What failed sessions cost, and what it costs to rescue them
Swedish chargers were observed available 95.6 percent of the time in Q2 2026. Only 87.4 percent of sessions succeeded. That 8.2 point gap is the difference between hardware being online and a driver actually getting charged, and it is where the money leaks. This model decomposes the failed 12.6 percent by cause, prices it, and answers the question operators avoid: if the backend is unreachable, should the charger start the session anyway?
Where attempts die
Every plug-in event is an attempt. It passes through authorization, payment, and negotiation before energy flows. The first bar is the whole population. The second rescales the failed slice so the causes are legible. Defaults are calibrated to the measured Swedish session success rate of 87.4 percent; the split between causes is estimated, since published measurement gives the total but not the taxonomy.
The decision
Backend-unreachable and roaming-token failures are the two classes a cached local authorization list can rescue. The charger delivers energy without live authorization and settles later. Some of those sessions never get paid. The question is how many is too many.
·
Annual ledger
Gross margin basis. Energy already delivered on an unpaid offline session is a real cash cost, not a foregone opportunity, so it is booked at cost plus recovery admin.
| Line | Sessions | SEK / year |
|---|
Portfolio
Method and where it is soft
- Breakeven bad-debt rate solves
(1−r)·m = r·(c+a)forr, givingr* = m / (m + c + a), wheremis margin on a capped offline session,cis the delivered energy cost, andais recovery admin per unpaid session. It is independent of volume, which is why it travels between operators. - Failure classes are treated as sequential independent gates, so the compound success rate is the product of survivals, not one minus the sum. At single-digit rates the difference is small but it is the correct construction.
- Only backend-unreachable and roaming-token failures are marked recoverable. Payment declines are not: the card said no, and starting anyway converts a failed session into a guaranteed unpaid one. Handshake and low-delivery aborts are vehicle-side or hardware-side and unaffected by authorization mode.
- The headline failure rate is measured, not assumed. Symbioen's Q2 2026 benchmark observed Swedish stations available 95.57 percent of the time against a session success rate of 87.40 percent, across 3.2 million status-derived sessions. Defaults here are calibrated so the compound survival across the five gates reproduces that figure. The split between the five causes is still my estimate: published measurement gives the total, not the taxonomy. Replace it with your own CDR data before quoting any per-cause number.
- The failure rate is going the wrong way. Sweden's observed station inventory grew 12.7 percent between Q1 and Q2 2026 while session success fell from 89.8 to 87.4 percent. Norway grew 4.2 percent and improved slightly. Expansion and service quality are not moving together, which means a portfolio modelled on last year's failure rate is understating the loss.
- Not modelled: reputational decay from repeat failures at the same site, the shift in lifetime value when a driver deletes the app, and idle-fee interactions. All three make failure more expensive than shown, so this reads as a floor.