Apartment Rental Scoring Function Analysis

Date: 2026-01-18 Focus: Apartments for Rent (category=1) in top Riyadh districts

Executive Summary

Analysis of 8,583 apartment rental listings in top Riyadh districts reveals that:

CTR decays dramatically from 95% (day 0-3) to 21% (day 60+)
CVR remains stable at 5-6% across all age buckets
From Owner = 3x higher CVR (13.95% vs 4.8%)
Favorites/Shares predict quality - 4x higher for top-converting listings

This suggests a scoring function that boosts fresh listings for 3-7 days, then ranks by engagement metrics (CTR, CVR, favorites, session rate).

Top Districts Analyzed

District	City	Active Listings
النرجس (An Narjis)	Riyadh	1,613
الملقا (Al Malqa)	Riyadh	1,414
العارض (Al Arid)	Riyadh	1,130
الرمال (Ar Rimal)	Riyadh	752
الياسمين (Al Yasmin)	Riyadh	718
العقيق (Al Aqiq)	Riyadh	620
ظهرة لبن (Dhahrat Laban)	Riyadh	606
المونسية (Al Munsiyah)	Riyadh	600
اليرموك (Al Yarmuk)	Riyadh	570
الحمرا (Al Hamra)	Al Khobar	560

Key Findings

1. CTR Decays Dramatically with Age

Age	Listings	Avg Impressions	Avg Views	CTR	CVR	Session Rate
0-3 days	268	440	45	95.4%	5.9%	23.5%
4-7 days	510	1,328	101	60.0%	5.8%	22.8%
8-14 days	651	2,880	211	39.9%	5.8%	23.4%
15-30 days	1,229	5,441	310	31.2%	5.6%	20.5%
31-60 days	1,732	10,798	581	26.2%	5.1%	19.9%
60+ days	4,193	33,488	1,310	20.8%	5.2%	18.0%

Key insight: CTR drops 4.6x from new to old listings, while CVR stays stable. The funnel bottleneck is at impressions→views, not views→contacts.

2. Freshness Boost Pattern (VPD by Age)

Days Since Created	Avg VPD	Boost Factor
0-1 days	33	3.5x
2-3 days	23.6	2.5x
4-7 days	12.5	1.3x
8-14 days	15	1.6x
15+ days	9.4	1.0x (baseline)

3. From Owner = 3x Higher CVR

Attributes	Listings	CTR	CVR
Has Images + From Owner	182	13.0%	13.95%
Has Images + Agent	4,615	21.4%	4.8%
No Images + From Owner	11	39.0%	12.46%
No Images + Agent	351	42.4%	6.3%

Owner listings have 2.9x higher CVR (13.95% vs 4.8%) despite lower CTR.

4. Engagement Signals Correlate with Quality

CVR Bucket	Listings	Avg Age	CTR	Session Rate	Favorite Rate	Share Rate
10%+ (High)	571	70	41.4%	32.4%	11.43%	3.01%
5-10% (Mid)	1,528	85	28.1%	22.6%	7.87%	1.88%
2-5% (Low)	1,924	92	23.1%	17.2%	5.05%	1.26%
<2% (V.Low)	1,072	76	25.6%	12.8%	2.82%	0.81%

Favorites and shares are 4x higher for top-converting listings!

5. Contact Type Breakdown

Contact Method	Share
WhatsApp	71.7%
Calls	25.5%
Messages	2.7%

Proposed Scoring Function

Design Principles

Freshness Boost (Days 0-7): New listings get temporary boost to collect engagement data
Engagement-Based Ranking (Day 7+): Rank by normalized engagement metrics
Quality Signals: CTR, CVR, session rate, favorites, shares
Owner Boost: Owner listings get explicit boost due to higher CVR

Freshness Boost Function

def freshness_boost(days_since_created):
    if days_since_created <= 1:
        return 3.5  # Day 0-1: maximum boost
    elif days_since_created <= 3:
        return 2.5  # Day 2-3: high boost
    elif days_since_created <= 7:
        return 1.5  # Day 4-7: moderate boost
    else:
        return 1.0  # Day 8+: no boost

Engagement Score (for listings with sufficient data)

def engagement_score(listing):
    # Normalize each metric to 0-1 range within segment
    ctr_score = normalize(listing.ctr, segment_ctr_p10, segment_ctr_p90)
    cvr_score = normalize(listing.cvr, segment_cvr_p10, segment_cvr_p90)
    session_score = normalize(listing.session_rate, segment_session_p10, segment_session_p90)
    favorite_score = normalize(listing.favorite_rate, segment_fav_p10, segment_fav_p90)

    # Weighted combination (weights based on correlation with CVR)
    return (
        0.25 * ctr_score +      # CTR: 1.6x correlation
        0.30 * cvr_score +      # CVR: direct conversion metric
        0.20 * session_score +  # Session: 2.5x correlation
        0.25 * favorite_score   # Favorites: 4x correlation
    )

Combined Scoring Formula

def listing_score(listing):
    days = days_since_created(listing)

    # Base score
    if days <= 7:
        # New listings: use freshness boost
        base_score = freshness_boost(days)
    else:
        # Established listings: use engagement score
        base_score = engagement_score(listing)

    # Apply multipliers
    score = base_score
    score *= 1.0 + (0.3 * listing.from_owner)  # +30% for owner
    score *= 1.0 + (0.2 * listing.has_images)  # +20% for images
    score *= 1.0 + (0.15 * listing.has_video)  # +15% for video
    score *= 1.0 + (2.0 * listing.is_premium)  # +200% for premium

    return score

SQL Implementation

-- Calculate listing score
SELECT
  l.id,
  l.district_id,
  days_since_created,
  -- Freshness boost (first 7 days)
  multiIf (
    days_since_created <= 1,
    3.5,
    days_since_created <= 3,
    2.5,
    days_since_created <= 7,
    1.5,
    1.0
  ) AS freshness_boost,
  -- Engagement score (normalized within segment)
  (
    0.25 * (ctr - segment_ctr_p10) / nullIf(segment_ctr_p90 - segment_ctr_p10, 0) + 0.30 * (cvr - segment_cvr_p10) / nullIf(segment_cvr_p90 - segment_cvr_p10, 0) + 0.20 * (session_rate - segment_session_p10) / nullIf(segment_session_p90 - segment_session_p10, 0) + 0.25 * (favorite_rate - segment_fav_p10) / nullIf(segment_fav_p90 - segment_fav_p10, 0)
  ) AS engagement_score,
  -- Combined score
  if (
    days_since_created <= 7,
    freshness_boost,
    greatest (engagement_score, 0.1)
  ) * (1 + 0.3 * from_owner) * (1 + 0.2 * has_img) * (1 + 2.0 * is_premium) AS final_score
FROM
  listings_with_metrics l
  JOIN segment_percentiles p ON l.category = p.category
  AND l.district_id = p.district_id
ORDER BY
  final_score DESC

Minimum Data Thresholds

Before using engagement-based scoring, ensure sufficient data:

Metric	Minimum Threshold
Impressions	≥ 100
Views	≥ 20
Days Active	≥ 7

For listings below thresholds, use freshness boost or segment average.

Expected Impact

Metric	Current State	Expected After
Avg CTR (new listings)	95% (too high, wasted impressions)	~60% (more targeted)
Avg CVR (overall)	5.5%	6.5%+ (better matching)
New listing visibility	Over-boosted day 0-3	Sustained boost day 0-7
High-quality listing rank	Based on total views	Based on engagement quality

Monitoring Metrics

Track these to validate the new scoring:

CTR by age bucket - Should become more stable
CVR by age bucket - Should improve for older listings
Favorite rate vs rank - Should correlate positively
From Owner performance - Should improve with explicit boost
New listing success rate - % converting to high-engagement after 7 days

SQL Queries Used

CTR/CVR by Age Bucket

SELECT
  age_bucket,
  count() AS listings,
  round(avg(total_impressions), 0) AS avg_impr,
  round(avg(total_views), 0) AS avg_views,
  round(avg(ctr) * 100, 2) AS avg_ctr_pct,
  round(avg(cvr) * 100, 2) AS avg_cvr_pct
FROM
  (
    SELECT
      l.id,
      coalesce(sum(s.impressions), 0) AS total_impressions,
      coalesce(sum(s.views), 0) AS total_views,
      coalesce(sum(s.calls + s.whatsapp + s.msgs), 0) AS total_contacts,
      total_views / greatest (total_impressions, 1) AS ctr,
      total_contacts / greatest (total_views, 1) AS cvr,
      dateDiff (
        'day',
        toDate (fromUnixTimestamp (l.create_time)),
        today ()
      ) AS days,
      multiIf (
        days <= 3,
        'a_0-3d',
        days <= 7,
        'b_4-7d',
        days <= 14,
        'c_8-14d',
        days <= 30,
        'd_15-30d',
        days <= 60,
        'e_31-60d',
        'f_60+d'
      ) AS age_bucket
    FROM
      sadb_listings l FINAL
      LEFT JOIN sadb_stats_listings_stats s FINAL ON l.id = s.id
      AND toDate (s.day_time) >= '2024-01-01'
    WHERE
      l.category = 1
      AND l.district_id IN (600, 570, 494, 446, 632, 504, 688, 590, 634, 1070)
      AND l.status IN (0, 4)
      AND l.published = 1
      AND l.hidden = 0
      AND l.rega_licensed = 1
      AND l._peerdb_is_deleted = 0
      AND toDate (fromUnixTimestamp (l.create_time)) >= '2024-01-01'
    GROUP BY
      l.id,
      days
  )
GROUP BY
  age_bucket
ORDER BY
  age_bucket

Engagement Signals by CVR Bucket

SELECT
  cvr_bucket,
  count() AS listings,
  round(avg(days), 0) AS avg_age,
  round(avg(ctr) * 100, 1) AS avg_ctr,
  round(avg(session_rate) * 100, 1) AS sess_rate,
  round(avg(fav_rate) * 100, 2) AS fav_pct
FROM
  (
    SELECT
      l.id,
      dateDiff (
        'day',
        toDate (fromUnixTimestamp (l.create_time)),
        today ()
      ) AS days,
      coalesce(sum(s.impressions), 0) AS impr,
      coalesce(sum(s.views), 0) AS views,
      coalesce(sum(s.calls + s.whatsapp + s.msgs), 0) AS contacts,
      coalesce(sum(s.session_30_seconds), 0) AS sessions,
      coalesce(sum(s.favorited), 0) AS favs,
      views / greatest (impr, 1) AS ctr,
      contacts / greatest (views, 1) AS cvr,
      sessions / greatest (views, 1) AS session_rate,
      favs / greatest (views, 1) AS fav_rate,
      multiIf (
        cvr >= 0.1,
        'a_10%+',
        cvr >= 0.05,
        'b_5-10%',
        cvr >= 0.02,
        'c_2-5%',
        'd_<2%'
      ) AS cvr_bucket
    FROM
      sadb_listings l FINAL
      LEFT JOIN sadb_stats_listings_stats s FINAL ON l.id = s.id
      AND toDate (s.day_time) >= '2024-01-01'
    WHERE
      l.category = 1
      AND l.district_id IN (600, 570, 494, 446, 632)
      AND l.status IN (0, 4)
      AND l.published = 1
      AND l.hidden = 0
      AND l.rega_licensed = 1
      AND l._peerdb_is_deleted = 0
    GROUP BY
      l.id,
      days
    HAVING
      views >= 50
  )
GROUP BY
  cvr_bucket
ORDER BY
  cvr_bucket