Start typing to search this publication.
techcirkle
Open menu

Subscribe to techcirkle

Get new posts delivered straight to your inbox.

Designing the Connected Layer of a Vehicle: Identity, Telemetry Ingestion and Offline-Safe Commands

Technical guide to the connected layer in car software development: device identity, telemetry ingestion, offline-safe commands and entitlements.

techcirkle
5 min read
Cover image for Designing the Connected Layer of a Vehicle: Identity, Telemetry Ingestion and Offline-Safe Commands
Connected vehicle interface representing the cloud and telematics layer of car software development
Connected vehicle interface representing the cloud and telematics layer of car software development

Most commercial software teams enter automotive through the connected layer, because it looks like the cloud and IoT work they already know. It is, but with constraints that punish shortcuts: clients that live for fifteen years, intermittent connectivity, hardware revisions that multiply across the fleet, and regulation that cares about every update.

This post is a technical tour of that layer: identity, telemetry, commands, entitlements and the data model. It is one thread of a broader guide, Car Software Development in 2026, and it deliberately ignores the safety-critical embedded side.

What this layer owns

Out of the six workstreams under the "car software" label (embedded control, driver assistance, cockpit, connected services, companion apps, dealer and aftersales), the connected layer sits in the middle. It talks to the vehicle on one side and to apps, dealers and analytics on the other.

Its responsibilities are:

  • Establishing who each vehicle is.

  • Ingesting telemetry from a flaky network.

  • Delivering commands and updates safely.

  • Deciding what each vehicle is entitled to do.

  • Keeping personal and operational data apart.

Device identity

Everything else depends on this. Each vehicle needs a strong identity, typically a certificate issued at manufacture or first provisioning, and a lifecycle plan for rotation and revocation. Plan for resale and ownership transfer from day one: identity belongs to the vehicle, while the link to an owner is a separate, revocable relationship.

vehicle:
  id: vin-or-surrogate
  certificate:
    issuer: fleet-ca
    rotateEveryDays: 365
    revokeOn: [theft, decommission, compromise]
ownership:
  vehicleId: vin-or-surrogate
  userId: user-123
  validFrom: 2026-01-01
  validTo: null

Treat the vehicle as untrusted for authorisation decisions. A certificate proves which device is calling; it does not prove that the request is permitted. Enforce authorisation server-side against the entitlement model below.

Telemetry ingestion

Cars generate bursts of data when connected and nothing when in a tunnel. A good ingestion pipeline therefore:

  1. Accepts batched, timestamped messages with a client-side sequence number.

  2. Deduplicates on sequence number, since retries are normal.

  3. Uses event time, not arrival time, for analytics.

  4. Validates against a versioned schema and upcasts older versions.

  5. Separates identifiers that tie data to a person from engineering telemetry.

The versioning point deserves emphasis. Vehicles in the field keep emitting whatever format they shipped with, so every API and telemetry schema should be versioned from day one. Breaking a schema in the backend creates an outage that no OTA can fix quickly, because the cars that need the fix are the ones that cannot report.

Commands that survive bad networks

Remote commands (unlock, pre-heat, start charging) should be modelled as durable messages, not RPCs.

type Command = {
  id: string;
  vehicleId: string;
  kind: "unlock" | "preheat" | "charge_start";
  issuedBy: string;       // authenticated user, not the vehicle
  issuedAt: number;
  expiresAt: number;      // too old to be safe -> dropped
  status: "queued" | "delivered" | "executed" | "expired" | "rejected";
};

The lifecycle is queue, deliver, confirm, and every transition is logged for audit. Expiry is a safety feature: a command typed in a car park without signal must not execute an hour later. Return a command id immediately and let the app poll or subscribe for status, so users see an honest state instead of a spinner.

Entitlements and new revenue models

Feature-on-demand subscriptions, premium connectivity and paid upgrades all need an entitlement system that the update pipeline can act on. These look like SaaS billing problems, but the vehicle adds offline states, resale, ownership transfer and regional legal differences. Our SaaS development work covers the billing side, and the key advice here is to design the entitlement model before the first paid feature, because retrofitting it across hardware generations is costly.

A workable model separates three things: what the hardware can do, what the owner has purchased and what regulation allows in the current market. Evaluate them together server-side and push the result to the vehicle as a signed, expiring entitlement.

Data model and privacy

Connected vehicles produce location, driving behaviour, component health and cabin interaction data. It is valuable for product improvement, insurance partnerships, fleet operations and predictive service, and it is tightly regulated. Practical design points:

  • Build consent, retention and anonymisation into the data model rather than layering them on.

  • Keep a pseudonymous key for engineering analytics and hold the mapping to a person in a separate, access-controlled store.

  • Tag every field with its purpose, so deletion and export requests can be answered automatically.

Operating the layer

A few habits separate stable platforms from fragile ones.

  • Design for heterogeneity. Hardware and firmware variants multiply, so compatibility rules belong in data, not in code branches.

  • Rehearse OTA rollouts with a pilot fleet before launch.

  • Run threat analysis and risk assessment on every connected interface, which ISO/SAE 21434 and UNECE R155 expect, and keep a documented vulnerability response process.

  • Make evidence machine-readable so UNECE R156 update-management reporting comes from the toolchain.

For the delivery side, see how we approach cloud application development and the broader guide linked above.

Cost and timeline sanity check

For planning, a companion app plus connected backend (non-safety-critical) often lands between roughly 150,000 and 600,000 USD for a first production release, and an OTA platform with signing, staged rollout and compliance traceability between 400,000 and 1.2 million USD. A first production release of an app and backend typically takes four to nine months. Budget 15 to 25 percent of build cost per year for maintenance, and more for anything connected.

Working with us

If you are building this layer, our custom software development services team delivers OTA backends, telematics platforms and owner apps, and we say clearly when a safety-critical component needs a specialist supplier. Get in touch with a short description of your platform, or read the full article at techcirkle.com/blog/car-software-development.

Frequently Asked Questions

Why model commands as durable messages instead of API calls?

Because vehicles lose connectivity. Durable messages with expiry and confirmation let you queue, retry and audit commands without executing stale ones at the wrong moment.

Where should authorisation decisions live?

On the server. The vehicle certificate identifies the device, but the entitlement and ownership models decide whether a given request is allowed.

How do we handle vehicle resale and ownership transfer?

Keep vehicle identity separate from the owner relationship. Ownership becomes a revocable, time-bound link, so transfer revokes old access and entitlements without touching the device identity.

How should telemetry schemas evolve?

Version them from day one, allow optional fields within a version, create a new version for breaking changes and keep supporting old versions until fleet data shows no vehicle still emits them.

Can this layer be built without ISO 26262 expertise?

Generally yes, since it is non-safety-critical, but it still falls under cybersecurity and update regulations such as R155 and R156, and it must integrate cleanly with safety-critical components built by specialists.

Subscribe to techcirkle