Deployment automation for small products: what CI/CD buys you and the smallest setup worth having
Deployment automation for small products: what CI/CD buys you, a minimal GitHub Actions pipeline, real costs and where automation stops helping.
By Rokibul Hasan8 min read
- Web development
- Automation

If your product goes live because one person follows a list of steps on their own laptop, that list is a risk to the business. Deployment automation replaces it with a process that runs the same way every time, starts with one action and records what went live and when.
This guide is for founders and product owners with a small web app, online store or Shopify app, usually with one to five people touching the code. It explains what CI/CD buys you, the smallest setup worth having, what it costs on GitHub Actions, and where automation stops helping.
There is one short YAML example, mainly to show that a basic pipeline fits on a single screen.
What deployment automation means
DORA, the research programme behind the widely used software delivery metrics, describes deployment automation as being able to release to test and production with a single action. Three terms get mixed up around it:
- Build automation turns source code into something that can run: installing packages, compiling, bundling.
- Continuous integration (CI) builds and tests every proposed change automatically, so a problem shows up while the change is still fresh.
- Continuous delivery and continuous deployment (CD) are close relatives. With delivery, any tested change can go live when someone presses a button. With deployment, it goes live without anyone pressing anything.
For a small product either form of CD is fine. What matters is that nobody repeats the steps by hand.
What a manual release really costs
| Manual step | How it goes wrong | With a pipeline |
|---|---|---|
| Build on a laptop | Works on one machine, fails on the server | A clean machine builds every time |
| Run the tests if there is time | Skipped when the deadline is close | Tests run on every change and block a failed release |
| Run the deploy commands | Wrong branch, wrong environment, a missed step | The same script runs each time |
| Keep keys on one computer | The person leaves and the access goes too | Secrets sit in an account you own |
| Remember what shipped | Nobody knows which version is live | Every release points to a commit |
Speed is the benefit people quote, but three others matter more to an owner.
You stop depending on one person. If your developer is ill or moves on, someone else can still ship a fix. Every plan I quote puts the code and accounts in your name and ends with handover notes, and a pipeline in your own account follows the same logic.
Releases get smaller. When a release is cheap, people ship one change at a time. When something breaks, there is far less to search through.
You get a record. "What changed on Tuesday?" has an answer.
Two of DORA's metrics are worth tracking even on a tiny team: how often you deploy, and how often a deployment needs an urgent fix. DORA warns against comparing these across different products, so compare yourself with last quarter.
The same source carries a warning. Automating a fragile process gives you a fragile automated process. Its simplest measure is to count the manual steps, so start there. If your list includes a step nobody can explain, fix that before you script it.
A minimal CI/CD pipeline in four parts
- Everything in version control. Code and configuration live in Git. Passwords and API keys do not.
- Checks on every proposed change. Install, type check, test, build. If one fails, the change does not merge.
- One route to production. Merging to the main branch deploys. Nothing else does, including a developer's laptop.
- A way back. Know how to undo a release before you need to, and try it once.
Here is the third part as a GitHub Actions workflow. The last step deploys to Cloudflare Workers with the action Cloudflare documents. On another host, only that step changes.
name: Deploy
on:
push:
branches: [main]
workflow_dispatch:
concurrency: production
jobs:
deploy:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
- uses: actions/setup-node@v7
with:
node-version: lts/*
cache: npm
- run: npm ci
- run: npm test
- run: npm run build
- uses: cloudflare/wrangler-action@v4
with:
apiToken: ${{ secrets.CLOUDFLARE_API_TOKEN }}
accountId: ${{ secrets.CLOUDFLARE_ACCOUNT_ID }}
In plain terms: on every push to main, a clean machine fetches the code, installs the exact dependencies, runs the tests, builds and deploys. If the tests fail, nothing ships. workflow_dispatch adds a Run workflow button in GitHub, and concurrency stops two deployments overlapping.
Two cautions for real use. GitHub's security guidance says to pin actions to a commit SHA, and the version tags above are there for readability. It also warns that anyone with write access to a repository can read its secrets, so keep that list short.
If you want a named person to approve each release, check your plan first. On GitHub Free, Pro and Team, required reviewers on environments only work on public repositories, and environment secrets on a private repository need Pro or above. A simple alternative is to remove the push trigger, so a release happens only when someone with write access presses the button.
You may already have most of this
Hosts and platforms include part of a pipeline. If yours covers what you need, a separate CI tool adds work with no payoff.
| Where the product runs | What you get | Watch for |
|---|---|---|
| Vercel | A preview URL for each branch or pull request, a production deploy on merge, Instant Rollback | The Hobby plan is non-commercial and rolls back one deployment only |
| Cloudflare Workers | Workers Builds deploys on push; Version URLs let you open a version before release | Version URLs are public unless you add Cloudflare Access |
| Shopify theme | The GitHub integration keeps a branch and a theme in step | The sync is two-way and cannot be turned off, and the code editor can overwrite a file without a conflict warning |
| Shopify app | shopify app deploy from a pipeline, using a service account token |
Extensions missing from the deploying environment are removed |
A brochure site that rebuilds on every push needs nothing more. Add your own pipeline when you have tests to enforce, more than one environment, or a step your host cannot run.
What GitHub Actions costs
GitHub's billing page, as of November 2026, sets the allowance in money instead of minutes: $12 a month on Free, $18 on Pro and Team. A standard Linux runner costs $0.006 a minute, so the Free allowance covers 2,000 Linux minutes. Standard runners on public repositories are free, and self-hosted runners do not draw on the allowance.
An illustration, using assumed numbers: an eight-minute pipeline runs 80 times a month, from three pushes on each of 20 pull requests plus 20 deploys. That is 640 minutes, or $3.84 of the $12. If you pass the allowance with no payment method on file, GitHub blocks further runs, so add a card before release day teaches you this.
macOS is the exception at $0.062 a minute, about ten times the Linux rate. Twelve dollars buys around 190 minutes, which an iOS build can use up quickly. My guide to mobile automation testing works through that bill.
So compute is rarely the expensive part. The cost is the time to set the pipeline up and keep its tests worth running, alongside the other running costs of a web app.
Where deployment automation stops
Data does not roll back. Undoing a release restores code. Cloudflare's rollback documentation says connected resources are not changed during a rollback, and warns that errors can occur if the structure of your data has changed. Vercel's rollback screen carries a similar reminder about databases and external APIs. So database changes should ship in two steps: add the new structure first, then remove the old one in a later release, once nothing depends on it. Take a backup before each.
App stores sit outside your pipeline. You can automate the build and the upload, but review and rollout belong to Apple and Google. Apple's phased release spreads an update over seven days, from 1% to 100% of users with automatic updates. Google Play lets you halt a staged rollout, but people who already have that version keep it, and Google's advice for a bad bundle is a new release with a fix. There is no instant undo.
A green pipeline only proves what it checked. If no test covers checkout, the pipeline will ship a broken checkout without complaint. Deciding what to test is a separate job, covered in Playwright automation testing.
A first version needs less. A few checks and one deploy route are enough while you find out whether people use the product, which is the argument of my MVP development guide.
Questions to ask your developer
- If you were unavailable tomorrow, how would the next release go out?
- Which account holds the pipeline and the secrets, and is it mine?
- What has to pass before a change goes live, and who can skip it?
- How do we undo a bad release, and have we tried it?
- How do database changes reach production?
- Where can I see which version is live right now?
FAQ
What is the difference between CI and CD?
Continuous integration (CI) means every proposed code change is built and tested automatically. Continuous delivery or deployment (CD) covers what happens next: with delivery, a tested change can go live with one action, and with deployment it goes live automatically. CI tells you whether a change is safe to release, and CD gets it to customers without anyone repeating steps by hand. Small products often start with CI plus a one-button release.
Does a small website need CI/CD?
Often not a custom pipeline. Hosts such as Vercel and Cloudflare can deploy automatically when you push to a branch, which is enough for a simple marketing site. A pipeline of your own earns its place once you have tests that should block a release, more than one environment, or a product people depend on daily, such as a web app or a store with custom code. Count your manual release steps to decide.
Is GitHub Actions free for private repositories?
Partly. As of November 2026 every GitHub plan includes a monthly compute allowance, $12 on Free and $18 on Pro and Team, which covers 2,000 to 3,000 minutes on a standard Linux runner at $0.006 a minute. Beyond that you pay per minute, or runs are blocked if no payment method is on file. macOS runners cost about ten times more, so mobile builds use the allowance much faster.
If you want releases that do not depend on one laptop, send me a short brief: what the product is, where it runs today, how a release happens now and what has gone wrong before. Adding a pipeline to an existing product is custom work quoted from a brief, and you can send yours through the contact form. You can also read about my full-stack web development work, or see what to test first in Playwright automation testing.


