CLIENT EXPERIENCES
What organisations
say after working with us
Accounts from clients who have been through the scoping process, the build, and the handover — including what worked and what they would have done differently.
CLIENT REVIEWS
Direct accounts
Farah Hanim
Head of Operations · Kuala Lumpur
We brought Mutiara AI in to help with our claims form extraction — we had two staff spending most of their day on manual data entry from PDFs. The system they delivered handled around 78% of forms without human intervention in the first month. The remaining 22% were flagged for review, which is roughly what they told us to expect based on the forms we sent them upfront. The handover documentation was better than we anticipated.
April 2025 · Document Intelligence System
Raj Nambiar
IT Manager · Petaling Jaya
The IT helpdesk assistant has been running since March. It handles password resets, VPN troubleshooting steps, and software access requests without involving our team. I'd say it handles about 60% of the ticket volume. The post-launch tuning period was useful — there were some query types we hadn't thought to test for. I appreciated that they communicated when something wasn't going to work rather than just trying to make it fit.
March 2025 · Conversational System Build
Lim Mei Yen
Risk & Compliance Lead · Shah Alam
We needed an independent view of a credit scoring model that had been in production for two years. The audit report was structured, specific, and identified three issues we hadn't considered — all related to how the training data was assembled. The severity ratings were clear, and the remediation plan gave us a realistic sequence to work through. I've seen internal reviews that were less thorough.
April 2025 · Model Audit and Risk Review
Zuraida Abdul Rahman
General Manager · Subang Jaya
The scoping process took longer than I expected — they spent a full week reviewing our documents before agreeing to the work. In hindsight, that was the right approach. The system they built performs well on the document types we discussed. The things that don't work as well are the ones that were outside the agreed scope, which they told us about beforehand. I'd recommend them for organisations that have a specific, bounded problem.
March 2025 · Document Intelligence System
Kelvin Sim
Head of Legal Technology · Kuala Lumpur
We process a high volume of commercial contracts each month. The extraction system reduced our data entry workload substantially — we now have a small review queue rather than full manual processing. They were clear that the accuracy would be lower on older, lower-quality scans, and that was accurate. The evaluation report they provided before deployment has been a useful reference during our internal monitoring.
April 2025 · Document Intelligence System
Nadia Amirah
Customer Experience Manager · Selangor
The customer triage assistant handles our initial product enquiries. It's honest with customers when it doesn't know something, which was a specific requirement we had. The three months of tuning caught some phrasing issues that the test set didn't surface. We're now looking at whether a second use case makes sense, and the team has been straightforward about whether that would be a good fit.
March 2025 · Conversational System Build
CASE STUDIES
Engagements in detail
Invoice and purchase order extraction for a logistics firm
CHALLENGE
A mid-sized logistics company processed approximately 2,400 invoices and purchase orders per month across seven different supplier formats. Three staff were allocated full-time to data extraction, and error rates from manual entry were creating downstream reconciliation issues.
WHAT WE DID
We analysed the seven document formats over two weeks and identified three that were well-suited to automated extraction and four with inconsistent formatting. We built a system that handled the three cleanly, flagged the four format types for human review, and delivered with full documentation covering which document characteristics drove lower confidence scores.
RESULTS — 10 WEEKS
- 68% of invoices processed without human intervention
- Manual review time reduced from ~3 FTE to ~1 FTE
- Extraction error rate on automated documents under 2%
"They were honest that the four inconsistent formats would need human review. That transparency was actually helpful for our planning." — Operations Director
Governance review of an automated loan assessment tool
CHALLENGE
A consumer finance firm had deployed a loan assessment model three years earlier. As the regulatory environment tightened, the compliance team needed an independent view of whether the model's evaluation regime and monitoring were adequate, and whether the training data still reflected current applicant characteristics.
WHAT WE DID
Over two weeks, we reviewed training data provenance, the evaluation methodology used at deployment, current production monitoring, and the operational processes surrounding model outputs. We produced a written report with severity ratings and a sequenced remediation plan — without being engaged to carry out the remediation itself.
DELIVERABLE
- Written audit report, 28 pages
- Five material issues identified, three rated high severity
- Sequenced remediation plan with estimated effort per item
"The report gave our board something specific to act on, rather than a general list of AI governance concerns." — Chief Risk Officer
REACH US
Contact details
Phone
+60 16-732 4859
Address
67, Jalan SS 21/35
47400 Petaling Jaya, Selangor
Hours
Mon–Fri: 9am–6pm
Sat: 9am–1pm MYT
Ready to start a conversation?
A short scoping call is the most efficient way to work out whether your situation maps to one of our services. No commitment required — just an honest exchange about the problem.
Get in Touch