Client voices
What clients say about
working with Aethernet
Unedited feedback from engagements across financial services, SaaS, and operations teams in Malaysia.
← Back to Home47+
Engagements
4.8
Avg rating / 5
94%
On-time delivery
6yr
In operation
Client reviews
From recent engagements
"The churn model they delivered flagged a segment we'd been missing for over a year. Within two months of using the output, our retention team had a 22% improvement in reactivation rate. The documentation made it easy for our internal team to take over."
Lim Zhi Wei
Head of Growth — SaaS platform, KL
March 2025
"We needed monthly compliance reports that used to take three days to assemble manually. Aethernet built a system that generates them in under ten minutes. The narrative sections read naturally — they didn't just dump raw tables."
Nurul Rashidah
Operations Manager — Financial services, Selangor
February 2025
"The transfer learning consultation was very useful for our team. We had a classification problem but only 400 labelled examples — Reza walked us through model selection clearly and the fine-tuned model performed substantially better than what we'd tried ourselves."
Tan Kai Sheng
Data Team Lead — E-commerce, Penang
January 2025
"Communication was the best part — one point of contact throughout, honest updates when there were data issues, and no surprises on scope or price. The final handover session was thorough and our analysts understood the system straight away."
Aiman Putera
VP Analytics — Telecom operator, KL
March 2025
"We'd had a bad experience with a larger vendor who delivered a black-box model we couldn't explain internally. Aethernet was the opposite — we understood exactly what signals the model was using and why. The documentation meant we could brief our board with confidence."
Sarah Wong
COO — Insurance intermediary, Johor Bahru
February 2025
"What struck me was that they told us upfront that our data for one part of the project wasn't adequate. Other vendors would have charged us to build something that wouldn't work. That honesty made us trust them for the rest of the engagement."
Mohd Hafizuddin
Director — Professional services firm, KL
January 2025
Case studies
Three engagements in detail
Case study 01
Churn model for a B2B SaaS platform
Challenge
A 60-person SaaS company was losing approximately 18% of accounts annually without visibility into which customers were at risk. Their CRM held usage data but no team had the capacity to analyse it systematically.
Solution
Aethernet built a churn prediction model using 14 months of login, feature usage, and support interaction data. The model was integrated directly into HubSpot, flagging at-risk accounts weekly with an explanation of the primary signals driving the score.
Results
After 90 days of operation, the retention team reported a 19% reduction in churn rate among flagged accounts. The model achieved 78% precision on held-out test data — substantially above the baseline. Delivered in 7 weeks.
Case study 02
Automated regulatory reports for a financial intermediary
Challenge
A licensed financial intermediary was producing monthly regulatory submissions manually — a three-day process involving multiple staff members pulling data from separate systems and formatting it into a specific template.
Solution
Aethernet built a pipeline that connects the client's core banking system and transaction database, processes the required data, and generates the full report — with narrative sections — as a PDF on a scheduled basis.
Results
Report preparation time dropped from 3 days to under 15 minutes. Staff redirected to higher-value work. The system has run without intervention for 5 months since handover. Delivered in 8 weeks.
Case study 03
Document classification for a legal services firm
Challenge
A mid-size legal firm needed to automatically classify incoming client documents into 12 categories. They had only 380 labelled examples — far too few to train a model from scratch effectively.
Solution
Using transfer learning consultation, Aethernet identified a multilingual transformer model, adapted it to the firm's document vocabulary, and fine-tuned it on their 380 examples with careful regularisation to avoid overfitting.
Results
Final model achieved 91% accuracy on a held-out test set, compared to 63% using a conventional TF-IDF approach. Client's team can independently fine-tune as they accumulate more labelled data. Delivered in 4 weeks.
Your turn
Ready to talk about your data?
Bring a problem. We'll give you an honest read on whether and how we can help.
Contact Aethernet