textured hair data.
Parfait
Selfie-driven wig customization. $5M seed from Upfront and Serena Williams' fund. Now licensing its AI to other hair brands, the same white-label path in this package.
Swivel Beauty
Already runs a marketplace matching women to stylists who specialize in natural textures, searchable by hair type.
HairHunt
General AI hairstyle app. Launched November 2025, reported 320,000+ users by April 2026. User reviews confirm face alteration at scale, the exact failure mode this package was designed against.
GloVerse
Celebrity-backed on-demand beauty marketplace, launched beta October 2026 (Grammy-nominated GloRilla). Supply-first GTM: onboarding beauty professionals now before consumer launch. Not a visualization competitor — but actively onboarding the stylist supply that Avalon's Phase 3 marketplace depends on. Integration candidate more than threat: Avalon renders, GloVerse books.
Here is the fair read. Nobody combines all of it. HairHunt has the visualization but changes faces. Myavana has the analysis and retail relationships but not visualization. Swivel has the stylists. Nobody has the feasibility piece at all. The position is not "we found an empty market." It is "the pieces exist and nothing is joined up." That is a normal and workable place to start from. It requires better execution rather than just being first.
05
The legal problem that is completely manageable and completely non-optional
Biometric Privacy Liability: Real Cases, Real Settlements
Charlotte Tilbury
$2.925M
Virtual try-on settlement
Kenvue (Neutrogena)
$4.7M
Facial-geometry skin tool
MAC Cosmetics
Proceeding
Case allowed to proceed June 2026
$1,000–$5,000 per person, no requirement that anyone was harmed. A pre-revenue MVP with 1,000 Illinois users is a theoretical seven-figure liability if the consent flow is wrong. Almost every one of those companies got caught for the same thing: a face-scanning tool with no proper consent screen and no published retention policy.
"Doing it right costs about a day of engineering and a conversation with a lawyer, at the beginning. There is no version where we add it in a later sprint."
06
The speed advantage
Swift builds with agentic development, which means a different relationship between time and scope than a traditional agency or internal team. The table below is the framework we use across all engagements. These are not marketing numbers. They are the actual ratios from our build log.
Size
Scope
Human hrs (mid)
Compression
Agent-directed
All-in
S
One surface, no schema change, known pattern
10
6x
0.7–2.7 hrs
2.2 hrs
M
One feature area, minor schema, up to one internal integration
28
5x
3.2–8.0 hrs
7.1 hrs
L
Multiple surfaces, schema migration, one external integration
70
4x
10–25 hrs
22.5 hrs
XL
New subsystem, multiple tables, multiple integrations, spike required
150
3x
33–67 hrs
70 hrs
Avalon MVP scope (Feature Matrix Rev F)
Mix: 10S + 38M + 10L + 0XL = 58 features
Human-equivalent hours (upper bound)~2,696 hrs
All-in agent-directed hours~517 hrs
Effective compression~5.2x
Sequential builds stack design, engineering, QA, and deployment. Agentic builds fan these out in parallel. That compression is what makes the MVP timelines below possible.
07
Investment
Phase 0
Consent + Legal
2 weeks
Phase 1 — MVP 0.0
Engine Technology Gate
2–3 months · $40K–60K
Phase 2
Consumer App
3–5 months
Phase 3
Scale + Stylist Marketplace
3–4 months
Biometric consent approved
Phase 0 exit
Engine go / no-go
MVP 0.0 gate
Beta launch
Phase 2 complete
White-label licensing
Phase 3 complete
Starting point: MVP 0.0
MVP 0.0 / Phase 0 + Phase 1 / Technology Gate
Engineering budget
$40K–60K
Timeline
3 months max
Features
26 of 57
MVP 0.0 is the first engagement: consent infrastructure (Phase 0) plus the full AI visualization engine (Phase 1). It answers the one question that must be answered before committing to the Phase 2 consumer app: does the engine produce results a stylist would actually use? If yes, Phase 2 proceeds. If not, the pivot happens before the larger budget is committed.
Achievable at: Tier 2 ($20K/mo → $60K over 3 months) or Tier 3 ($30K/mo → $60K over 2 months). Tier 1 reaches MVP 0.0 in 4-6 months at $40-60K.
Full MVP and beyond
MVP 0.0 ($40K–$60K, Phase 0 + Phase 1) is the fixed starting engagement and technology gate. After MVP 0.0 clears, three flat monthly retainer options for Phase 2 and Phase 3:
Tier 1
Steady Build
$10,000/mo
Engineering retainer
15 dev hrs/wk + 5 PM & QA hrs/wk
MVP 0.0 / Phases 0-1 (4-6 mo)$40K–$60K
Phase 2 (5 mo)$50,000
Phase 3 (4 mo)$40,000
Engineering total$130K–$150K
Tier 2, Recommended
Accelerated
$20,000/mo
Engineering retainer
30 dev hrs/wk + 8 PM & QA hrs/wk
MVP 0.0 / Phases 0-1 (2-3 mo)$40K–$60K
Phase 2 (3 mo)$60,000
Phase 3 (2.5 mo)$50,000
Engineering total$150K–$170K
Tier 3
Full Sprint
$30,000/mo
Engineering retainer
48 dev hrs/wk + 12 PM & QA hrs/wk
MVP 0.0 / Phases 0-1 (1.5-2 mo)$40K–$60K
Phase 2 (2 mo)$60,000
Phase 3 (1.5 mo)$45,000
Engineering total$145K–$165K
Rev F note: Phase 3 expanded by approximately 1 month at each tier due to the acceleration of stylist marketplace features (ST-03/05/07) from Phase 4 and the addition of the GloVerse/Swivel booking integration and on-demand mobile service mode. These were Phase 4 in Rev D. The supply-chain risk from GloVerse's October 2026 launch makes earlier activation the better risk profile.
Non-engineering costs (required, separate from retainer)
These are not optional line items. They exist regardless of which tier you choose and regardless of whether Swift or someone else builds the software.
| Item | Low | High |
|---|---|---|
| Corpus production shoot | $57,000 | $129,000 |
| Legal: BIPA compliance + entity formation | $16,000 | $46,000 |
| Liability insurance | $5,000 | $25,000 |
| Compute and infra during build | $2,000 | $6,000 |
| Non-engineering total | $80,000 | $206,000 |
The corpus shoot is the largest single line item and the only asset a competitor cannot replicate quickly. A production-quality library across the full range of textured hair types and skin tones, properly released, is what turns this from a demo into a defensible product.
Combined totals (engineering + non-engineering)
Tier 1 (13-15 months to Phase 3)
$210K–$356K
$130K-$150K engineering + $80K–$206K non-eng
Tier 2 (8-9 months to Phase 3)
$230K–$376K
$150K-$170K engineering + $80K–$206K non-eng
Tier 3 (5-6 months to Phase 3)
$225K–$371K
$145K-$165K engineering + $80K–$206K non-eng
Post-launch infrastructure runs approximately $399/month (hosting, storage, communications) plus usage-based AI agent costs that scale with volume. This is the cost floor at launch, not the run rate at scale.
08
The business case
Market size
$8–9B
Global textured hair care market
2025–2026. North American share is approximately 40%, or $3.5B addressable from a US base.
~40M
Black women in the US
Change style every 4–8 weeks. Typical annual spend: $1,800–$3,200 per consumer, at $100–$400 per install.
Unit economics (supply side, who pays at launch)
The stylist is the first paying customer, not the consumer. The corpus contribution program is simultaneously the acquisition channel and the distribution channel, which is why CAC is near zero.
$0–$30
Stylist CAC
$1,470
Stylist LTV (30-mo avg)
>49:1
LTV:CAC ratio
At $49/month average and 30-month average retention, the math is not subtle. The question is not whether the unit economics work. It is whether we can activate enough stylists to reach the supply threshold that makes the consumer product compelling.
Break-even
175
Stylists to break-even at $29–$49/month average
Post-launch operating costs run approximately $2,000–$5,000 per month. At $29–$49 average per stylist, break-even is 175 subscribers. 175 stylists is 0.09% of independent stylists in the US. This is not a volume problem. It is an activation problem, which is a different kind of problem and a solvable one.
Revenue model (four lines, sequenced)
Do not try to turn on all four at once. Each is a different relationship and a different trust level with a different audience.
- 1Stylist subscription, $29–$79/month. Fastest path to revenue. Value is legible to the stylist: she gets leads, her portfolio looks professional, and she has a reference tool for consultations. Charge for the lead.
- 2Consumer subscription, $6–$10/month. Cost recovery and demand signaling, not the business. It tells us which styles and textures consumers actually search for, which shapes the catalog. Do not treat this as the primary revenue line.
- 3Product affiliate. Attach at the moment of intent, when she has seen the style and the feasibility layer is recommending products to support it. The constraint is keeping the recommendation honest. The moment the affiliate relationship is visible in the ranking, the recommendation loses its value.
- 4Aggregate demand data, Phase 4+. Highest margin. Near-zero delivery cost. Brands currently have no clean source of data on what textured hair consumers are actually searching for at the intent moment. We will. This is the long bet, not the launch bet.
Partnership and exit framing
Before deciding whether to build a company, decide whether you want to own one long-term. Both of the following are realistic outcomes within three to five years, and knowing which one you are aiming for changes some early decisions.
Myavana as acquirer
14 years of textured hair data, Ulta-integrated, retail relationships in place. What Myavana lacks is visualization and stylist supply. Once Avalon has both, it fills the exact gap in their stack. An acquisition is not a stretch outcome. It is the obvious outcome if the product lands.
Swivel as distribution partner
Swivel already runs stylist matching for natural textures. Avalon feeds their booking funnel rather than competing with it. A partnership that gives Swivel's stylists the visualization tool while Avalon gets their network is worth modeling before building the marketplace in-house.
Neither of these require you to design for an exit from day one. But if you do not want to run a company forever, one of these is the path, and the early decisions that differ are entity structure and data ownership, which are exactly the questions in section 09.
09
The infrastructure
Avalon runs on the Avalon Brain - a seven-layer cognitive substrate that routes every platform component through a single hub. The visualization engine, the stylist corpus, the intake flow, and the consumer app all connect through it. There are no direct component-to-component connections outside the Brain. This is what makes the corpus self-reinforcing: every stylist contribution sharpens recommendations across the whole platform before it reaches any user.
Platform hub architecture
Core platform
Stylist corpus (the moat)
Client-facing layers
AI inference
Brain substrate layers
The Brain is seven stacked layers. What makes Avalon a T3 tenant is that all seven are provisioned at full capacity, tuned to the hair domain, and running on Cliff's hardware - not shared cloud infrastructure.
Four additional layers (context compression, memory, vector store) are documented in full in the Brain Infrastructure strategy doc.
What the T3 tier delivers
T3 is not a software license. It is a deployment of the full Brain stack - all seven layers - configured for the hair domain, installed on Cliff's hardware, and handed over as an owned asset. This means data sovereignty: client photos, consultation history, and stylist contributions never leave Cliff's infrastructure.
You get what we run in production - installed on your hardware, owned permanently. As the platform improves, upgrades are your choice on your timeline. You are never starting from scratch.