Client context
Major European utility
Blackdown's role
BESS optimisation, reserve integration, AI explanation and governance
Commercial outcome
Measured increase in BESS revenues
Publication status
Generalised reference architecture
The exact revenue uplift, portfolio composition and optimisation methodology are commercially sensitive and are not reproduced here.
The challenge
Every battery is running several businesses at once.
A grid-scale battery may move between wholesale arbitrage, reserve availability, balancing opportunities and imbalance management within the same operating day.
These value streams compete for the same limited power, energy and state of charge. A profitable-looking discharge can remove the headroom needed for a reserve obligation. Charging into a negative-price period can create value while also weakening readiness for a later upward response.
The question is not: “Where is the highest price?”
It is: “Which combination of dispatch, availability and market exposure creates the highest risk-adjusted value while respecting every physical, contractual and operational constraint?”
That is a deterministic optimisation problem. But the result is useful only when the people accountable for the asset can understand and supervise it.
The failure mode
A good optimiser that nobody trusts will be overridden.
The weakness often sits between the optimisation engine and the human trading or dispatch team responsible for its output.
A trader sees that the battery is not discharging into a visible price spike, but cannot immediately see the reserve commitment, state-of-charge floor or later opportunity driving the decision.
The optimiser protects value or compliance that the desk cannot immediately see.
A trader changes the schedule without a clear view of the revenue, state-of-charge or reserve consequences.
The business cannot later determine which inputs, constraints, forecast assumptions or interventions produced the realised outcome.
Once recommendations are routinely changed without a shared constraint picture, the intended strategy and the actual operating strategy quietly diverge.
Combining optimisation, market integration and governed AI
Blackdown's contribution went beyond adding a conversational interface to an existing optimiser. The work connected the dispatch mathematics, the additional market service and the controls needed for people to understand and govern the result.
Optimisation development
Developed and improved deterministic logic for evaluating dispatch choices across market, battery and contractual constraints.
Revenue stacking
Modelled competing value streams together rather than treating arbitrage, reserve and balancing as isolated strategies.
Reserve integration
Incorporated an additional reserve service into the optimisation and operational decision process.
Explanation and governance
Designed the layer that surfaces binding constraints, explains commercial trade-offs and records informed human interventions.
The objective was not to make the optimiser appear intelligent. It was to make the dispatch decision commercially stronger, operationally clearer and easier to govern.
Reference architecture
Five layers with clear responsibility boundaries
01 — Inputs
Market and asset state
- Price forecasts
- Auction results
- Asset telemetry
- State of charge
- Availability
- Existing commitments
- Efficiency and degradation assumptions
02 — Deterministic
Deterministic optimisation
- Dispatch schedule
- Revenue allocation
- Physical constraints
- Reserve requirements
- Expected commercial value
03 — AI support
AI decision support
- Commercial explanation
- Binding constraints
- Opportunity cost
- Conflicts
- Uncertainty
- Desk brief
04 — Human
Human review and approval
- Accept
- Request analysis
- Propose override
- Record rationale
- Approve binding action
05 — Execution
Controlled execution
- Existing trading systems
- Scheduling systems
- Asset-control systems
- Established permissions
The AI layer does not have unrestricted write access to trading, scheduling or asset-control systems. It consumes approved optimiser inputs and outputs, prepares an explanation and supports a controlled human decision.
A dispatch brief, not a solver output
The decision-support layer should give an experienced trader or asset manager enough information to understand the proposed plan in less than a minute.
- Recommended plan
- Charge, hold, discharge or remain available by settlement period.
- Forecast value
- Expected contribution from wholesale, reserve, balancing and other modelled value streams.
- Binding constraints
- State of charge, reserve commitments, power limits, asset availability and degradation limits.
- Principal conflict
- The strongest competing use of the same battery capacity.
- Opportunity deliberately forgone
- The best alternative strategy and why it was rejected.
- What could change the decision
- Price movement, forecast uncertainty, auction result, revised asset availability or market instruction.
- Required action
- Accept the plan, review an exception, quantify an override or escalate a breach.
Example explanation
The battery is not scheduled to discharge during the 18:00 wholesale peak because upward capability is being preserved for the contracted reserve window. Releasing that headroom would increase forecast wholesale revenue but leave the asset unable to satisfy the reserve requirement under the current state-of-charge trajectory.
Override governance
Human intervention remains essential. Silent intervention does not.
Forecasts change. Traders may hold information that is not represented in the model. Asset conditions may deteriorate. The objective is not to eliminate overrides, but to ensure that the consequences are visible before an override is accepted.
Before — Original recommendation
- Original dispatch
- Expected revenue
- Reserve availability
- State-of-charge trajectory
- Degradation and cycling impact
After — Proposed override
- Proposed change
- Incremental forecast value
- Revised reserve position
- Revised state of charge
- New commercial or operational risk
Approval record
Repeated overrides become a feedback set. They may reveal missing desk information, biased forecasts or constraints that have been modelled too conservatively.
Turning a market specification into operating logic
A new reserve service is not simply another market-data feed. Its rules must become explicit, version-controlled optimisation constraints and operating controls.
Quick Reserve is used here as an illustrative GB market example. The same architectural pattern applies to other reserve and flexibility products.
Scenario testing
Measured client outcome
Higher BESS revenues
The delivered work produced a measured increase in BESS revenues.
The commercial result followed optimisation development and reserve-service integration in client operations. The exact uplift, asset population and proprietary optimisation methodology are not published.
An additional reserve service was incorporated into the optimisation and operational process.
Dispatch recommendations were supported by clearer explanation of constraints, conflicts and commercial trade-offs.
Human changes could be assessed against the optimiser's value and constraint picture rather than recorded as unexplained schedule changes.
The value did not come from generating a more polished summary of an unchanged dispatch decision. It came from improving the optimisation and market-participation capability underneath, then making that capability easier for the desk to understand and govern.
A common pattern for multi-market flexible assets
Although the architecture is framed around grid-scale batteries, the pattern applies wherever a deterministic model makes high-value decisions that people must understand, supervise and occasionally override.
Battery portfolios, hybrid renewable-plus-storage assets and other flexible generation or demand.
Wholesale, balancing, reserve, capacity and contractual-flexibility decisions.
Trading, scheduling, dispatch and asset-control workflows.
Commercial review, model governance, risk oversight and post-event analysis.
Does the optimiser and the desk speak the same language?
Blackdown connects the optimisation, market and human-control layers around commercially consequential battery decisions.
Discuss your battery portfolioReference-architecture disclosure
This is a reference architecture informed by BESS optimisation, reserve-service integration and AI decision-support work delivered for an anonymised major European utility.
The architecture has been generalised. Commercially sensitive client information, optimisation methods and performance figures have been omitted. Quick Reserve is included as an illustrative market example and should not be read as confirmation of the service used in the underlying engagement.