Shriram Finance
less than 3–4 hours — Onboarding Turn-Around-Time (TAT)
Independently reportedBanking & finance · Process automation · Classical ML · In production
Process automation
GreenifyAI · EasyDeploy AI
Vendor-reportedConstruction & real estateIn productionClassical ML
Vendor-reported. The customer is named and the numbers are quoted from the source page, but the account comes from the vendor. No independent confirmation.
Reported by easydeploy.ai (vendor self-report). Checked against the source page on 2026-09-05. 7 of 7 figures below appear on that page word for word.
The problem
Grading MLS listings for sustainability meant one LLM call per listing, which was slow and expensive to scale.
What was deployed
- Vendor
- EasyDeploy AI
- Products
- EasyDeploy
- Technique
- Classical ML
- Build or buy
- Bought off the shelf
- Deployment
- API
- Data used
- 928 listings that an LLM had already graded became training data.
- Scale
- 19,484 listings scored out of 113,398 total catalogue
What changed
Each row is quoted from the source. Figures we could not find on the page in those words are marked — they are kept, not deleted, so you can judge them.
19,484 listings scoredsustainability coveragefrom 928 labeled listings
“GreenifyAI turned 928 labeled listings into 2,000% more sustainability coverage”
Quoted word for word from the source
18,556 listingsscoring throughputfrom 267 listings on their busiest single day
“6,850% more scoring throughput”
Quoted word for word from the source
6.9 sectime to score backlog listings
“6.9 sec to score all 18,556 backlog listings”
Quoted word for word from the source
a single $49 prediction packcostfrom about $1,405 in per-call LLM pricing
“96.6% lower cost than standard per-call LLM pricing, across the full 113,398 listing catalogue.”
Quoted word for word from the source
69.8%accuracyfrom 42.6% for always guessing the most common grade
“Accuracy 69.8% Against 42.6% for always guessing the most common grade.”
Quoted word for word from the source
0.705Macro F1
“Macro F1 0.705 Balanced across all three grades, including the smallest.”
Quoted word for word from the source
6 min 30 sectraining time
“6 min 30 sec”
Quoted word for word from the source
Difficulties and limits
- An LLM bill grows with the catalogue. A trained model does not. Every listing sent through the LLM adds to the invoice, priced per call, with no ceiling.
- Every miss landed on the grade right next door, never mistaking a top-rated listing for a bottom-rated one.
Sources
- easydeploy.ai — vendor self-report, 2026-08-04
Similar deployments
GE Aerospace
Independently reportedManufacturing · Process automation · Classical ML · Scaled
Apollo Tyres
under 10 minutes — Time for root cause analysis
Independently reportedManufacturing · Process automation · AI agents · In production
Broadcom · Moveworks
88% — autonomous IT resolution rate
Independently reportedTechnology & software · Process automation · NLP · Scaled
icare · SAS
increased by 15% — accuracy
Independently reportedInsurance · Process automation · Classical ML · Scaled · 2019
Amazon · Amazon
10% more — shopper spend
Independently reportedRetail & e-commerce · Process automation · Computer vision · Scaled · 2024