Building a physical product involves a set of constraints that software businesses have largely escaped, and the resulting failure modes are distinct.
The capital structure
The fundamental difference. Software can be iterated at negligible marginal cost. Hardware cannot.
Tooling — the moulds and fixtures for manufacturing — represents substantial upfront cost, and modifying it after production has begun is expensive.
Which means design decisions become effectively irreversible at a point, and a mistake discovered afterwards is either lived with or paid for.
Inventory ties up capital before any revenue. You build units, pay for them, and hold them until they sell. Getting demand forecasting wrong is expensive in both directions — too much stock is capital sitting in a warehouse, too little means missing the window.
The prototype-to-production gap
Where a great many hardware projects fail, and it's consistently underestimated.
A working prototype demonstrates that something is possible. Manufacturing at scale requires that it be possible repeatedly, by people who didn't design it, with consistent quality, at a cost that permits a margin.
Those are different problems. Techniques that work for one unit — hand assembly, selected components, individual adjustment — don't survive scale.
Design for manufacture is a discipline in itself, and projects that treat it as a final step rather than as a constraint from the start reliably encounter problems.
The industry observation that getting from a working prototype to reliable production takes longer and costs more than everything before it is repeated frequently enough to be taken seriously.
Supply chain
Every component is a dependency, and a hardware product may have hundreds.
Component availability changes. Parts are discontinued, allocated to larger customers, or subject to shortages. Any of these can halt production.
Lead times are long, which means committing to orders months before knowing whether demand materialises.
Minimum order quantities mean small volumes are disproportionately expensive, which is a particular problem for a company still finding its market.
And a single-sourced component with no alternative is a risk that only becomes visible when it fails.
The crowdfunding problem
A specific pattern worth understanding, since it's how a substantial number of hardware products have been launched.
Crowdfunding validates demand and provides capital, which addresses two real problems.
It also fixes the price before manufacturing costs are known, commits to a delivery date before manufacturing challenges are understood, and creates a large number of customers who paid long ago and are waiting.
Analyses of crowdfunded hardware have found substantial rates of delay and non-delivery, and the pattern is consistent enough to be structural rather than a matter of individual failures.
The commitments that seem reasonable when a prototype works frequently prove impossible once manufacturing realities emerge, and by then the price is set and the money is spent.
Certification and regulation
Frequently underestimated in both cost and time.
Products require certification for electromagnetic compatibility, electrical safety, radio emissions where applicable, and various regional requirements.
Each certification costs money and takes time, and failure requires redesign and retesting.
Regional variation means selling into multiple markets multiplies the requirement, and some markets have requirements that materially affect design.
Products in regulated categories — anything medical, anything for children, anything involving safety — face substantially higher barriers, and the timelines are measured in years.
Support obligations
A hardware product creates ongoing obligations that software doesn't.
Warranty claims, returns, replacement parts, and in many jurisdictions statutory obligations lasting years.
A defect discovered after shipping cannot be patched. It requires either living with the consequences or a recall, which for a small company is frequently fatal.
And customers reasonably expect support for the product's lifetime, which for physical goods is considerably longer than software companies are accustomed to committing to.
What appears to work
Patterns among hardware companies that have succeeded.
Manufacturing expertise early. Bringing in people who have shipped physical products, before the design is finalised rather than after.
Conservative first production runs. Small batches to identify problems before committing to volume.
Standard components where possible. Custom parts increase cost, lead time and risk. Every custom component should justify itself.
Realistic timelines. Doubling estimates is a commonly offered heuristic and is frequently insufficient.
A software or service component. Businesses combining hardware with recurring revenue have better economics than those selling a device once, which is why so many hardware products now come with subscriptions.
That last point explains a great deal about the current market, including the cloud dependency problems that follow from it.
Contract manufacturing
The relationship that determines a great deal, and it is frequently entered into naively.
Contract manufacturers are not extensions of your company. They have their own priorities, other customers, and minimum volumes below which you are not commercially interesting.
A small startup is generally the least important customer of any manufacturer large enough to be capable, which affects scheduling, attention when problems arise, and how quickly issues get resolved.
What experienced hardware people advise: visit the factory, have someone present during initial production runs, define quality criteria in writing before starting, and understand that the relationship requires management rather than delegation.