Read first · full change report

Harmonia investor deck rewrite

The source deck should not be circulated. The rewrite replaces a broad, contradictory scale story with a narrow, falsifiable validation-seed case.

Source: 21-slide PowerPoint
Send-ahead: 16 main + 11 appendix
Live: 11 main + 7 Q&A
Status: Local review only

Executive decision

The source deck mixed a broad product scope, three different funding asks, unverified competitive claims, crisis-support language, and precise 2030–2035 financial outcomes without the operating evidence needed to defend them.

The rewrite changes the financing thesis: Harmonia is seeking a provisional €400k validation seed for an 18-month target runway to prove one narrow Croatian parenting workflow. Both figures are internal assumptions until a founder-approved monthly cash ledger reconciles hiring, product, governance, pilot, GTM, legal, security, and reserve costs.

Can one bounded, governed workflow safely convert a recurring parenting moment into a useful next step, retain paying families, and acquire them through one repeatable trusted channel?

Delivered narrative architecture

Detailed send-ahead: 16 main slides plus a diligence appendix.
CEO live deck: 11 main slides plus a Q&A appendix and presenter notes.
One rendering engine, but separate slide definitions; the live deck is not the send-ahead deck with hidden paragraphs.
Five visible evidence states: verified fact, internal evidence, assumption, target, and product intent.
Every rewritten slide retains source-slide provenance back to the 21-slide PowerPoint.

What changed and why

1. The customer and job became narrow

BeforeParents of children aged 2–17, multiple needs, B2C/B2B/B2B2C, and multi-country expansion.

AfterAn explicit hypothesis—Croatian-speaking caregivers of children aged 3–7 testing one recurring non-emergency transition, boundary, or routine job.

WhyThe source scope could not generate fast, interpretable learning. One cohort, one job, one geography, and explicit exclusions reduce product, clinical, acquisition, and measurement ambiguity.

2. The product became a governed workflow, not “AI advice”

BeforeBroad personalised support, proactive suggestions, and a “Red Button” crisis protocol whose model, source, ownership, and escalation design were unclear.

AfterMinimum-necessary context → governed source/protocol path → one bounded action → outcome reflection. Low-risk explanation may be generated; defined immediate guidance is pre-approved; out-of-scope indicators trigger refusal and the appropriate human or emergency route.

WhyGeneration is commodity infrastructure. The investable product claim must be the constrained workflow, protocol governance, evaluation, incident handling, and measurable learning loop. Crisis, diagnosis, treatment, and autonomous safeguarding claims were removed because the evidence does not support them.

3. “Unique” became an earned-defensibility thesis

Before“Only Harmonia” and feature checkmarks implied uniqueness.

AfterLocal trust is presented as a wedge, not a moat. Potential defensibility is sequenced as governed protocol rights, an expert-labelled evaluation set, consented longitudinal outcomes, a safety history, and repeatable trusted distribution.

WhyGood Inside already markets personalised, real-time parenting support, and Joy announced a funded AI parenting product. Features and prompts are copyable; governed evidence and distribution may compound only if Harmonia actually earns them.

4. Workshop reach stopped masquerading as product traction

BeforeWorkshops involving 100+ children sat inside the validation story.

AfterThey are labelled unverified internal evidence that may demonstrate problem access and programme delivery—not app usage, retention, paid demand, AI quality, or product-market fit.

WhyAn investor will separate founder access from product evidence. The method, dates, institutions, permissions, outcomes, and founder/adviser claims must be verified before external use.

5. The roadmap moved from delayed launch to fast falsification

BeforeMVP in 2027, soft launch later in 2027, commercial launch in 2028, then long-range geographic and financial milestones.

AfterMonths 0–3 define scope, governance, data map, protocol rights, evaluation, and a concierge prototype; months 4–12 run a 100-parent, 90-day programme, closed alpha, comparator tests, and paid beta; months 13–18 repeat the winning channel and launch regionally—or narrow, rework, or stop.

WhyAn AI-native company cannot spend 10–18 months avoiding meaningful customer and safety evidence. The round now buys staged de-risking, not an eight-year forecast.

6. The proof programme gained denominators and stop conditions

BeforeValidation was described through activity and future releases.

AfterA four-week active protocol plus eight-week follow-up, with internal decision targets of at least 60% activation, at least 35% week-four retention of activated parents, at least 15% paid continuation by week six, and zero unhandled high-severity safety events. Expert/general-AI comparison, consent, deletion, escalation, refunds, and channel economics are included.

WhyThese targets can disprove the thesis. They are deliberately labelled internal thresholds—not benchmarks or achieved results.

7. GTM became one measurable acquisition loop

BeforeMultiple channels, markets, and business models ran in parallel.

AfterB2C subscription first; workshops and trusted parent/professional relationships recruit the initial cohort. The funnel is qualified reach → activation → retention → paid continuation → contribution margin. B2B2C is a 20-account design-partner test, not booked enterprise revenue.

WhyLists of channels conceal accountability. The rewrite forces one route to prove reach, conversion, payback, and operational fit before expansion.

8. Market sizing became bottom-up and gated

BeforeTop-down TAM/SAM/SOM and international expansion created false precision.

AfterThe market equation is eligible households × reachable share × activation × paid conversion × annual price. Regional and English-language markets are contingent upside only after safe paid retention and a repeatable channel are demonstrated.

WhyCategory size does not prove reachable revenue. Household counts and conversion ranges remain held until a reproducible sourced model exists.

9. Financials became driver-based

Before€250k, €350k, and €400k asks appeared across the deck; a €1m future round was embedded in the plan; positive EBITDA in 2030 and approximately €40.6m revenue/~50% EBITDA margin in 2035 were presented with insufficient sensitivity evidence.

AfterOne provisional local-review case: €400k, 18 months, and no promised EBITDA year. The base sensitivity uses €11.99 monthly price, €7.24 contribution per paid parent after a 25% tax/payment/refund reserve and €1.75 variable cost, and €25k mature monthly fixed cost—implying approximately 3,452 concurrent paid parents for B2C-only operating break-even. Across the low/high cases, the range is approximately 2,634–5,008 paid parents. The €25k figure is a mature sensitivity, not seed-period run-rate; the seed model caps operating spend at €360k over 18 months and holds €40k as contingency.

WhyBreak-even must be an output of price, retention, CAC, support, inference, staffing, safety, and fixed-cost evidence—not a date chosen for the headline. Every displayed input remains an assumption pending measured cohorts, vendor quotes, finance/tax review, and the monthly cash ledger.

10. The use of funds was tied to evidence

BeforeThe source deck did not consistently connect capital to the proof package.

AfterThe provisional ceiling is €130k core team/programme delivery; €85k product/engineering; €45k clinical protocol, safeguarding, and quality; €30k pilot operations; €30k channel tests; €20k privacy/legal/finance; €20k tools/admin; and €40k contingency. Operating spend is capped at €360k over 18 months; the reserve is not counted as operating spend.

WhyThis split is a local-review hypothesis, not an approved budget, and must be replaced by the monthly ledger before circulation.

11. Safety, privacy, and accountability moved into the core case

Before“Psychologist validated” carried too much implied assurance, while data flows, consent, retention/deletion, model-training boundaries, incident handling, and accountable technical/safeguarding ownership were unclear.

AfterThe main story and appendices state the exclusions, minimum-data principle, refusal/escalation boundary, governance requirements, evaluation loop, incident review, and rollback rule. Design baselines cite UNICEF and European Commission guidance without claiming compliance or demonstrated safety.

WhyLongitudinal child/family context is a central diligence risk. Trust cannot be asserted; it must be designed, owned, tested, and evidenced.

12. The team slide became an accountability slide

BeforeBiographies implied coverage of clinical, technical, product, privacy, security, and safeguarding work.

AfterThe founders’ reported commercial, psychology, family-support, and programme experience is stated as internal evidence, while missing accountable technical, privacy/security, evaluation, and safeguarding ownership is explicit.

WhyInvestors fund execution responsibility. Adviser biographies and domain proximity do not substitute for committed owners.

Source-slide disposition

Source	Original job	New destination	Decision
1, 21	Opening and close	Opening and provisional ask	One consistent validation-seed thesis; conflicting asks removed
2–4	Problem, why now, market gap	Decision gap and alternatives	Generic/unsourced claims removed; one urgent job retained
5–7, 10	Solution, proactive support, workflow, methodology	Workflow, architecture, safety, data/evaluation appendices	Features decomposed into governed paths and explicit boundaries
8–9	Competition and advantage	Alternatives, defensibility, competition appendix	Unique-feature moat removed; sources and earned moat substituted
11–12	Audience and market	Narrow ICP and bottom-up market method	Ages 2–17 and top-down precision removed
13–14	GTM and geography	One trusted-channel funnel and contingent expansion	Parallel channels/markets replaced by evidence gates
15	Validation and roadmap	Internal evidence, 90-day programme, 18-month gates	Activity separated from product proof; feedback accelerated
16	Revenue model	B2C-first test and pricing/economics appendices	Simultaneous models replaced by one primary engine
17–19	Financials and costs	Driver-based economics, use of funds, stop conditions	2030/2035 precision and automatic follow-on removed
20	Team	Accountable execution and diligence gaps	Résumé volume replaced by ownership and verification

Claims removed or downgraded

Good Inside 100,000+ subscribers and ~$34m revenue: removed; the source trail was insufficient for investor use.
Joy ~50,000 users, $41.5m total funding, and ~$5–7m revenue: removed; only the retrievable $14m Series A announcement is cited.
“Only Harmonia”/unique personalised real-time support: removed.
Red Button crisis support in 2–3 seconds: removed and not renamed as an emergency product.
Workshops as app validation: downgraded to unverified internal problem access.
Croatia/regional expansion as low-CAC: downgraded to a channel hypothesis.
~80% gross margin and 3.0x LTV:CAC: removed as outcomes; underlying inputs must be tested.
Positive EBITDA in 2030 and 2035 scale economics: removed from the pitch.
The €1m follow-on round: treated only as a possible future financing dependency—not part of the base case or proof.

Evidence and sources

Preserved source: RESEARCH/HARMONIA_PITCH_DECK_SOURCE_2026_08_03.pptx
Source SHA-256: 4024c8d6bebbfe2a00b302abcaf5d434a92b153ea3695ff4d96f68b23d9b6a9d
Integrated audit: RESEARCH/HARMONIA_INVESTOR_DILIGENCE_AUDIT_2026_08_03.md
Claim ledger: RESEARCH/HARMONIA_CLAIM_SOURCE_LEDGER_2026_08_03.md
Independent investor memo: RESEARCH/HARMONIA_RUTHLESS_INVESTOR_MEMO_2026_08_03.md
GTM/economics rebuild: RESEARCH/HARMONIA_GTM_ECONOMICS_REBUILD_2026_08_03.md
Narrative/production spec: OUTBOX/HARMONIA_DECK_NARRATIVE_AND_PRODUCTION_SPEC_2026_08_03.md
Technical audit: RESEARCH/HARMONIA_SOURCE_DECK_TECHNICAL_AUDIT_2026_08_03.md

External references displayed in the deck include Good Inside product/pricing, Joy’s Series A announcement, UNICEF guidance on AI and children, and European Commission GDPR, child-data, and AI Act guidance. Retrieval dates are stored with the source records.

External-circulation blockers

Local-test ready is not investor-circulation ready. Before any send, founders must verify the product state, founder/adviser roles and consent, workshop method and permissions, protocol rights, clinical/safeguarding ownership, architecture and data map, privacy/legal/security position, escalation/incident process, monthly cash ledger, final ask, use of funds, and model assumptions.

No investor contact, publication, product release, pilot recruitment, price test, financing commitment, or terms are authorised by this rewrite.

Delivery evidence

Repository: REPOS/harmonia-investor-decks
Branch: feature/investor-deck-scaffold
Deck rewrite baseline: edfa5504d23892d53b8ec12ce233c5a2f98da811

Clean working tree on feature/investor-deck-scaffold.
npm test: 9/9 rendered-route and static-export tests passed.
npm run lint, npm run build, and npx tsc --noEmit passed.
Routes built: /, /report, /send-ahead, /live, and /live/notes.
A cPanel-ready static package reproduces all five routes with Apache configuration and upload instructions; it requires no Node.js process, database, or application server.
16 and 11 main-slide contracts; 11 and 7 appendix-slide contracts.
Maximum visible main-slide density: ≤70 words send-ahead and ≤35 words live.
Source mapping is present on every delivered slide and covers source slides 1–21.
All 11 live main slides have separate presenter notes; note payload is excluded from the audience deck.
The 1200×630 social-preview image and metadata are present without hosting.

The in-app browser backend was unavailable, so Tom’s local browser test remains the visual acceptance gate. That limitation does not invalidate the deterministic route, density, navigation, print, accessibility, or build evidence above.