# Writing hardware product requirements engineers can use

> Vague hardware product requirements cost months later. How to write measurable, testable requirements for a connected device before design starts.

- Author: Milo Perović, Founder & lead engineer
- Published: 2026-10-25
- Topic: Process

Most connected-hardware projects that slip by months do not fail on a hard engineering problem. They fail on a sentence nobody questioned early: "long battery life", "rugged", "works anywhere". Writing hardware product requirements that an engineer can design against, and a tester can later prove, is the cheapest risk reduction available to a founder or product manager.

This post is for teams about to brief a hardware build, whether in-house or with a studio like ours. It covers what a usable requirement looks like, which numbers matter most for a connected device, and how to keep the document alive once design starts.

## Why vague requirements are expensive in hardware

In software, an ambiguous requirement usually costs a sprint. In hardware it can cost a board respin, a new mould insert or a failed certification run. Each of those carries lead time measured in weeks, not days, because it involves fabrication, assembly or a booked lab slot.

The ambiguity also compounds. "Outdoor use" quietly decides the ingress rating, the operating temperature range, the plastic grade, the UV stabiliser, the battery chemistry and whether a vent membrane is needed. If nobody writes those decisions down, each engineer makes a different assumption, and the conflicts surface at the first integration test.

## What good hardware product requirements look like

A usable requirement has four properties. It is specific, it is measurable, it states the conditions under which it applies, and it has a verification method attached. If any of the four is missing, it is a wish rather than a requirement.

Compare these two versions of the same intent:

- **Wish:** "The battery should last a long time."

- **Requirement:** "The device shall operate for at least 12 months on one 2,400 mAh Li-SOCl2 cell, reporting every 15 minutes over LoRaWAN at SF9 or lower, at an average ambient of 20 °C. Verified by measured average current on the bench, multiplied out with a 20% derating for self-discharge and temperature."

The second version is longer, but it tells the hardware engineer which power budget to design to, tells the firmware engineer how often the radio wakes, and tells the tester exactly what to measure. It also exposes the assumptions, so that when someone later asks for a five-minute reporting interval, everyone can see what it costs.

We use "shall" for mandatory requirements and "should" for goals that are desirable but negotiable. It is an old convention from systems engineering, and it saves a surprising number of arguments.

## The numbers a connected device brief needs first

Not every requirement deserves the same attention at the start. For a typical connected product, these are the ones that drive architecture, and they are worth pinning down before anyone opens a schematic tool:

- **Power source and target life.** Mains, rechargeable or primary cell, and the life or charge interval expected. This decides the MCU family, the radio and the sleep strategy.

- **Connectivity and data pattern.** How much data, how often, how far, and what happens when the link is down. A few bytes every hour suggests LoRaWAN or NB-IoT; streaming to a phone suggests BLE; continuous high throughput suggests Wi-Fi or cellular.

- **Environment.** Operating and storage temperature, humidity, ingress protection (for example IP54 or IP67), UV exposure, drop height and vibration. Name a standard where one applies, such as IEC 60529 for ingress or IEC 60068 for environmental tests.

- **Size, weight and mounting.** Envelope dimensions, how the device is fixed and what it must not obstruct. This constrains antenna placement and battery size more than most people expect.

- **Target markets and certification.** CE RED for Europe, FCC Part 15 for the United States, plus any sector-specific standards. Markets change the radio bands, the test plan and the label.

- **Unit cost and volume.** A target bill of materials cost at a stated volume. A device built for 500 units a year is a different design from one built for 50,000.

- **Service life and updates.** How many years the product must stay in the field, and whether firmware must be updated remotely. If yes, that implies flash headroom, a bootloader and a backend from day one.

If a founder can answer these seven well, the rest of the document is mostly detail. If they cannot, that is the most useful thing to discover in week one.

## Separating needs from solutions

A common trap is writing the solution into the requirement. "The device shall use an ESP32-S3" is a design decision dressed as a requirement. Unless there is a real reason, such as an existing firmware codebase or a pre-certified module the team already owns, it closes off options before the trade-offs are understood.

A better approach is to state the need, then record the constraint separately with its reason. "The device shall pair with iOS and Android phones without an app store install for initial setup" is a need. "We must reuse our existing STM32 firmware stack" is a constraint, and writing down why it exists lets the team challenge it later if it turns out to be expensive.

This separation is the same idea behind the first phase of our [CAEC process](https://miloperovic.com/journal/what-the-caec-process-looks-like-on-a-typical-project/): understand what the product should be before deciding how to build it.

## A simple format that works

You do not need requirements management software for a first product. A shared spreadsheet or document with a consistent structure is enough, as long as every line carries the same fields.

FieldPurposeExample

IDStable reference for tests and ticketsPWR-003

RequirementOne testable statement using "shall" or "should"Standby current shall not exceed 15 µA at 25 °C

RationaleWhy it exists, so it can be challenged sensiblyNeeded to meet the 12-month battery target

PriorityMust, should or couldMust

VerificationHow it will be provenBench measurement with a current profiler, five units

Owner and statusWho answers questions, and whether it is agreedProduct lead, agreed

Keep one requirement per line. If a line contains "and", it often contains two requirements that will be verified differently, and it is worth splitting.

## Keeping the document alive during design

Requirements change, and that is normal. A thermal simulation shows the sealed enclosure runs hotter than expected, or a supplier quote pushes the bill of materials over target. The problem is not change itself but untracked change, where the design drifts and the document stays frozen.

A few habits keep it honest:

- Version the document and note what changed and why at each design review.

- Link every test in the bench plan back to a requirement ID, so gaps are obvious in both directions.

- When a requirement is relaxed, record the decision and who made it, rather than quietly deleting the line.

- Flag requirements that are still assumptions, such as an unmeasured average current, until real data replaces them.

At the end of the build, the same document becomes the backbone of the verification report: each requirement, the test that covered it, and whether it passed.

## A checklist before you commit to a build

Before signing off a brief and starting detailed design, we would want to see the following in place:

- The seven architecture-driving numbers above are written down, even if some are ranges.

- Every mandatory requirement has a measurable threshold and a stated test condition.

- Every requirement has a verification method, even a rough one.

- Constraints are separated from needs and each has a reason.

- Target markets and certifications are named, not implied.

- Someone owns the document and has authority to agree changes.

None of this needs to be perfect on day one. A clear draft that the team argues over is far more valuable than a polished document nobody reads. If you are preparing a brief and want a second pair of eyes on it, a [free 30-minute scoping call](https://miloperovic.com/contact/) is usually enough to spot the gaps that matter.
