AI doporučení nemovitostí 2026: explainable scoring + 6 archetype

1. Hyper-personalization — proč explainable scoring vs black-box ML

Property recommendation systems globally split into 2 paradigms: black-box ML (collaborative filtering, deep learning embeddings) vs explainable rule-based scoring. **CZ 2026 Jistybyt uses explainable scoring** for legal + practical reasons:

**Legal: GDPR Article 22 automated decision-making transparency:**

• Article 22 GDPR — individuals have right to NOT be subject to decision based solely on automated processing

• If recommendations significantly impact financial decisions, individuals have right to obtain meaningful information about logic involved

• Black-box ML (neural networks, ensemble models) often cannot provide meaningful explanation

• Explainable scoring with weighted factors + rationale per recommendation = compliant

**Practical: User trust + adoption:**

• Users want to understand WHY specific archetype recommended

• Black-box "computer says urban-apartment" = 25-40 % adoption rate

• Explainable "young-couple + transit + central careers = urban-apartment matches profile" = 65-80 % adoption rate

**CZ market reality 2026:**

• ~85 000 ročních property purchases

• ~30 % buyers report archetype misalignment (purchased without proper life stage analysis)

• Misalignment = repeated transactions within 5-7 let (selling apartment to buy house, etc.)

• Each repeat transaction costs 200-500k Kč (transfer costs + advokát + opportunity)

• Personalization potential value: 60-150k Kč per buyer (avoidance of repeat transactions)

**6 PropertyArchetype framework:**

• urban-apartment (4-12M Kč)

• periphery-apartment (3-7M Kč)

• townhouse-rowhouse (6-12M Kč)

• family-house (8-25M Kč)

• cooperative-share (2,5-8M Kč)

• investment-rental (3,5-8M Kč)

2. Scoring algorithm — multi-factor weighted

Jistybyt scoring algorithm uses transparent weighted factors per archetype. Each archetype has base score (30-60) + adjustments based on profile match.

**Scoring factors:**

• **LifeStage** (6 categories): young-single → urban +25, family-with-kids → family-house +20, retirement → urban +10 (healthcare proximity)

• **CommutePreference** (5 categories): walking-only → urban +20, periphery -15 (incompatible)

• **NumberOfChildren**: 2+ → family-house +30, urban -15 (space limit)

• **FinancialPriority** (6 categories): minimum-cost → periphery/cooperative +15, rental-yield → investment +35

• **RiskTolerance** 0-100: low (< 30) excludes investment archetype, high (> 70) opens commercial

• **OwnershipPreference** 0-100: high (> 80) excludes cooperative-share, flexibility favors rental

• **plansLongTermStay**: yes → family-house/townhouse +15, vlastnictví preference

• **acceptsLongerCommute**: enables periphery + house in remote locations

**Example calculation pro young-couple s 2 dětmi + transit + cost-priority:**

• Urban-apartment: 50 base + 25 (young-couple) + 20 (transit) - 15 (children) = 80 score

• Periphery-apartment: 60 base + 20 (children OK) + 15 (cost) = 95 score

• Family-house: 30 base + 30 (children) + 0 (no commute tolerance) - 25 (low income gen) = 35 score

• Townhouse: 40 base + 20 (couple-family transition) + 15 (children fit) = 75 score

• Cooperative-share: 30 base + 25 (cost-priority) - 30 (high ownership) = 25 score

• Investment-rental: 20 base - 15 (children focus) = 5 score

• **Top: periphery-apartment 95**, runner-up urban-apartment 80

**Rationale generation:** every score adjustment generates rationale text explaining factor + impact. Users see "Children 2 — periphery balances space + affordability (+20 score)" allowing transparent decision-making.

3. Borrowing capacity — ČNB Doporučení 1/2022 integration

Recommendation engine integrates ČNB Doporučení 1/2022 DSTI/DTI/LTV limits pro borrowing capacity calculation:

**DSTI 45 % limit applied** (mid-range default — could be 50 % < 36 let or 45 % ≥ 36 let):

• Max monthly payment = householdIncome × 0,45 - existingDebt × 0,05/12

• Annuity formula reverse: principal = payment × (1 - (1+r)^-n) / r

• 25-year mortgage assumption + 5 % p.a. rate = principal × 0,065 / month

**Borrowing capacity calculation example:**

• Income 100k Kč/měs

• Max payment 100k × 0,45 = 45k Kč/měs

• 5 % rate × 25 years = annuity factor 153 (1 / 0,00653)

• Principal = 45k × 153 = **6,9M Kč borrowing capacity**

**Total purchasing power:**

• Borrowing capacity (6,9M)

• + Savings (1,5M typical)

• - Existing debt (0)

• = **8,4M Kč total purchasing power**

**Maps to archetype price ranges:**

• Urban apartment 4-12M → upper end

• Periphery apartment 3-7M → fully accessible

• Townhouse 6-12M → lower-mid range

• Family-house 8-25M → starter range

• Cooperative 2,5-8M → easily accessible

• Investment 3,5-8M → entry-level investment

**Critical considerations:**

• If savings < 20 % of purchasing power: ČNB LTV 80 % limit may apply

• Pro < 36 let: 90 % LTV výjimka 1× za život available

• High existing debt reduces capacity proportionally

• ČNB stress test +2 % rate buffer recommended

4. 6 PropertyArchetype detailed profiles

**Urban-apartment (city center byt 1+kk-3+kk, 4-12M Kč):**

• Base score 50

• Bonuses: young-single +25, young-couple +25, walking-only +20, transit +20, retirement +10

• Penalties: 2+ children -15, short-drive without commute tolerance triggers warning

• Use cases: career-focused professionals, social life priority, transit-dependent

• Examples: Praha 1 Staré Město, Praha 2 Vinohrady, Praha 7 Holešovice

**Periphery-apartment (suburban byt 2+kk-4+kk, 3-7M Kč):**

• Base score 60 (highest baseline — good fit for many profiles)

• Bonuses: 1-2 children +20, cost priority +15, family + commute tolerance +10

• Penalties: walking-only -15 (incompatible)

• Use cases: starting families, commute-tolerant workers, cost-conscious buyers

• Examples: Praha 4 Modřany, Praha 5 Stodůlky, Praha 9 Letňany, Praha 11 Háje

**Townhouse-rowhouse (řadovka/klínovka, 6-12M Kč):**

• Base score 40

• Bonuses: young-couple/family transition +20, 1-2 children +15, long-term-value +10

• Use cases: starting families wanting house-like layout, middle-ground apartment vs full house

• Examples: Praha 6 Břevnov, Praha 5 Stodůlky, suburban developments

**Family-house (rodinný dům 4+1/5+1 + zahrada, 8-25M Kč):**

• Base score 30 (lower baseline — selective fit)

• Bonuses: 2+ children +30, family-with-kids +20, plans long-term +15

• Penalties: low income (< 80k) -25, walking-only/transit triggers warning

• Use cases: established families, multi-generational living, gardening priority

• Examples: Praha periphery (Lhotka, Šeberov, Lipence), regional centers

**Cooperative-share (bytové družstvo, 2,5-8M Kč):**

• Base score 30

• Bonuses: limited equity + lower income +25, long-term-stay +10

• Penalties: high ownership preference -30

• Always-warning: bank mortgage NOT available pro pure cooperative shares

• Use cases: limited cash equity, flexibility-tolerant buyers, future SVJ conversion path

**Investment-rental (3,5-8M Kč):**

• Base score 20 (specialty profile)

• Bonuses: rental-yield/long-term-value priority +35, savings 2M+ +15, income 120k+ +10

• Penalties: 2+ children -15 (focus dilution)

• Use cases: high-income diversification, retirement-supplement income, real-estate investment

• Examples: secondary apartments, well-located studios, university-area rentals

5. Lifestyle insights + financial considerations engine

Beyond archetype matching, recommendation engine generates context-aware insights:

**Lifestyle insights triggered by:**

• young-single + remote-flexible: "Geography flexibility — můžete optimize hodnotu/komfort vs lokalita"

• family-with-kids 2+: "Space + zahrada + školy priority. Periphery family-house often optimal"

• plansLongTermStay: "5+ let stay — vlastnictví financially preferable vs nájem"

• retirement: "Healthcare proximity + amenity walkability essential pro mobility"

• empty-nester: "Right-sizing opportunity — sell large house, downsize to apartment"

**Financial considerations triggered by:**

• High savings + good income: "Premium archetype options accessible"

• Limited savings + good income: "ČNB LTV 80 % limit apply, consider 90 % výjimka pokud < 36 let"

• High existing debt: "DSTI capacity reduced, consider debt consolidation pre-purchase"

• Low risk tolerance + investment archetype matched: "Avoid investment, stable rental yield strategy preferable"

• High income + cooperative match: "Bank mortgage unavailable for cooperative — consider direct ownership instead"

**Decision tree integration:**

• Top recommendation = highest scored archetype

• Backup recommendations = 2nd + 3rd scored

• If top score < 50: "No archetype strong fit — consider profile reassessment OR custom search"

• If multiple archetypes > 70: "Multiple good fits — secondary factors decisive (school district, neighborhood, specific amenities)"

**Output format:**

• Top recommendation card: archetype name + match score + price range + monthly cost + 3 rationale

• Alternative archetypes table: 5 alternates s scores + rationale + warnings

• Lifestyle insights paragraph (3-5 personalized observations)

• Financial considerations checklist (3-7 items)

• Action plan: next steps (engage advokát, get mortgage pre-qual, view properties)

6. Future enhancements — vector embeddings + collaborative filtering

Current explainable scoring V1 will evolve while maintaining explainability:

**V2: Vector embeddings (planned 2026 Q4):**

• Profile vectors (multi-dimensional preferences) compared to archetype vectors via cosine similarity

• Dimensionality reduction (PCA / UMAP) for explainability

• Each dimension labeled (e.g. "urban-density", "family-orientation", "investment-mindset")

• Combined with rule-based scoring for compliance

**V3: Collaborative filtering (planned 2027):**

• Anonymous similar buyer profile matches (privacy-preserving — no PII)

• "Buyers like you also considered:" recommendations

• Differential privacy + k-anonymity (k ≥ 50) compliance

• Aggregate-level insights only

**V4: Property-level integration (planned 2027):**

• Match archetype + specific listings on /listings

• Pre-filter properties by archetype scoring

• Personalized search ranking

**V5: Behavioral learning (planned 2028):**

• Track recommendation acceptance / rejection

• Refine scoring weights via A/B testing

• Personalized weights per user (with consent)

**Privacy architecture:**

• All ML training on anonymized aggregates

• User profile data never trains models without explicit consent

• EU AI Act 2024 compliance (high-risk system documentation)

• Regular bias audits per EU directive 2024

**Explainability principle preserved across V1-V5:**

• Every recommendation includes rationale

• Users can request "why" explanation

• Override factors visible (user can adjust weights)

• Decision trail audit-able

7. 7 doporučení + 5 chyb

**7 strategických doporučení:**

1. **Use recommendation engine before search** — start s archetype identification (Jistybyt kalkulátor 5 minut), then narrow to specific listings (saves 10-30 hodin browsing).

2. **Verify recommendation rationale** — review WHY archetype recommended. If rationale doesn't match self-perception, reassess profile inputs.

3. **Consider top 2-3 archetypes**, not just #1 — secondary factors (school district, specific amenity) often decisive.

4. **Borrowing capacity calculation** — verify against actual bank pre-qualification. Jistybyt ČNB DSTI 45 % is conservative estimate, banks may approve more or less.

5. **Match life stage to long-term plans** — buying for current life stage AND 5-10 year evolution. Children growing? Career change? Retirement timing?

6. **Investment archetype only after primary housing solved** — don't mix investment property when primary housing not yet stable. Rule of thumb: invest only after 12+ months in primary home.

7. **Re-run recommendation every 3-5 years** — life stages evolve. Young-couple → family-with-kids → empty-nester transitions warrant re-evaluation.

**5 typických chyb:**

1. **Skip life stage analysis** — young-single buying family-house = poor fit. Same applies family with 3 kids buying urban-apartment.

2. **Ignore commute preference** — buying remote location with transit-only preference = daily friction + likely sale within 3-5 let.

3. **Borrow at maximum DSTI 50 %** — ČNB limit ne comfortable. Aim 30-35 % comfort zone, leaves buffer for life events.

4. **Ignore children growth planning** — 1 child today + planning 2nd = need extra bedroom + green space. Plan for 2-3 children if applicable.

5. **Mix investment vs primary housing** — investment-rental needs ownership mindset (returns focus), primary housing needs lifestyle fit. Different goals, different criteria.

8. Závěr — recommendation strategy + doporučená kombinace nástrojů

**Klíčové insighty:**

• **CZ 2026: ~85 000 ročních property purchases**, ~30 % archetype misalignment

• **Repeat transactions cost 200-500k Kč** each — personalization saves 60-150k Kč per buyer

• **Explainable scoring (vs black-box ML)** — GDPR Article 22 compliance + 65-80 % adoption rate

• **6 PropertyArchetype framework**: urban/periphery apartments, townhouse, family-house, cooperative-share, investment-rental

• **7 scoring factors**: LifeStage, CommutePreference, FinancialPriority, RiskTolerance, OwnershipPreference, NumberOfChildren, plansLongTermStay

• **ČNB Doporučení 1/2022 DSTI 45 %** integration pro borrowing capacity

• **Lifestyle insights + financial considerations** engine — context-aware recommendations

• **Future enhancements**: V2 vector embeddings, V3 collaborative filtering, V4 listings integration, V5 behavioral learning

• **5 chyb** — skip life stage analysis, ignore commute, borrow at max DSTI, ignore children growth, mix investment vs primary

Doporučená kombinace nástrojů: /api/property-recommendation (6 PropertyArchetype × 6 LifeStage × 5 CommutePreference × 6 FinancialPriority + scoring engine) → /api/buy-vs-rent (cluster pre-existing — financial decision after archetype identified) → /api/avm (cluster pre-existing — single property valuation post-search) → /api/dsti-dti (cluster pre-existing — borrowing capacity verification) → /api/listings (cluster pre-existing — search by archetype filter).

Jistybyt je jediná CZ platforma, která spočítá **AI doporučení nemovitostí s 11 parametry** (householdMonthlyIncomeCzk, savingsCzk, existingDebtCzk, lifeStage, numberOfChildren, commutePreference, financialPriority, riskTolerance, ownershipPreference, acceptsLongerCommute, plansLongTermStay) → 6 PropertyArchetype × 6 LifeStage × 5 CommutePreference × 6 FinancialPriority matrix + explainable weighted scoring engine + ČNB DSTI 45 % borrowing capacity + lifestyle insights + financial considerations + actionable next steps. **Bez kalkulátoru riskujete: skip life stage analysis (archetype misalignment 30 % CZ market = 200-500k repeat transaction costs), ignore commute preference (daily friction + 3-5 let sale cycle), borrow at maximum DSTI 50 % (ne comfortable, buffer needed), ignore children growth planning (1 → 2-3 child need extra space). S kalkulátorem máte transparency o per-archetype matching score + rationale + warnings + price range + monthly cost projection + alternative archetypes + lifestyle insights + financial considerations engine + GDPR Article 22 compliant explainable AI.**