What Clients Say
About Working With Us
Accounts from businesses that have completed Damai Logic engagements — what the work involved, what was delivered, and how it was received by the teams operating the systems.
Back to HomeRecent Accounts
We engaged Damai Logic for a demand forecasting system across our twelve distribution points. The thing that stood out was the scope document — it was specific enough that we could have evaluated the deliverable against it without any interpretation. The uncertainty ranges were a change from what we expected; our team now uses them as part of the weekly planning conversation rather than just looking at the point forecast.
The anomaly detection engagement addressed a specific problem with our production line telemetry — we were getting too many alerts from a rule-based system and most of them were not actionable. Damai Logic worked with our maintenance team to define what actually mattered before building anything. The false-positive rate they documented turned out to be accurate once we ran it live for a month. There were some edge cases in the definition phase that took longer to resolve than expected, but the process was transparent about it.
We brought in the Reading Group for our architecture team of seven. The pre-reading format worked well — sessions were genuinely discussion-led, and the practitioner who led them was direct about where current methods have limitations. The session on evaluation methodology was the most practically useful; several of us had been accepting vendor benchmark figures without understanding what they actually measured.
Cash position forecasting was something we had tried to address with spreadsheet models but they were becoming unmaintainable. The Damai Logic engagement produced a system our finance team could run on their own after about a week of familiarisation. The written documentation is thorough — more thorough than I would have asked for, but that turned out to be valuable when one of our team members who'd been on the handover left the company three months later.
Infrastructure telemetry anomaly detection was a problem we had been meaning to address for two years. The engagement ran slightly over the four-week estimate because our logging infrastructure was less consistent than we had described in the scoping conversation — Damai Logic flagged this early and we agreed on a revised timeline. The final system catches the class of incidents we defined and the alert volume is manageable. Nothing fancy, but it works reliably.
The Reading Group gave my leads a grounding in RAG and fine-tuning that has been directly applicable to decisions we've had to make about our product direction. Before the programme, the team was relying on secondhand summaries of conference talks. The governance session was particularly useful — it was honest about what current regulatory frameworks do and don't require, which helped us calibrate our own risk discussions.
Selected Engagement Summaries
A logistics company with twelve regional distribution points needed weekly demand forecasts to plan vehicle allocation. Their existing approach was a moving average applied by a warehouse supervisor each Monday — useful but not calibrated to seasonal variation in the region.
We examined two years of weekly shipment records and assessed the strength of seasonal patterns. A seasonal decomposition model was evaluated against the moving-average baseline on held-out data from the previous six months. The model improved on the baseline; the team was briefed on the retraining schedule for the following year.
The system has been running for eight months without a retraining cycle being required. Mean absolute percentage error on held-out data was 11%, compared to 19% for the previous moving-average approach. Vehicle allocation decisions are now made with the forecast range visible alongside the point estimate.
"The uncertainty ranges changed how we communicate internally about planning decisions. We stopped arguing about whether the forecast was right and started discussing what to do given the range it showed." — Operations Manager
A precision parts manufacturer had a rule-based alert system that generated roughly 140 alerts per shift — so many that the maintenance team had stopped acting on most of them. They needed a system that would surface genuinely anomalous readings at a volume the team could investigate.
We reviewed twelve months of production line telemetry alongside the maintenance log. Together with the maintenance supervisor, we defined four classes of anomaly that had historically preceded downtime. The detection system was tuned to achieve a stated false-positive rate of less than three alerts per shift for those specific classes.
Alert volume dropped to approximately two per shift. The maintenance team now investigates every alert that comes through. In the first ten weeks of operation, the system flagged two incidents that were caught and addressed before causing downtime. One class of anomaly proved harder to detect than initially expected and was noted in the limitations document.
"When we went from 140 alerts to two, the team started trusting the system instead of ignoring it. That trust took a few weeks to build, but it's now part of how the shift operates." — Maintenance Supervisor
A financial services firm's engineering team of nine — including four engineering leads — was being asked to evaluate vendor proposals for ML-based fraud detection without a shared vocabulary or framework for doing so. The CTO wanted the team equipped to engage technically rather than relying solely on vendor demonstrations.
The six Reading Group sessions were held fortnightly over three months. Topics included embedding spaces relevant to transaction data, evaluation methodology with specific reference to imbalanced classification problems, deployment patterns, and regulatory considerations under BNM's existing guidance. Pre-reading was drawn from published papers and technical documentation.
The team subsequently issued a structured vendor evaluation framework drawing on the evaluation methodology discussed in session three. Two of four vendor proposals were declined on technical grounds that the team could articulate specifically. One team member noted that the session on evaluation had changed how they interpreted the performance figures in the RFP responses they received.
"Our team went from nodding through vendor demos to asking specific questions about evaluation methodology. That shift took about three sessions to become visible." — CTO
Reach Damai Logic
Have a Question That Data Might Help Answer?
A brief initial conversation about your operational situation costs nothing and takes thirty minutes. From that conversation we can usually tell whether a quantitative approach makes sense and what it would involve.
Book a Discussion