Recurring revenue — what this shop could actually sell twice¶

Status: proposal — awaiting decision, 2026-09-14. Written after Angel asked what business models would give recurring income. Opinionated on purpose, and grounded in what the shop already has rather than what a consultancy deck would suggest.

The filter¶

Project work pays once and stops when the customer stops. Recurring revenue is worth having, but most of the models people reach for do not survive contact with a two-person shop, because they require someone else's honesty (royalties) or an obligation that never sleeps (hosted services). So each idea below is judged on three things:

Does it use something we already have?
Can we enforce it by delivering, rather than by auditing or suing?
What does it oblige us to do at 2am? An obligation with no staff behind it is a liability dressed as revenue.

The three worth doing¶

1. Obsolescence and compliance watch — the strongest fit¶

An annual fee per product to watch the customer's bill of materials and tell them before a part bites: NRND and end-of-life notices, lead times stretching, single-source risk, and a yearly written review of what needs designing out.

Why it fits this shop specifically: the tooling is already built. The DigiKey and Mouser keys, the part cache, the lifecycle watch, the ERP holding real BOMs — that work is done and currently serves only us. The marginal cost of watching a customer's BOM alongside our own is close to zero, and the value to them is obvious the first time it saves a respin.

It also attaches naturally to work already delivered: every board this shop designs is a candidate the day it ships.

And there is now a second reason to buy it. The EU Cyber Resilience Act's reporting obligation started on 11 September 2026, and its full requirements land in December 2027 — a software bill of materials, a way to ship signed updates, and somebody watching CVE feeds for the libraries a product ships with. That is a recurring obligation the customer now has by law, and it is exactly the shape of thing a small manufacturer will pay someone else to carry. The note published on 2026-09-14 is already the marketing for it.

What it obliges us to do: send a report on a schedule and an alert when something breaks. Nothing at 2am.

2. Test as a service — fixture plus per-unit¶

Build the customer's test fixture once (paid), then test every unit they build, priced per unit. The Universal Tester is the asset; the fixture is the lock-in, and an honest one — they keep the results, we keep the throughput.

Why it fits: it turns a capability into volume revenue that grows with their success rather than with our hours. It is enforced by us physically doing the work.

What it obliges us to do: turn units around on a schedule. Real, but bounded, and it is the business the shop is already in.

3. Repeat build agreements — boring, and the biggest number¶

Not a new model: a standing agreement to build the same board on a replenishment schedule, with a blanket order and agreed lead times, rather than being re-quoted each time. Predictable revenue, predictable purchasing, and it makes component buying cheaper because quantities are known in advance.

Why it matters more than it sounds: it converts the customers already here into a base load. That is worth more than a new product line and costs nothing to try — it is a conversation, not a build.

Worth doing later, or in a specific case¶

Universal Tester software subscription. Annual for updates, new instrument drivers and the analytics export. Sell the station outright and subscribe the software. Deliberately parked: universal-tester-product already decided against licence enforcement for now, and shipping the product matters more than monetising it twice.
Fractional hardware engineer. A fixed number of hours a month at a retainer rate. Smooths cash flow, and honest about what it is. The cost is that it fills the calendar with somebody else's priorities, which is exactly what makes the tester slip.
Licensing a design to a manufacturer. Only with minimum annual payments — see terms-and-warranty. Depends on someone else's unit reporting, which is the thing this shop cannot police.
Spares and consumables. Pogo pins, fixtures, adapters, cables. Small, but it recurs on its own and needs no new capability.

Not now, and why¶

Hosted device infrastructure — OTA update servers, telemetry dashboards for customers' products. Genuinely recurring, genuinely valuable, and an obligation that never sleeps: an outage becomes their product failing. Not with two people and one server.
Selling this internal platform as software. Tempting, and it comes up in roles-and-skills and task-board. It is a second company, not a second revenue line: support, multi-tenancy, onboarding, someone else's data. The platform should earn its keep by making this shop faster first.
Training or a course. Cheap to say, expensive to make, and it competes for the same writing time the content pipeline already uses to sell the actual services.

What to do about it¶

Nothing before PCB Carolina on 11 November — the tester is what earns this year, and all three good ideas above are sales conversations rather than builds.

Then, in order:

Offer the obsolescence watch to every existing customer with a board in the field. One email, an annual price, and the report template mostly exists already.
Propose a build agreement to whichever customer reorders most often. A conversation.
Price test-as-a-service once the tester is shipping and its throughput is known.

The first two are worth trying because failing costs an afternoon.

Log¶

2026-09-14: written.