Five programmes delivered under Ivan Teh's leadership, described by the problem they were brought in to solve rather than by the technology used.
Retail and consumer goods
A multi outlet retail chain was ordering against last season's pattern and marketing to its entire base with the same message. Stock sat in the wrong stores while other outlets ran dry, and promotional spend was diluted across customers who were never going to convert.
Demand forecasting was rebuilt on transaction level history combined with local signals, and customer segmentation was derived from behaviour rather than demographics. Allocation decisions moved from a monthly cycle to a continuous one, with store managers given direct visibility of the recommendation and the reason behind it.
The organisation reported a twenty per cent increase in sales within six months, attributed to better availability and to promotions reaching the customers most likely to respond.
Healthcare
A hospital group had capacity data, diagnostic data and scheduling data sitting in systems that did not speak to each other. Bottlenecks were only visible after they had already formed, and clinical staff spent time reconciling records that should have reconciled themselves.
Patient flow and diagnostic data were brought into a single analytical layer with predictive load modelling on top. Scheduling was adjusted against forecast demand rather than historical averages, and diagnostic support surfaced relevant prior cases at the point of decision.
The group reported a thirty per cent reduction in patient waiting times alongside improved diagnostic accuracy, with better documented outcomes for patients.
Banking and financial services
A major bank was running fraud detection on static rules. The rules caught known patterns and missed new ones, and the false positive rate was high enough that genuine customers were being blocked, which carried its own commercial cost.
A machine learning detection layer was introduced alongside the rules engine, scoring transactions against behavioural baselines specific to each customer. The system was designed so that analysts could see why a transaction had been flagged, which was a condition of the risk function accepting it.
The bank reported a forty per cent decrease in fraudulent transactions in the first year, with stronger risk management overall and improved customer trust.
Education and medical training
A medical and health sciences institution could evaluate teaching effectiveness only retrospectively, once results were in. Students who were struggling were identified too late for intervention to help them.
A personalised learning analytics programme was developed in collaboration with IMU, applying analytics and artificial intelligence to learning behaviour so that both student experience and teaching effectiveness could be observed while a course was still running.
The programme gave educators a live view of engagement and comprehension, moving intervention from post hoc to in course.
Trade, associations and SMEs
Halal sector SMEs and technology sector members of a national association had commercial need for digital capability but not the capital to build it individually.
Two collaborative platforms were delivered: work with Alliance Islamic Bank's Halal in One programme gave halal SMEs access to shared digital tools and resources, and work with PIKOM produced a digital engagement platform serving more than a thousand member companies for product promotion, business meetings and virtual exhibitions.
Capability that had been out of reach individually became viable collectively, with the cost spread across the membership rather than carried by each business alone.
Several were delivered under confidentiality terms that cover the commercial detail. Where a programme has been publicly announced by the partner, such as the work with IMU, PIKOM and Alliance Islamic Bank, it is named.
The percentage outcomes are figures reported by the client organisations themselves following deployment. They are presented as reported, not as independently audited results.
In every one, the data already existed. The constraint was that it was fragmented, delayed, or presented in a form the decision maker could not use. The work was less about collecting more and more about closing the distance between signal and decision.