These are accounts from people who have been through an engagement with us. Not every project goes perfectly — but every client knows what happened and why.
Back to HomeProjects delivered
Client satisfaction rate
In the Malaysian market
Sectors served
Dates reflect when the engagement concluded. Names and details shared with permission.
"We had customer feedback data going back three years that nobody had the time to read properly. Cognitara built a sentiment system that runs automatically now. The discovery phase took a bit longer than expected because our data was messier than we'd realised, but they were upfront about it and adjusted the timeline accordingly."
Lim Wei Xuan
Head of CX, Retail Group — KL
March 2026
"Our quality inspection team was doing manual visual checks on packaging — about 400 units an hour. After the vision integration went live we're running at 1,800 units with fewer misses. The camera consultation piece was something I hadn't expected to be useful, but it saved us from a positioning mistake that would've created problems downstream."
Norzaida binti Faridah
Operations Manager, Manufacturing — Shah Alam
February 2026
"The predictive model they built for our procurement planning has genuinely changed how we approach quarterly stock decisions. What surprised me was how much time they spent explaining the model logic rather than just handing over a dashboard. My team actually understands what it's doing, which means they trust it."
Arjun Krishnamurthy
Head of Supply Chain, Distributor — Petaling Jaya
April 2026
"I'd been burned by an AI project before — over-promised, under-delivered, no documentation left behind. Cognitara was different from the first call. They asked questions that made me think they actually cared about whether this would work, not just whether they'd get the contract. The NDA was signed before we shared anything."
Zulaikha binti Hashim
Digital Lead, Financial Services — KLCC
March 2026
"We engaged Cognitara for a document classification project — our compliance team was manually sorting hundreds of contracts a week. The model isn't perfect, maybe 91% on edge cases, but they told us that upfront rather than inflating the numbers. For the volume we handle, it's already saved us significant review time."
Danny Tan Boon Kheng
Compliance Director, Professional Services — KL
January 2026
"What I appreciated most was that they were willing to tell us our data wasn't ready for a predictive model yet. They recommended a three-month data collection phase first, which delayed our timeline but meant the engagement actually worked when we came back. I've seen vendors push ahead regardless — that's not what happened here."
Siti Yusra bt Azman
Analytics Manager, Logistics — Selangor
February 2026
These case summaries describe what was attempted, what was built, and what the outcome looked like.
A regional bank's contact centre was receiving 2,000–3,000 customer messages per day in a mix of English and Bahasa Malaysia. Manual triaging meant routing delays averaging 4 hours. Staff morale was suffering from repetitive categorisation work.
An intent classification and routing model fine-tuned on 18 months of historical message data, covering 23 intent categories. Integrated directly into the bank's existing ticketing platform via API. The model also flags messages requiring human review above a confidence threshold.
Average routing time dropped from 4 hours to under 12 minutes. Approximately 78% of messages are now handled without any manual routing step. Staff were redeployed from categorisation to resolution. Engagement duration: 7 weeks.
"The confidence threshold feature was their idea — it meant we weren't just automating blindly. That detail made a real difference to the team's trust in the system."
A Penang electronics manufacturer was seeing a 3.2% defect escape rate at end-of-line inspection — defects identified only after packaging. Manual inspection at speed was inconsistent, particularly for micro-solder defects invisible to the naked eye at typical line speeds.
A custom computer vision model trained on annotated images of acceptable and defective units. Camera positioning was redesigned as part of the consultation phase. The model runs on edge hardware at line speed and triggers an automatic divert gate for flagged units. Training dataset: 12,400 annotated images.
Defect escape rate reduced to 0.4% over the first three months of operation. False positive rate held at 1.8%, accepted by the production team as within tolerance. Engagement duration: 9 weeks.
A mid-sized FMCG distributor was managing inventory for 340+ SKUs across 5 warehouses using spreadsheet-based forecasting. Overstocking and stockouts were both common, tying up capital and causing service failures with key retail accounts.
A demand forecasting platform covering all SKUs across all warehouses, incorporating historical sales, promotional calendar, and seasonal patterns. A planning dashboard was built to surface reorder recommendations alongside confidence intervals, connected to the client's existing ERP system.
Stockout events fell by 34% in the first quarter post-deployment. Overstock value reduced by approximately RM 280,000 across the warehouse network. The planning team reports spending significantly less time on manual forecasting and more on exception management. Engagement duration: 11 weeks.
We're always open to a conversation about what you're working on. Even if it's early days and you're not sure what AI involvement would look like, we're happy to help you think through it.
Let's start with a discovery call. No commitment, no sales pressure — just a conversation about your situation.
Book a Discovery Call