Book a demo

Case study · Retail

Turning 230,000+ reviews into actions for every store manager at a leading jewellery brand

We built an AI layer and a set of custom reports that tell each store, area and regional manager what to fix, whom to reward and where competitors are winning.

Google reviews analysed in under a year
230,631
Response rate on platform-triggered surveys
~50%
Competitors benchmarked in every region
5

The challenge

What wasn't working

  • NPS feedback and Google reviews were analysed by hand, so reports were slow and inconsistent.
  • Raw feedback didn't tell any manager what to fix next.
  • There was no fair, data-backed way to recognise staff that customers praised.

The solution

What we did

  1. One data pipeline

    NPS files and Google reviews, for the brand and its top five competitors, flow in automatically.

  2. AI analysis

    Every review is tagged by category and sentiment, with staff names and critical reviews detected.

  3. Mapped to the org chart

    Each insight rolls up from store to area to region, so every manager sees their own stores.

  4. Reports managers use

    Next best actions, staff scoring, competitor benchmarks and NPS, scheduled to each manager.

From raw feedback to a manager's reportThree sources, one pipeline, one model for the brand and its competitors.
  1. Sources
    • Post-purchase NPS
    • Google reviews: the brand
    • Google reviews: top 5 competitors
    • Mystery-shopper surveys
  2. Ingest & store
    • SFTP delivery
    • Sync to S3
    • Processing jobs
    • PostgreSQL + Metabase
  3. AI analysis
    • Category & sentiment
    • Staff-name detection
    • Critical-review detection
    • Mapping to store, area, region
  4. Delivered
    • Custom reports
    • Scheduled to each manager
    • Competitor benchmarks
    • Survey triggers

Results

What changed

  • 230,631 Google reviews categorised by sentiment, April 2025 to February 2026.
  • Staff rewarded through the Staff Scoring Report, based on what customers wrote.
  • Managers see which issues got better or worse between periods, for their own region and area.
  • 4 surveys triggered from the platform, with a response rate of about 50%.
  • Scheduled reports replaced the manual analysis teams used to do.
230,631 reviews by sentimentGoogle reviews, April 2025 to February 2026.
  • 160,727Positive · 69.7%
  • 67,831Neutral · 29.4%
  • 2,073Negative · 0.9%
What praise and complaints are aboutEach topic's share of positive and negative reviews. Staff service leads both; value for money is a far bigger share of complaints than of praise.
  • Share of positive reviews
  • Share of negative reviews
  • Staff service & knowledge
  • Product type & variety
  • Value for money
  • Product quality
  • Policy, process & billing
  • Product design
  • Size issues
  • Offers & promotions
Next best action: what movedOne region, two consecutive weeks in October 2026: each topic's share of all feedback, and the change.
TopicWeek 1Week 2−10 pts0+10 pts
Making charge30.38%22.35%−8.03
Plain gold: design & variety17.58%10.96%−6.62
Product display0.18%1.84%+1.66
Payment process1.50%0.79%−0.71
OTP & network issues1.15%0.44%−0.71

Share grew: flagged for attention Share fell

Regional benchmarkPositive-review share, the brand vs the regional leader, February 2026.
  • The brand
  • Regional leader
Strongest region
99.08%100%
Weakest region
94.9%99.47%

Axis starts at 94%. Dashed line: brand-wide average, about 98%.

One store's NPSThe NPS report for a single store, October 2026.

87.47NPS from 6,064 responses

  • 5,514Promoters · 90.93%
  • 340Passives · 5.61%
  • 210Detractors · 3.46%

Next case study

All case studies
HospitalitySelling room upgrades and add-ons before guests arrive, for a leisure resort groupWe connected guest messaging to the group's PMS in both directions, and built a segment-based journey on WhatsApp and email that sells upgrades before guests arrive.Read the case study

See what Falcon finds in your feedback.

Book a 30-minute demo, or take the interactive tour of the workspace.

Book a demo