Your team lives in a handful of screens inside every platform you pay for. We rebuild exactly those, as software that belongs to you — deployed in weeks, at a price agreed on the first call.
It starts with your invoices. The audit ranks every subscription by what a replacement would return, and you keep that ranking whether or not we build anything.
A working internal tool used to fill a quarter. It now fits in weeks, and the estimate holds well enough to fix a price against it.
Owned software costs the same at ten users as it does at a thousand. Your headcount grows, your hiring plan speeds up, and the line item stays put.
For most tools worth replacing, the records are yours already. You are paying for an interface over them, and interfaces are the part that got cheap.
Pieter Levels — Nomad List, Photo AI, twelve years of building in the open — posted the full list of SaaS he replaced with his own code, and what each one had been costing him. One person, his own products, his own figures. We read it as a price list for what has become buildable.
| SUBSCRIPTION CANCELLED | BUILT INSTEAD | REPORTED SAVING |
|---|---|---|
| Human moderators | Moderation running on a model API across every site | ~$10,000 / mo |
| Customer support | Featurebase plus an AI support bot | ~$5,000 / mo |
| Google Vision NSFW detection | NudeNet, self-hosted | ~$2,500 / mo |
| Image resizing API | Sharp and Redis on his own server | ~$1,500 / mo |
| Mapbox | OpenFreeMap | ~$500 / mo |
| Cloudflare Stream | Nginx on a VPS | ~$250 / mo |
| ApiFlash | A screenshot service for social images | ~$250 / mo |
| New Relic and IPHub | error2telegram — monitoring, blocking and fleet management | ~$250 / mo |
| ScrapingBee and SerpAPI | Playwright driving real Chrome on a VPS, about $1 a month | ~$250 / mo |
| IFTTT | A Telegram bot | ~$100 / mo |
| Photopea, embedded as an editor | His own photo editor | ~$100 / mo |
| Uptime Robot | Uptime Kuma on a $5 box | ~$50 / mo |
| Ghost | His own blog platform | Not stated |
| Weather API | Norway’s public meteorological API | Not stated |
This is the half of the list worth reading twice. Every one of these could be self-hosted, and he decided against it — because the alternative is becoming a registrar, a mail administrator, or a datacenter. It is a working definition of where rebuilding stops paying.
We use the same cut. If the vendor is selling scale, compliance, or a reputation you cannot rebuild, keep paying them.
A tool has to show most of these before we will quote a build against it. Failing them is not a problem to solve — it means the vendor is doing something for you that is genuinely hard to do yourself.
Open the tool and count the screens your team touches in a normal week. Three or four out of forty is the shape we are looking for. You are funding the other thirty-six so that somebody else’s enterprise customer can have them.
Per seat, per record, per run, per API call. The invoice grows with your headcount and your traffic, while the cost of serving you barely moves. That gap is the thing you are actually buying out.
If the product is mostly a view over records you own and can export, the expensive half of the problem is already solved. The view is the part that got cheap.
Invoicing, approvals, reporting, file processing. Boring and settled is buildable. A domain that changes every quarter is one you want a vendor chasing on your behalf, and they will do it better than you will.
Three numbers decide this: what you pay now, what running your own version costs every month, and what the build costs once. Everything after that is a conversation.
Illustrative inputs. Your real quote follows a technical review.
Include support, monitoring, and third-party services in running costs. Contract overlap, usage changes, and unexpected work can extend payback.
Not a service menu. These are the categories where the arithmetic tends to work, drawn from the tools clients actually bring to the first call.
Scheduled reports, live metrics and client portals, wired straight to your own database instead of a per-seat bill.
Approvals, task routing and back-office screens, including the integrations your current platform refuses to do.
Image resizing, transcoding, thumbnails and document processing, running at your own compute cost.
Geocoding, maps, weather, currency. Metered endpoints that often have a free public source behind them.
Health checks, error capture and alerting into the channel your team already reads, without a per-host price.
Content checks, tagging and routing, done by a model you call directly rather than a per-request vendor.
Extraction, screenshots and lookups. Usually the largest gap between what a vendor charges and what it costs to run.
Blogs, docs and landing pages, on your own domain and in your own repository, with no seat count on the editor.
Triage, first replies and routing, sitting on your history rather than a helpdesk’s per-agent tier.
We do not replace a stack. We replace the single subscription with the best ratio, prove the number on a real invoice, and then look at the next one with you.
Four stages. The subscription stays live through all of them, and you decide when it goes.
You send the subscription list and what each one costs. We sort it by what it would take to replace and what that would return, and hand the whole ranking back to you.
OUTCOME / A RANKED LIST, YOURS TO KEEPWe take the strongest one and read it properly: the data model, the exports, the integrations, the permissions. Then a fixed scope and a fixed price, or a recommendation to leave it alone.
OUTCOME / A SCOPE AND A PRICEYou use it while it is being built, against a copy of your real data. Feedback still changes what ships, because nothing has been cut over yet.
OUTCOME / WORKING SOFTWAREBoth systems run on the same data until the outputs match and your team stops opening the old one. You cancel at your renewal date. We hand over the repository, the deploys and the runbook.
OUTCOME / A CANCELLED INVOICEThe parallel run is the whole thing.
Migrations fail at the cutover, not the build. So we do not have a cutover — we have an overlap you end when you are ready.Then we say so and you keep paying it. That happens often, and it is the most useful outcome of an audit.
We would rather be the studio that talked you out of a build than the one that sold you a tool with a nine-year payback.
Your team, or us on a defined monthly arrangement. It is agreed before the build starts, not discovered afterwards.
If neither of those has an answer, the honest recommendation is to stay on the subscription.
Migration is part of the scope, and the export path is the first thing we check in the audit.
If a vendor has no usable export, that is a finding in itself, and it changes what the replacement is worth to you.
Divide the build cost by what you save each month after running costs. The estimator on this page does exactly that.
Add a quarter for contract overlap and the answer is close enough to decide on.
Yes. The repository is in your organisation from the first commit, along with the deploys, the environment and the runbook.
The point of the exercise is that nobody can put a price on it again.
You can, because the subscription is still running. Nothing is cancelled during the build or the parallel run.
The cancellation is your decision and your calendar, at your renewal date.
AI is in our process, and it is a large part of why the arithmetic changed. It is not the deliverable.
What ships is reviewed, tested and readable by whoever inherits it.
Yes. The audit stands on its own and you keep the ranking whether or not we build anything.
Bring the invoice list to the call and we will start there.
Thirty minutes on your actual subscription list. You leave with a ranking, whether or not we build anything.
Your subscription list with what each one costs, and a rough count of who logs into them.
Sort them by what a replacement would return, and mark the ones you should keep renting.
A ranked shortlist, a break-even for the top one, and a fixed price if it clears.
Free, and there is no follow-up sequence. If every subscription on your list is worth keeping, that is what the call will conclude.
Four fields. It rides along with your booking, so the call opens on your numbers instead of an introduction.
Name the tools you are paying for, roughly what they cost, and which screens your team actually opens. We read it before we join.
Nothing is sent until you confirm a time on the calendar above.