Caldris
Managed removal of unauthorised AI likenesses, fake endorsements and impersonation campaigns — for public figures and the people who represent them.
Key outcome
Live and selling a promise. Positioning, tiers and a consultation funnel are public; the monitoring and enforcement operation behind them is not built yet, and caldris.live is still being connected.
02 / The problem
Generative tools made impersonation cheap, fast and convincing. Public figures find synthetic video ads, cloned voices and fake endorsements running under their name, and have no reliable way to find them, prove them, or get them removed and kept down.
03 / The hypothesis
“If a discreet, human-verified service can show a representative exactly what is running against their client and get it removed with evidence that holds, they will pay a monthly retainer for ongoing protection rather than a one-off takedown fee.”
04 / The solution
- →A marketing site with six content pages (platform, how it works, services, industries, legal enforcement, for agencies) plus about, privacy, terms and a sign-in page.
- →A six-step process sold as the product: detect → verify → preserve → enforce → escalate → report, with two explicit rules — AI confidence is never treated as proof, and a submitted notice is never counted as a removal.
- →Coverage stated as Meta platforms, TikTok, YouTube, X, public websites and landing pages, search results and supported advertising libraries, with further channels added on request.
- →Three retainers: Essential €1,000/mo (up to 15 managed cases), Executive €3,500/mo (up to 60, dedicated account manager, agreed SLA), Global from €10,000/mo (up to 200, custom countries and languages). Legal Enforcement is a separate add-on on custom quote, with external legal fees and court fees explicitly excluded.
- →A confidential consultation request covering who you represent, client category, threat type, urgency, region, individuals needing protection and an optional incident URL.
05 / Build components
14 routes: home, platform, how-it-works, services, industries, legal-enforcement, for-agencies, consultation, about, sign-in, privacy, terms.
Tiered retainers with monthly case caps (15 / 60 / 200) so delivery effort scales with price.
Client-side validation only (email format, URL format, consent tick) then a 'Request received' screen. No server function, no inbox, no database — submissions currently go nowhere.
Sign-in page states the portal 'is not yet available'; access by invitation only.
Described on-site (timestamped capture, SHA-256 content hashes, secure case records) — not built.
Described on-site as analyst-verified across the listed channels — not built.
Brand name, key copy and tier pricing live in src/config, so a rebrand or repricing is a one-file change.
Support email is still a placeholder: confidential@example.com.
06 / AI & LLM stack
- Build platform
- Lovable (TanStack Start)
- LLM(s)
- [Add model]
- AI use case
- None in product today — the offer is deliberately human-verified, and AI confidence is excluded from proof.
- Automation
- None yet — channel monitoring and takedown submission would be the first.
- APIs / integrations
- None — no payments, email, analytics or database connected.
- Hosting / infrastructure
- Lovable hosting; Lovable Cloud not enabled
07 / Distribution & GTM
- →Direct: talent managers, agents and business managers who already own a client's brand deals.
- →Agencies: a dedicated /for-agencies page pitching protection across a whole roster.
- →Legal: law firms and family offices that need enforcement capacity rather than advice.
- →[Add outbound tests, if any]
08 / KPIs & results
Visitors
TBC
Signups
TBC
Paid conversions
TBC
Revenue
TBC
CAC
TBC
LTV
TBC
Conversion rate
TBC
Gross margin
TBC
Payback period
TBC
Retention / repeat
TBC
Consultation requests
TBC
Values marked TBC are pending real data.
09 / Scale-or-kill
Undecided.
Too early to judge demand. The page is live but has no traffic source, no analytics and no working lead capture, so there is nothing to measure against the hypothesis yet.
What worked
- +The site reads as credible and discreet — no deepfake fear-mongering, and the copy makes a point of not overclaiming.
- +Retainer pricing sits high enough to cover human labour, which most 'AI detection' tools get wrong.
What didn’t
- −The funnel leaks: a consultation request is never delivered to a human.
- −Nothing behind the promise exists yet, so every claim is a manual service sold before it can be delivered.
Next time I’d test
- →Wire the consultation form to a real inbox and a calendar link before writing another word of copy.
- →Deliver the first three cases manually and free, to learn what a real takedown actually costs in hours.
10 / Total cost & credits
- Lovable credits
- TBC
- AI / LLM credits
- TBC
- Hosting
- TBC
- Domain
- TBC
- Paid ads
- TBC
- Tools & subscriptions
- TBC
- Contractor / creative
- TBC
- Total experiment cost
- TBC
11 / Lessons learned
- 1.A service can be sold before it exists — but only until the first customer expects it.
- 2.A lead form that silently discards submissions is worse than no form at all.
- 3.Selling human verification as the premium is a defensible position while everyone else sells AI confidence scores.
12 / Next experiment
Connect caldris.live, put the consultation form behind a real inbox, then run one paid test: ten conversations with talent managers and sports agents to see whether anyone will pay €1,000/mo before the monitoring exists.