Nothing erodes trust in a charging operation faster than a wrong invoice. A driver charged for energy they didn't receive, a fleet billed twice for one session, a site host's revenue share that doesn't reconcile with the meter — each one costs more in credibility than it does in refunds.
And almost every billing dispute, when you dig, is a data dispute. The tariff maths was fine; the inputs weren't. Here's where charging billing data goes wrong, and the engineering that keeps it honest.
The pipeline, and its weak joints
A charging invoice is the end of a long pipeline:
Rendering diagram…
Each joint has a characteristic failure:
Meter → OCPP. The charger's meter is the source of truth, but what leaves the charger is a sampled report of it — meter values at intervals, plus start/stop readings. Chargers differ in sampling cadence, units (Wh vs kWh has caused real invoices to be out by a factor of a thousand), and which measurands they send. Normalise units at ingestion, validate against the charger's known configuration, and reject the physically impossible — a 22 kW AC post reporting 300 kWh in an hour is a data fault, not a big session.
OCPP → session record. The classic failures are incompleteness and duplication. Connectivity blips at session end and the stop event arrives late, twice, or never — creating the zombie sessions that either overbill (clock kept running) or underbill (energy delivered, never closed). The disciplines that matter: idempotent event handling keyed on transaction identity (a retried stop event must not close a session twice), and offline reconciliation — chargers queue transaction data locally, so when one reconnects after an outage, its backfilled sessions must merge cleanly rather than duplicate.
Session → tariff. Time-of-day boundaries, idle fees, per-session versus per-kWh components — tariff bugs are amplified by the previous stage's data quality. A session with a missing stop time meets an idle-fee rule and generates the £400 invoice that ends up on social media. Tariff application should refuse to price a session that fails integrity checks, parking it for review instead.
Tariff → settlement. Multi-party now: driver payment, site-host revenue share, roaming partners settling over OCPI. Same session, three consumers, three chances to disagree — unless they all read the same sealed session record.
The disciplines that make it boring
1. Immutable session records with explicit status. A session moves through states — active, completed, pending-review, billed — and once billed, the underlying record never changes; corrections are new, linked adjustment records. This single rule turns "the numbers changed since last month" disputes into audit-trail lookups.
2. An integrity gate before pricing. Sessions must pass checks before they're billable: start and stop readings present, energy monotonically increasing, duration plausible, meter delta matching reported total. Fail → quarantine queue with a reason code. Operators are always surprised by what the gate catches in week one — and by how quiet it goes once upstream fixes land.
3. Idempotency everywhere money is created. Every event that can create or modify a billable record must be safely replayable — retried webhooks, backfilled offline data, platform migrations. We covered the consumer side in streaming telemetry; the producer side is the same rule with higher stakes.
4. Reconciliation as a scheduled habit. A periodic job comparing session energy totals against meter deltas per charger, flagging drift. Drift means a meter fault, a data fault, or theft — all things you want to find before the site host's accountant does.
5. Keep the raw messages. When a dispute does arrive, the raw OCPP log is the evidence that settles it in minutes: here's the meter start, here's every sample, here's the stop. No raw history, no defence — just competing recollections.
The commercial payoff
Billing integrity feels like plumbing until you see its absence priced: support hours on disputes, goodwill refunds issued because the data couldn't prove otherwise, tender questions about billing accuracy answered with adjectives instead of architecture, site hosts commissioning their own check-metering because they stopped trusting yours.
The compounding asset is the opposite reputation: invoices that reconcile, every time, with evidence on demand. It's not glamorous engineering — which is exactly the point. Money data should be the most boring data you have.
AmpNexus applies integrity gates, idempotent ingestion and full OCPP history to every session — see sessions & transactions or the API reference.