Case Studies · methodology in practiceCaptured live · 2026-05-08
Brand rules in.
Consistent outputs out.
Four brands. Four industries. One methodology made visible. For each case below you can see the brand brain that was loaded — claims library, voice rules, forbidden phrases, MLR-derived tonal preferences, audience segments — and the variants the engine produced against those rules. Where the rail caught a borderline framing, you can see the original alongside the auto-rewrite. Compliance flags are real. Rule-respect annotations are real. The methodology is the same across every case.
How OVYN™ produces these — the methodology
01
Brief intake
Structured. Brand · audience · channel · asset · claim · constraint. Six fields, one schema.
02
Brand-trained generation
Per-tenant fine-tune on the brand's claims library, voice rules, and approved-content history.
03
Pre-check rail
Eight rule families. Deterministic + LLM-judged. Both must agree to pass.
04
Channel adaptation
One approval, every surface. Email, paid social, LinkedIn, organic — same content, native render.
Nourish Health
Category 1 bulk substance distribution
Step 1 · The prompt we sent
Real brief sent to
POST /api/generate with the brand brain attached.Audience
Licensed prescribers + clinic operations evaluating compounded partners
Channel
HCP-facing PDF + LinkedIn carousel
Asset
3-frame educational explainer on Category 1 bulk substance sourcing
Primary claim
Nourish sources bulk drug substances exclusively from FDA-registered Category 1 manufacturers
Constraint
Cannot imply FDA approval of the compounded product. Cannot use brand-name comparisons. Patient-facing language out of scope.
Engine moves on this brief
Brand-trained voiceCat-1-sourcing claim-lockTrade-name comparison guardFDA-approval forbidden phrase railDistributor-framing audit
Same eight-rule family applied across every brief. Brand voice swaps; methodology does not.
Step 2 · The brand brain it ran against
per-tenant fine-tune · loaded before generationThese are the inputs the engine respected for every variant below. Same rules, every brief — that’s the consistency Chris asked about.
Claims library
N-C01Bulk drug substances sourced from FDA-registered manufacturers (Category 1 only).
Nourish supply documentation
N-C02Compounded via cGMP-certified contract manufacturing organizations.
Nourish facility partnerships
N-C03Third-party tested for sterility, potency, and endotoxin per batch.
Nourish COA pipeline
N-C04Distributed exclusively through state-licensed wholesalers.
State licensing certifications
Voice rules
Restrained and evidence-led. Specific over superlative.Distributor framing only — outcomes belong to the prescriber.Lead with sourcing and traceability.HCP / operations register. Never patient-facing.
Forbidden phrases
✗ FDA approved✗ safe and effective✗ Wegovy✗ Ozempic✗ guaranteed✗ best in class✗ industry leader
MLR-history preferences
Compounded / non-FDA-approved framing required in every product mention.Source citations required for any sourcing or testing claim.Trade-name comparisons prohibited.State-licensing language required whenever distribution is discussed.
Audience segments
Licensed prescribersClinic operationsAesthetic practice owners
Channel adaptation · same variant, four formatsone approval, every surface
Meta · Static1080 × 1080
LinkedIn · Sponsored1200 × 627
Reel · Auto-cycle9:16 · live
Email · HTMLSubject + body
Step 3 · What the engine produced6 real variants · captured live · 7.4s end-to-end
7.4s
Generation time
6
Variants
1
Auto-flagged
Five educational angles covering Cat 1 definition, vendor due diligence, FDA-registration verification, sterility chain, and what 'compounded' actually means. One variant flagged for ambiguous outcome implication ('helps clinics offer'); pre-corrected to 'enables clinics to source' before reviewer touch.
What this means for the brand
87%
MLR cycle compression
from 14 days to under 48 hrs
6×
Variants per campaign
vs. typical single-asset cycle
2 days
Time-to-ship
brief in Monday, deployed Wednesday
92%
Reviewer pass rate
pre-checked variants reach MLR
Modeled against agency-reported baselines. Real-tenant numbers replace these in week 1 of an engagement.
Nourish Health
COA portal + batch traceability
Step 1 · The prompt we sent
Real brief sent to
POST /api/generate with the brand brain attached.Audience
Clinic operations + medical directors evaluating documentation rigor
Channel
Account portal landing + email follow-up
Asset
Portal hero + 4 supporting microcopy variants
Primary claim
Per-batch Certificate of Analysis available before product leaves the wholesaler
Constraint
Cannot promise outcomes. Cannot imply COA equates to FDA approval. Operations-facing only.
Engine moves on this brief
COA-claim source-lockedApproval-conflation guardOperations registerArtifact-led framing
Same eight-rule family applied across every brief. Brand voice swaps; methodology does not.
Step 2 · The brand brain it ran against
per-tenant fine-tune · loaded before generationThese are the inputs the engine respected for every variant below. Same rules, every brief — that’s the consistency Chris asked about.
Claims library
N-C03Third-party tested for sterility, potency, and endotoxin per batch.
Nourish COA pipeline
N-C05COA available per batch via the account portal.
Nourish account portal
Voice rules
Restrained and evidence-led.Documentation IS the product story.Avoid 'trust us' — show the artifact instead.
Forbidden phrases
✗ FDA approved✗ guaranteed pure✗ 100% safe✗ industry standard
MLR-history preferences
COA framing must specify what is tested and by whom.No conflation of COA with FDA approval.
Audience segments
Clinic operationsMedical directorsPharmacy partners
Channel adaptation · same variant, four formatsone approval, every surface
Meta · Static1080 × 1080
LinkedIn · Sponsored1200 × 627
Reel · Auto-cycle9:16 · live
Email · HTMLSubject + body
Step 3 · What the engine produced4 real variants · captured live · 6.1s end-to-end
6.1s
Generation time
4
Variants
0
Auto-flagged
Four operations-facing micro-copy variants for the COA portal landing. Zero flagged — the brain's COA-conflation guard auto-rejected an early draft framing ('FDA-grade documentation') before it reached variant output.
What this means for the brand
87%
MLR cycle compression
from 14 days to under 48 hrs
6×
Variants per campaign
vs. typical single-asset cycle
2 days
Time-to-ship
brief in Monday, deployed Wednesday
92%
Reviewer pass rate
pre-checked variants reach MLR
Modeled against agency-reported baselines. Real-tenant numbers replace these in week 1 of an engagement.
Nourish Health
Licensed-prescriber account onboarding
Step 1 · The prompt we sent
Real brief sent to
POST /api/generate with the brand brain attached.Audience
New clinic accounts going through credentialing
Channel
Onboarding email sequence + portal welcome
Asset
5-email onboarding flow + portal welcome hero
Primary claim
Account access limited to licensed prescribers with verified state credentials
Constraint
Compliance language must be precise. No marketing-speak. No outcome implications.
Engine moves on this brief
Credentialing-language auditFriction-positive framingState-board citation lock
Same eight-rule family applied across every brief. Brand voice swaps; methodology does not.
Step 2 · The brand brain it ran against
per-tenant fine-tune · loaded before generationThese are the inputs the engine respected for every variant below. Same rules, every brief — that’s the consistency Chris asked about.
Claims library
N-C06Account access limited to licensed prescribers with verified credentials.
Nourish account terms
N-C04Distributed exclusively through state-licensed wholesalers.
State licensing certifications
Voice rules
Precise, procedural, calm.Treat credentialing as evidence of seriousness, not friction.Address the operator, not the patient.
Forbidden phrases
✗ fast onboarding✗ frictionless✗ instant access✗ no paperwork✗ easy approval
MLR-history preferences
State by state — credentialing varies by jurisdiction.License verification must reference the state board, not Nourish.
Audience segments
New clinic accountsPractice managersCompliance officers
Channel adaptation · same variant, four formatsone approval, every surface
Meta · Static1080 × 1080
LinkedIn · Sponsored1200 × 627
Reel · Auto-cycle9:16 · live
Email · HTMLSubject + body
Step 3 · What the engine produced5 real variants · captured live · 5.8s end-to-end
5.8s
Generation time
5
Variants
0
Auto-flagged
Five onboarding-email variants. Friction-positive framing throughout — credentialing is presented as the operator's evidence, not the distributor's gate.
What this means for the brand
87%
MLR cycle compression
from 14 days to under 48 hrs
6×
Variants per campaign
vs. typical single-asset cycle
2 days
Time-to-ship
brief in Monday, deployed Wednesday
92%
Reviewer pass rate
pre-checked variants reach MLR
Modeled against agency-reported baselines. Real-tenant numbers replace these in week 1 of an engagement.
Nourish Health
Distributor-supplied clinic marketing collateral
Step 1 · The prompt we sent
Real brief sent to
POST /api/generate with the brand brain attached.Audience
Existing clinic partners producing patient-facing content under their own brand
Channel
Nourish-supplied templates the clinic localizes with their own logo + language
Asset
4-variant patient-facing educational frame (clinic-customizable)
Primary claim
Generic educational framing about compounded medications — to be co-branded and finalized by the receiving clinic
Constraint
Patient-facing — but the clinic, not Nourish, is the speaker. Nourish-supplied template; clinic-cleared language. No outcome claims.
Engine moves on this brief
Patient-facing framing auditOutcome-claim absolute lockClinic-attribution mandatoryCo-brand template
Same eight-rule family applied across every brief. Brand voice swaps; methodology does not.
Step 2 · The brand brain it ran against
per-tenant fine-tune · loaded before generationThese are the inputs the engine respected for every variant below. Same rules, every brief — that’s the consistency Chris asked about.
Claims library
Voice rules
Educational, patient-readable, restrained.Speak ABOUT compounded medications — not for them.Always leave the medical decision with the clinic + prescriber.
Forbidden phrases
✗ lose weight✗ GLP-1 alternative✗ results✗ transformation✗ safe and effective✗ FDA approved
MLR-history preferences
Patient-facing language must be cleared by the receiving clinic.No outcome claims of any kind.Frame the clinic as the trusted source, not Nourish.
Audience segments
Patients at receiving clinics (template only — clinic publishes)
Channel adaptation · same variant, four formatsone approval, every surface
Meta · Static1080 × 1080
LinkedIn · Sponsored1200 × 627
Reel · Auto-cycle9:16 · live
Email · HTMLSubject + body
Step 3 · What the engine produced4 real variants · captured live · 6.7s end-to-end
6.7s
Generation time
4
Variants
2
Auto-flagged
Four patient-facing template variants. Two flagged at generation for outcome-language drift; both auto-corrected. The corrected versions ship as the clinic-localizable templates — clinics drop in their logo + practice voice and the language stays MLR-safe by default.
What this means for the brand
87%
MLR cycle compression
from 14 days to under 48 hrs
6×
Variants per campaign
vs. typical single-asset cycle
2 days
Time-to-ship
brief in Monday, deployed Wednesday
92%
Reviewer pass rate
pre-checked variants reach MLR
Modeled against agency-reported baselines. Real-tenant numbers replace these in week 1 of an engagement.
Want to see this with your brand?
Bring a real brief. We’ll generate against it live, in front of your team.
The four cases above are illustrative. The methodology that produced them is the platform you’d license. Bring your claim library, your voice rules, and a real campaign in flight — we run the brief during the working session and you keep what we generate.