-
TL;DR • Project at a glance
Unified fragmented ops tooling so operators could see live position risk without reconciling across multiple systems.
Designed for speed and attention under pressure using progressive disclosure and deliberately constrained alerting.
Automated routine nominations selectively while keeping human judgement for exceptions and external failure modes.
Outcome: Intraday balancing dropped from ~4 hours per operator per day to < 30 minutes combined, alongside faster batch submission (~12 minutes → ~2 minutes) and a lower error rate (~15% → < 3%).
As part of the Power Operations transformation programme, collaborated with the Product Owner and engineering leads to define and deliver a unified position management and nominations platform, replacing fragmented legacy monitoring tools.
The platform was designed to support increasing trading volumes, tighter gate closures and evolving compliance requirements, while reducing operational dependency on disconnected systems and manual reconciliation.
— The Challenge
Ten minutes to fix the problem — or pay for it
At 2:15 a.m., a power operations operator noticed something was wrong.
To understand the issue, they had to reconcile data across three separate systems, cross-check spreadsheets, and refresh a regulator portal — all while the clock was ticking. If the imbalance wasn’t corrected within ten minutes, the batch would be rejected and penalties would follow.
This wasn’t an edge case. It was routine.
As trading volumes increased and regulation pushed the market towards finer time granularity, operational risk was amplified by fragmented legacy tooling. Operators weren’t failing because they lacked expertise — they were failing because systems made it impossible to see and act fast enough.
The problem wasn’t speed or scale.
It was decision-making under pressure, with incomplete visibility.
“At 2:15 a.m., I'd be copying data between spreadsheets and an old scheduling tool while refreshing a regulator portal. If anything went wrong, I had 10 minutes to fix it before penalties kicked in.” —Senior Operator, Power Operations
— What success had to look like
If operators only had minutes to prevent penalties, the system had to do three things — reliably and without hesitation.
- Make risk visible immediately.
Operators needed to see imbalances and anomalies the moment they emerged, without searching across tools or reconciling data manually. - Focus attention on what actually mattered.
In high-pressure moments, the system had to reduce noise — not add to it — so operators could act decisively instead of second-guessing. - Remove manual work from the critical path.
Routine tasks, particularly nominations, had to be automated so human effort was reserved for exceptions and judgement.
Success wasn’t measured by feature delivery.
It was measured by whether an operator, at 2 a.m., could understand the situation and act before the clock ran out.
— Approach adopted
Designing for people, not just processes
- Co-designed with operators and supervisors through workshops and shadowing to uncover hidden manual work and failure points.
- Used sandbox environments so users could test safely and build trust before rollout.
- Ran weekly sprint demos to keep delivery aligned to real operational needs.
— Solution overview
A phased transformation
Phase 1 — Live position management
A single interface for intraday and day-ahead positions, with progressive drill-down so operators could move from system health to specific trades without losing context. Visual cues flagged imbalances and anomalies early enough to act before penalties.
Phase 2 — Nominations automation (selective by design)
Nominations were the biggest operational bottleneck. The obvious response was full automation. I resisted that.
In cross-border power trading, failures can originate outside internal systems. Full automation would have removed human judgement precisely when it mattered. We automated only what was predictable and verifiable: routine cases flowed automatically; exceptions required explicit human confirmation.
Automation was capped at around 75%, by design. The goal wasn't throughput. It was trust.
“Now, my team focuses on the issues that matter most, instead of getting lost in hundreds of alerts.” — Supervisor, Cross-Border Operations
— Design decisions
Balancing scale, speed, and clarity
Why we optimised for speed over completeness
From the outset, operators asked for everything: full trade history, all counterparties, every time slice, every reconciliation state — in one place.
Technically, that was achievable. Operationally, it would have been dangerous.
In peak trading windows, the cost of a slow or cluttered interface isn't inconvenience — it's missed action. Rendering thousands of rows with full detail increased latency and made it harder to identify what actually required intervention.
I chose to prioritise speed and clarity over completeness. Summary views surfaced system health first. Detail was available on demand through drill-down, filtering, and aggregation — never by default.
This meant accepting a trade-off: some information was hidden by default, in exchange for instant response under load.
From that point on, if a view couldn't remain fast and legible at peak volume, it didn't ship.
Consistent, scalable UI components
Multiple modules needed to work together seamlessly and expand over time. We standardised core components to keep interaction patterns consistent and reduce long-term maintenance overhead.
Why we deliberately limited alerts
In early prototypes, we exposed every detectable issue in real time. On paper, this seemed safer. In practice, it created a new risk.
Operators were seeing dozens of simultaneous alerts during peak windows. Instead of improving awareness, the system increased cognitive load — critical issues were being lost in noise.
The problem wasn't alerting accuracy. It was attention management under pressure.
I made the call to limit what the system surfaced, even when more data was available. We introduced tiered alerting (Critical, Warning, Informational) and constrained the interface to highlight only the three most urgent issues at any moment.
The trade-off was intentional: some information is de-emphasised by default, so operators can act decisively without hesitation.
From that point on, every screen had to earn the operator's attention.
— Metrics & outcomes
Consequences of decisions
We didn’t optimise for metrics. We optimised for judgement under pressure. The numbers improved because the system made better decisions easier to execute.
Intraday balancing improved because we prioritised speed and progressive disclosure.
Nomination effort fell because we automated selectively and kept humans in the loop for exceptions.
Batch submission time dropped because alerts were constrained to what deserved attention.
Error rates fell because issues became visible early enough to resolve before penalties applied.
| Task | Before | After |
|---|---|---|
| Intraday balancing | ~4 hrs/operator/day | < 30 mins combined |
| Nomination processing | ~70% manual | > 75% automated |
| Batch submission time | ~12 minutes | ~2 minutes |
| Error rate | ~15% | < 3% (mostly external) |
— Strategic outcomes
Preparing for what comes next
Beyond efficiency, the platform positioned EDF Trading for regulatory and volume-driven change.
- Regulatory readiness: Prepared for EU 15-minute granularity requirements.
- Legacy retirement: Three outdated tools sunset, reducing maintenance overhead.
- Scalability: Built to handle rising volumes as trading strategies evolve.
“We've gone from firefighting to forward planning. The platform doesn't just help us keep up — it lets us lead.” — Head of Power Operations
