Overview
Designing for the fleet, not the first unit
Most connected-product pain appears after launch: devices that cannot be identified reliably, configuration that has to be applied by hand, updates that cannot be rolled back, and failures that cannot be diagnosed without shipping the unit back.
These are design decisions, made early, usually by omission. We make them explicitly: how a device proves what it is, how it receives configuration, how it reports health, how it is updated and what happens when an update fails.
The result is a product that can be manufactured, commissioned, monitored and maintained at scale — which is usually what determines whether a connected product is profitable.
Device mechanisms
What a manageable device needs
Six mechanisms that decide whether a deployed fleet is supportable.
-
Identity and provisioning
A reliable way for each device to prove what it is, established at manufacture rather than improvised at installation.
-
Configuration management
Parameters that can be set, validated, changed remotely and recovered safely when something is set incorrectly.
-
Connectivity management
Connection handling, retry behaviour, fallback and defined operation during outages of any length.
-
Remote diagnostics
Health data, fault codes and logs that let a support engineer understand a failure without visiting the device.
-
Secure update
Firmware update with integrity verification, staged rollout, rollback and defined behaviour on interruption.
-
Lifecycle support
Versioning across a mixed fleet, backwards compatibility and a plan for devices that will still be running in ten years.
Failure modes
Fleet problems and the design decisions that prevent them
Each of these is a design decision made — or omitted — before the hardware is frozen.
| Problem after launch | Root cause | Design decision that prevents it |
|---|---|---|
| Devices cannot be told apart | Identity improvised at install time | Unique identity provisioned during manufacture |
| Update bricks units | No verification or rollback path | Signed updates, integrity check, atomic switch and rollback |
| Failures cannot be explained | No retained diagnostic data | Structured fault logging sized to the platform, retrievable remotely |
| Configuration drift across fleet | Manual configuration with no record | Central configuration model with validation and reporting |
| Mixed versions become unmanageable | No compatibility strategy | Versioning and backwards-compatible interfaces defined up front |
Fleet process
Engineering a device for its whole service life
Each stage below is where an unmanaged fleet usually starts to hurt.
-
Lifecycle definition
How devices are manufactured, commissioned, operated, updated, diagnosed and eventually retired.
-
Identity design
Device identity and credentials established during manufacture, in a process production can actually run.
-
Configuration model
What is configurable, who changes it, how it is validated and how a bad configuration is recovered.
-
Update architecture
Update packaging, delivery, verification, rollback and behaviour if power or connectivity is lost mid-update.
-
Diagnostics design
What the device records, what it reports, and how support staff retrieve it in the field.
-
Fleet validation
Mechanisms exercised across multiple devices and firmware versions, not just on one bench unit.
Scale
Designing for the fleet, not the first unit
Shipping one connected device is an engineering exercise. Shipping ten thousand and still being able to fix them two years later is a different problem.
Identity, configuration, diagnostics and update are decided early, because retrofitting them to deployed hardware is often impossible.
Applications
Where device management decides the outcome
Where it decides the outcome
- Large fleets of deployed sensors or controllers
- Devices installed in locations that are expensive to visit
- Products with long service lives and evolving firmware
- Equipment requiring commissioning by non-specialist installers
- Products where field failures must be diagnosed remotely
- Existing connected products that have become unmanageable
Mechanisms
- Device identity
- Provisioning
- Remote configuration
- OTA update
- Diagnostics
- Fleet versioning
FAQ
Connected device questions
What teams ask once they start thinking past the first unit.
Partially, and it depends on what the existing hardware and bootloader support. Adding update capability or identity to a deployed fleet is far harder than designing it in, which is why we raise it early.
Integrity verification before anything is applied, an update path that can be interrupted without leaving the device unusable, rollback to a known-good image, and staged rollout so a bad release does not reach the whole fleet at once.
Usually you do, or a platform vendor does. We engineer the device side to integrate with it. Where no platform exists yet, we can help define what the device needs it to do.
It should be simple enough for a non-specialist installer: the device identifies itself, receives its configuration and reports success. If commissioning needs an engineer, the design usually needs revisiting.
Next step
Planning a fleet you will still be supporting in five years?
We will map the identity, configuration, update and diagnostic mechanisms it needs before the hardware is fixed.