Inside Octopus Energy's flex auction

For three Tuesday evenings in March 2026, an automated bidding system at Octopus Energy ran the largest residential-flexibility auction yet conducted in the UK. By the end of the third window, 312 MW of distributed capacity had cleared at an average price of £4.18/kWh of dispatched energy, with a settlement floor of £1.00/kWh and a single window peak at £8.20/kWh.

That sounds esoteric. The plain reading is more interesting: half a million households agreed to shift around 9–18% of their evening electricity demand to a different time, in exchange for being paid roughly 20× the wholesale price for the energy they did not consume during a specified hour.

I sat down with two of the engineers who designed the auction and one operations lead who runs it during the dispatch window. What follows is the auction's anatomy and what its clearing pattern says about residential demand response in 2026.

What the auction actually procures

The auction does not buy generation. It buys commitments not to consume during a forecast-stressed hour, settled against each household's own counterfactual baseline (which Octopus calculates from the customer's own meter history, regressed against weather and day-type).

Each clearing window targets a one-hour delivery period 36–48 hours ahead. Octopus's trading desk decides the size of the call based on its imbalance position and ESO's day-ahead price curve. The auction itself is a uniform-price clearing — every accepted bid receives the marginal clearing price, regardless of where it bid in.

The bid stack

This is the part that gets glossed over. The bid stack is not made of individual household decisions; it is made of tariff cohorts auto-bidding under rules the customer set once and forgot about. The composition in March 2026 was approximately:

• Cosy heat-pump cohort — 91 MW. Highest reservation price (mean £3.10/kWh) because pre-heating cycles are expensive in terms of comfort. About 14% of total.
• Intelligent Octopus Go (EV) cohort — 162 MW. Lowest reservation price (mean £1.20/kWh) because delaying an overnight EV charge by an hour costs the household effectively nothing.
• Saving Sessions opt-in (no controllable load) — 47 MW. Bids only on prosocial sign-up; cleared as a price-taker.
• Battery cohort (Powerwall, Givenergy, Tesla) — 12 MW. Bid into the high tail; clears only in tight windows.

The reason the EV cohort dominates is structural: EVs can defer charging by hours at essentially zero customer cost, and an Intelligent Go customer has pre-authorised Octopus to shift charging windows freely within an overnight envelope. The "auction" for that cohort is, in practice, an automated rearrangement of when energy is delivered to the car — the customer sees a notification at most.

What the clearing price tells us

The £4.18/kWh average sits dramatically above the spot wholesale price during the same delivery windows (£0.08–0.21/kWh). That gap is the value Octopus is willing to pay because the alternative — buying short-notice imbalance energy via the Balancing Mechanism — costs them more, sometimes considerably more.

Two patterns from March:

1. Cold-snap windows priced higher. On 14 March, with overnight temperatures forecast to drop to −3°C in the south-east and ESO modelling a tight evening peak, the auction cleared at £7.30/kWh against 41 MW called. The marginal accepted bid was a Cosy heat-pump household; the inframarginal rent on the EV cohort that window was meaningful (~£190k aggregate).
2. Mid-week mild windows undersubscribed. On 22 March, with mild weather and abundant Scottish wind, the auction was called at 22 MW but received 67 MW of bids; the clearing price collapsed to the £1.00/kWh floor.

Both are doing exactly what an efficient flexibility market should do — price-discovery in the tail, price-taker behaviour in the body.

The questions this raises

Octopus is not the only retailer running residential flex, but at 312 MW cleared it is now a settlement-relevant slice of GB system flexibility. That has three implications worth flagging:

One — DNO-level optimisation. The auction is currently optimised for ESO-level (national) signals: wholesale price, imbalance volume. It does not yet co-optimise against DNO flexibility tenders procured at the 33 kV substation level. Greg Jackson has said publicly that "stacking DSO flex on top of the national auction is on the 2026 roadmap"; a flex engineer I spoke to put it more carefully: "We need a DNO settlement signal that arrives before our auction closes. Right now it doesn't."

Two — settlement integrity. Baselining is the technical heart of every flexibility market. Octopus uses a 10-day regression model with day-of-week, weather, and a participation-day dummy. Two competing aggregators (Kaluza and Energy Pool) use different baselines. There is no national reference baseline. Until there is, comparative settlement is hard.

Three — interactions with the capacity market. A 312 MW DSR resource that consistently shifts evening demand is, by EMR Delivery Body definition, a capacity-market-eligible asset class. Octopus pre-qualified roughly 180 MW of residential DSR for the upcoming T-1 auction. If that volume clears, the same household demand-shift will earn capacity payments and auction revenues — explicitly permitted under the rules, but a regulatory body will eventually have to decide whether the aggregate compensation is overpaying for a single megawatt-hour of restraint.

So what does 312 MW prove

It proves residential flexibility is no longer a pilot. The numbers are settlement-grade, the bid stack is layered, the clearing prices behave like a market. The unresolved question is whether the rest of the GB regulatory architecture — settlement, DSO procurement, capacity market — adapts fast enough that this volume becomes a planning assumption, not a happy accident.

Octopus runs the next auction on 4 June. The number to watch is not the cleared volume; it is the marginal accepted bid. If it falls below £3/kWh, the residential supply curve is thickening. If it rises above £6/kWh, demand for flexibility is outrunning supply. Either way, the signal is more interesting than any forecast.