The following use cases are an illustrative example per sector. Each snapshot shows what the agent evaluates before acting, what it executes autonomously when it can do so with guarantees, and what the analyst receives when it cannot.

Prescriptive Decision AI for Utilities
Utilities
Renewable generation and demand forecasting

Forecast models produce a number. The trader does not know why the model is failing in that specific period, which signal is distorting the forecast, or how much it will cost to ignore it. In renewables, uncertainty is not symmetric: overestimating wind generation during low-pressure periods has a direct, quantifiable imbalance cost before it happens. AyGLOO adds an agentic layer on top of that model: the agent evaluates forecast reliability in that specific weather context, computes the expected cost of each possible bidding position, and adjusts automatically when it can do so with guarantees. When it cannot, it alerts the trader with full context and quantified exposure to decide in minutes.

1
The agent adjusts the bid automatically in periods where reliability is total. The trader intervenes only where meteorological uncertainty requires human judgement.→ Avoidable penalty €3,000–€4,000 per adjusted period
2
ISA detects hourly and weather micro-segments where the model systematically overestimates, using each segment’s historical MAPE as a double-check before acting.→ MAPE 14.6% in critical segment vs 4.2% portfolio average
3
CF shows which variable is distorting the forecast and how much it would change if that signal normalised — informing intraday hedging decisions before execution.→ Dominant signal identified before closing the position
4
The agent chooses between three actions — adjust the bid, recommend intraday hedging, or alert the trader — based on what minimises expected cost in each specific case. It does not apply the same response to all uncertainty.→ Strategic decision, not just a technical model correction
Today: what a trader typically receives
Forecast #FC-20260218-14h 342 MWh · MAPE: significantly above portfolio average Day-ahead market · Zone: Northwest · No additional context
With AyGLOO. Same forecast, fully enriched
Forecast #FC-20260218-14h · Zone: Northwest 342 MWh · Day-ahead market · Window: 18–22h The agent acts
XAI Why this forecast Twin Reliability-level rules ISA Model error segment CF Variable distorting the forecast Econ Economic decision function Action Agent decision PDF Regulatory traceability
1. Why this forecast. Twin model (depth 7 · fidelity 97.1%)
100%IF production_variable_A < seasonal_avg AND meteorological_indicator_B = true AND season = reference → Lower the bid automatically. Full reliability: act without manual review.
81%IF demand_variable_C > seasonal_avg AND consumption_pattern_D = true → Possible high demand. Review demand position before closing the bid.
59%IF temperature > seasonal_avg with no other factors → Weak signal in isolation. Do not adjust the bid on this variable alone.
A full-confidence rule triggers autonomous adjustment by the agent: it executes the bid correction without waiting for the trader. Lower-confidence rules are routed to the trader, with exposure already quantified to decide in minutes.
XAITwin
2. Model error segment (ISA)
Micro-segment: "late-evening hours, low pressure, winter" · Historical MAPE in this segment: 14.6% vs 4.2% portfolio average · The model systematically overestimates wind generation in low-wind periods · No drift detected in the last 8 weeks · High reliability: act with confidence.
ISA is the agent’s double-check before automatic adjustment: it confirms the error in this segment is systematic and stable, not a one-off noise. If the light were amber — drift detected or unstable pattern — the agent would not adjust automatically. It would route to the trader.
ISA
3. CF. Which variable is distorting the forecast
Remove low-wind signal: forecast increases +18.3 MWh · If wind recovered to seasonal average: forecast +14 MWh · If temperature normalised: demand −6 MWh · Wind is the dominant factor; temperature is secondary
CF is not a correction: it explains which variable is pulling the forecast down and how much it would change if that condition normalised. It informs intraday hedging decisions if the trader chooses to complement the automatic adjustment.
CF
4. Economic decision function. Expected cost by bidding position
Position
Bid MWh
Expected penalty
Saving vs no adjustment
No adjustment
342 MWh
€4,200–€5,000
estimated deviation without correction
—
Automatic adjustment ✓
323–326 MWh
–16 to –19 MWh
€800–€1,200
€3,000–€4,000
Intraday hedge
342 MWh + hedge
€300–€600
residual penalty
€X hedge premium
→ In this case the agent adjusts directly: maximum saving with no extra cost · If residual exposure exceeded hedge cost, the agent would choose intraday hedging
The agent does not optimise MAPE: it optimises expected penalty cost. The business metric is not forecast accuracy — it is the cost of deviation. That changes the objective function and, with it, the bidding decision.
Econ
5. Agent decision
The agent adjusts the bid — no trader intervention
The full-confidence rule holds and ISA confirms the error in this segment is systematic and stable · Executes bid adjustment: –16 to –19 MWh for the 18–22h window · Expected penalty reduced from €4,200–€5,000 to €800–€1,200 · Decision traceability exportable for regulatory audit
or, if direct adjustment does not cover exposure or the intraday hedge premium is lower than expected residual penalty
→ The agent recommends intraday hedging: residual exposure after adjustment exceeds premium cost · the agent computes optimal hedge volume and presents it to the trader ready to execute · final trader decision because it implies an active intraday market position
or, if reliability is lower or ISA detects drift in the model
→ The agent alerts the trader: forecast with elevated uncertainty, segment MAPE, CF with dominant variable, expected exposure in euros, and compared adjustment/hedge scenarios · the trader decides the position with exposure quantified in minutes
The agent has three paths, not two: adjust directly when error is systematic and adjustment covers exposure; recommend hedging when it does not; and alert the trader when uncertainty requires human judgement. In trading this is crucial: the difference between bid adjustment and intraday hedging is not technical — it is a strategic decision with distinct cost and market-position implications.
ActionPDF

Illustrative example. Each deployment is adapted to each client’s models, data, and operating procedures.

Estimated impact · DA/ID markets
€3–4K
Saving per automatically adjusted period
Difference between penalty without adjustment (€4,200–€5,000) and with adjustment (€800–€1,200) · directly attributable to each agent decision
−65%
MAPE in high-error segments identified by ISA
Systematic correction in micro-segments where the model overestimates wind generation
80%
Bids resolved with no trader intervention
The trader acts only where weather uncertainty or hedging strategy requires human judgement