AI-led products for Iberian energy operations
A suite of AI-led products built directly against the three operational pillars. Forecasting that learns from your book. Execution that runs against OMIE and MIBEL with the trader in the loop. Reconciliation that closes the month inside two working days. Each product ships as a working system, deployed in your tenancy, operated jointly.
The advisory pillars tell you what good operations look like. These pages cover the products that actually deliver it. Each product is built against a specific operational workstream, ships as a working system rather than a deliverable, and runs in your tenancy rather than ours.
The 13 products map one-to-one to the 13 sub-topics in the advisory pillars. Three suites, organised by pillar.
Why an AI-led delivery model
Three reasons.
Unit economics that work. A 500 GWh retailer running demand forecasting, guarantee monitoring and invoice reconciliation as people-driven workstreams typically spends €300k to €600k a year on the function. The same workstreams run on AI-led products against the same operational standard cost a fraction of that, with better discipline and a defensible audit trail.
Operational depth the team does not need to hire. The architecture for an OMIE execution layer, an agentic invoice reconciliation engine, or a regulatory radar is non-trivial. Most retailers under 1 TWh do not have, and cannot economically hire, the depth of engineering to build these themselves. A working product sidesteps the question.
A control surface the regulator will accept. Every product ships with a model and agent registry, an authorisation policy enforced at the runtime boundary, and an immutable audit log. The regulatory posture is built in, not bolted on. See A reference architecture for agentic AI in the regulated enterprise for the underlying pattern.
The three product suites
AI-led purchasing products
Four products covering the energy purchasing pillar.
- Forecast. AI-led hourly demand forecasting, calibrated to the retailer's customer mix and the Iberian weather and calendar.
- Execute. Agent-mediated execution against OMIE day-ahead, MIBEL intraday, MEFF futures and the GdO secondary market.
- Margin. Real-time trading guarantee monitoring across OMIE, MEFF, REE and any cross-border CCP positions.
- Reconcile. AI-led invoice reconciliation against DSO ATR settlements, REE balancing, OMIE clearing and MEFF mark-to-market.
AI-led operations products
Four products covering the operational management pillar.
- Switch. AI validation and exception handling on the SIPS data exchange with the eight Spanish DSOs and Portuguese e-Redes.
- Office. AI-augmented back-office covering invoice generation, direct debit dunning, bad-debt management and CRM hygiene.
- Serve. Customer service co-pilot tuned to the three Iberian retail customer populations (residential PVPC and free-market, SME bilateral, corporate structured).
- File. Regulatory filing assistant covering the MITECO, CNMC, REE and OMIE submission calendar, plus the DGEG and ERSE equivalents.
AI-led analytics products
Five products covering the analysis and pricing pillar.
- Price. AI-augmented pricing platform with scenario engine across the MEFF forward curve, customer mix shifts, weather and regulatory scenarios.
- Brief. AI-generated weekly market briefing covering OMIE, MIBEL, MEFF, GdO and cross-border flows, plus monthly outlook with explicit hedging recommendations.
- Board. Analytics platform integrating the customer, purchasing, regulatory and financial data domains into the dashboards the management team actually uses.
- Radar. Regulatory monitoring agent covering MITECO, CNMC, REE, OMIE, EU directives, plus ERSE/DGEG. Same-day notification, structured impact assessment, implementation tracking.
- Report. Bespoke AI reporting for the retailer-specific needs (PE covenant packs, sustainability-linked loan reporting, counterparty credit committees, green-finance verification).
The underlying architecture
Every product in the suite is built against the same five-layer reference architecture.
Layer 1: foundation models. Substitutable. Azure OpenAI, Anthropic, Mistral, or a sovereign EU deployment depending on the data residency requirement. The product does not lock you to a specific vendor.
Layer 2: model serving. Deployed in your hyperscaler tenancy in the EU jurisdiction your regulatory posture requires. Not in our tenancy. Your data, your audit trail, your control plane.
Layer 3: agent runtime. A bespoke runtime built on FastAPI, Postgres and the relevant queue infrastructure. Owned by the architecture function, not by a vendor.
Layer 4: tool gateway. Mediated integration into your existing systems: OMIE SIOM, MEFF clearing, REE balancing platforms, DSO data exchanges, CNMC filing portals, your CRM and billing engine. Policy enforcement, audit logging and rate limiting at the gateway.
Layer 5: domain products. The 13 modules above. Each scoped to a specific operational workstream, with a defined authorisation envelope and a named accountable senior manager on your side.
For the full reference architecture, see A reference architecture for agentic AI in the regulated enterprise, Platform strategy for agentic AI and Cyber guardrails for AI agents in regulated workflows.
How engagements work
Three engagement models, depending on what the retailer wants to own.
1. Build for you. I build the product in your tenancy. You operate it. Engagement length depends on the product (typically 6 to 16 weeks). Fixed scope, fixed price. After delivery, an optional 90-day support window. After that, you own and operate.
2. Build and operate jointly. I build the product in your tenancy. We operate it together. You run the business-day workstream; I keep the model and integration layer current, handle the regulatory changes, and carry the on-call. Monthly retainer.
3. Reference design only. You build it yourself, guided by the reference architecture. I review at defined milestones. Sits well for retailers with strong in-house engineering capacity who want the architecture pattern without the build.
The three models can be mixed across the 13 products. Most retailers start with one or two products under model 2 to demonstrate the operating model, then extend.
Why I am credible to build these
I am Head of Architecture at Sonnedix, an Iberian-active solar IPP. The systems my function builds and operates sit against OMIE, MEFF, the REE balancing platforms and the DSO data exchanges every working day. I have personally built two production AI-native enterprise platforms in the last 24 months: Meridian (the EA platform) and CANVAS (the application and vendor approval workflow). Both are running in production at Sonnedix with full audit trails, real users and real commercial outcomes.
The products covered on these pages are a separate engagement track from my day role. The technical patterns carry across; the client relationships do not.
How to start
The right starting point is a diagnostic across the three pillars. Two weeks, fixed fee, on-site or remote, with a written assessment of where the operation benefits most from an AI-led product layer.
Send me a message on LinkedIn or via the contact form on the homepage. Mention the pillar most pressing for you, the size of your book or portfolio, and a couple of sentences about your current operating model.
I will come back within one working day with availability and any clarifying questions.