Predictive operations across an upstream asset at a Tier-1 national oil company.

At a glance.

All client identifying details are anonymised under NDA. A senior lead reference call is available to qualified prospective clients on request.

The client operates one of the largest upstream assets in the region. The asset's reliability function was strong on paper — a mature CMMS, a well-staffed control room, vibration and acoustic sensing on most rotating equipment — but unplanned shutdowns on compressors, pumps and turbines were costing the asset a material number of production hours per year. The reliability team had a backlog of analytical projects that had never reached production, mostly because the path from data-science notebook to operator-trusted alert had no owner.

The client had spent four years and a meaningful eight-figure budget with a global engineering-services vendor attempting to industrialise predictive maintenance. The output was a portfolio of disconnected dashboards that the control room did not use, because the alerts were noisy, the false-positive rate was unacceptable to a shift supervisor whose decision authority is binding, and the models had been built off-site by a team the operators had never met. The client had also been told repeatedly that the work could not be done with national engineers because the talent did not exist.

We mobilised a twenty-two-person pod into the asset's operations centre for fourteen months. Eleven of the engineers were Qatari nationals — six recruited directly from Qatar University through the wesolutions Academy intake, and five hired into the firm and seconded to the engagement. The pod reported to the asset's vice president of operations, not to its IT function. We rebuilt the alerting stack from first principles: a small number of high-signal models per asset class, an explicit suppression logic owned jointly by reliability engineers and shift supervisors, and a closed feedback loop in which every alert acknowledged by the control room becomes a labelled example. The pod ran weekly working sessions with the shift supervisors for the full duration. The system was deployed into one compressor train in week fourteen and expanded to the full asset over the following nine months.

Unplanned downtime on instrumented assets fell by thirty-one percent over the engagement, translating to an annualised production uplift of approximately USD eighty-six million as audited by the client's finance function. Four hundred and twelve critical assets are under predictive control at handover. The shift supervisor population — who at the outset were the most sceptical constituency — became the system's strongest internal advocates, principally because the suppression logic gave them ownership of the alert volume. Eleven of the Qatari engineers transferred onto the client's permanent payroll at the end of the engagement, by mutual agreement, and now run the platform internally. The remaining wesolutions engineers stepped down to a small steady-state support pod.

Three. First, the failure mode of the prior vendor was organisational, not technical: building models off-site for an operations team you have never met is structurally incapable of producing operator-trusted alerts. Second, the binding constraint on predictive maintenance is alert-suppression governance, not model accuracy — once the shift supervisors owned the suppression logic, the rest followed. Third, Qatari engineering talent at the required level absolutely exists, and the constraint was the firm willing to recruit, train and embed it; that is a deliberate choice the prior vendor had not made.