What Clients Say About Working with Novalith

We were processing several hundred customer tickets a day in a mix of Bahasa Malaysia and English, and routing them manually. Novalith built a classification model that handles both languages properly. It took about twelve weeks from first conversation to deployment. The improvement in routing accuracy was meaningful — we'd tried a generic solution before and it genuinely couldn't cope with the language mix.

February 2026 · Language Model Development

The sentiment platform they built for us monitors brand mentions across a few social media channels. I was sceptical about how well it would handle informal Bahasa Malaysia — people write very differently online compared to formal text — but the calibration work they did made a real difference. I do wish the initial discovery phase had been a bit longer, but they were responsive when we flagged things that needed adjustment post-delivery.

January 2026 · Sentiment & Text Analysis

We were entering invoice data manually from three different vendor formats. The extraction system Novalith built now handles all three formats and outputs clean structured data into our finance system. The accuracy rate is around 96% — the remaining edge cases are mostly vendor-specific formatting quirks we'd agreed up front might need manual review. Saved us real time each week.

March 2026 · Document Extraction

What I appreciated most was the pre-delivery evaluation report. They were upfront about a couple of document types where the model wasn't as strong, and we made a decision together about how to handle those. Most vendors would have glossed over that. The documentation they provided was also genuinely useful — not just a README.

December 2025 · Document Extraction

We run customer feedback surveys in both Bahasa Malaysia and English, and analysing them consistently was always a challenge. Novalith's sentiment platform now processes both together. The topic extraction feature is particularly useful — we can see which service areas are generating negative feedback without manually reading through everything. Straightforward to work with throughout.

February 2026 · Sentiment & Text Analysis

Contract review is time-consuming work and we were looking at whether NLP could assist with extracting key clause information. Novalith were honest from the start about what was realistic — they didn't oversell. The extraction model they built handles standard commercial agreements well. For more complex bespoke contracts, it flags for human review rather than attempting extraction it's not confident about. That's the right design.

January 2026 · Document Extraction