Systems that talk to each other.

We connect your CRM, payments, email and data in code. Every integration is built for the day a message arrives twice, arrives late, or the other side changes its API.

Three clusters of light points joined by arcs that meet at one glowing junction, for System Integrations

What it is

The easy part is the connection.

Connecting two systems takes an afternoon: one sends, the other receives. API integration services earn their fee on everything after that. The message that is sent twice. The one that arrives when your side is down. The vendor that changes its API with a few months of notice.

We build integrations in code, with those cases handled from the start. Some are simple, such as a website form that lands as a lead in a CRM. Some carry money, such as payments and subscriptions. The care is the same.

Where it breaks

Six ways an integration fails.

None of these are our opinion. Each is written in the documentation of the systems you would connect. Select one to see what the vendor says and what we build against it.

Receive

It arrives twice

What the vendor says
“Webhook endpoints might occasionally receive the same event more than once.”

Stripe, webhooks documentation

What we build

Every incoming event is recorded by its own id before anything else happens. A repeat is recognised and skipped, so the same event is not acted on twice.

In something we run

In a subscription product we built, each payment event is stored once by its id. If it cannot be stored, we tell the sender it failed, so it is sent again.

Quotations from each vendor’s own documentation, read on 15 November 2026. The examples are from products and systems we built and operate.

Sources: Stripe, webhooks (opens in a new tab), Pipedrive, rate limiting (opens in a new tab), HubSpot, private apps (opens in a new tab), HubSpot, date-based API versioning (opens in a new tab), Salesforce, API end of life (opens in a new tab)

What we connect

Systems we have wired up.

Each of these runs today in a client site, in one of our own products or in our own operations.

CRM

Website and quote forms that arrive in the CRM as a lead, after spam has been filtered out.

Payments

Stripe checkout, subscriptions, invoices and identity checks, with the payment events handled as described above.

Email, both ways

Outgoing mail, incoming replies read back into the system, and delivery events that say what bounced.

Google’s data

Search Console, PageSpeed, Analytics and Ads pulled on a schedule into one database.

Alerts for people

Messages to the team’s chat or task board when something needs a decision, limited so the same alert is not repeated.

AI and search APIs

Language models and web-search services called from code.

Beyond one connection

Data pipelines and internal tools.

Data pipelines

Data that moves on its own, every day.

A data pipeline is an integration on a timer: collect from several sources, clean it, put it in one place, and repeat. Ours pulls search data, rankings and speed measurements for the sites in our SEO system into one database, which the dashboards and the monthly reports then read.

The engineering is in the limits. Our hosting platform stops a scheduled job after fifteen minutes, and one of our jobs outgrew that. We split it so it handles a share of the sites each day and every site is still covered each week. A pipeline that silently stops halfway is worse than no pipeline.

  • Scheduled, in small runs that can fail safely
  • One database that dashboards and reports share
  • A run that fails is reported, not skipped

Internal tools

A back office instead of a spreadsheet.

An internal tool is the screen your own team works in: review a queue, approve, send, look up a customer. We built one for a subscription product of ours, where staff handle requests, customers, billing and identity checks, and one for our clients, who see their leads and reports in it.

What a spreadsheet cannot give you is a record. In the back office we built, every change a staff member makes is logged with who made it. In the spreadsheet audits collected by researcher Raymond Panko, errors turned up in 24% of the spreadsheets overall, and in most of those examined in the later, more thorough audits. Fixed fields and a log do not stop mistakes, but they make each one traceable.

  • Roles: who may see and change what
  • Every change logged
  • Built on the same data as your integrations

Sources: Cloudflare, platform limits (opens in a new tab), Panko, Spreadsheet Errors: What We Know (opens in a new tab)

Code or platform

When a platform is enough.

We have built both: an outreach system on an automation platform, with manual checkpoints, and payment and email flows in code. The choice depends on what happens when it goes wrong.

Build it in code

  • Money moves through it, or customer records are created.
  • A duplicate or a lost message would cost you.
  • Volume is high enough for per-step fees to add up.
  • It has to keep working for years, across vendor changes.

A platform is fine

  • A notification or a copy of a record, where a miss is harmless.
  • A process your team wants to change themselves, often.
  • Low volume, with a person checking the result anyway.
  • A trial to see whether the process is worth automating.

Keys and data

Who holds the keys.

An integration holds the credentials to your systems. This is general information, not legal advice.

Credentials in your accounts
API keys are created in your own accounts, so you can revoke one yourself. We keep them as stored secrets, out of the code.
Outside data is not trusted
OWASP’s API security list warns that developers trust data from third-party APIs more than user input. We validate what comes in from another system the same way.
Put the roles in writing
Whoever runs an integration that moves personal data is your processor. The rule is that a processor acts only on your documented instructions and brings in no other party without your authorisation. That belongs in a processing agreement before work starts, and we sign one with you whenever we can reach your customers’ data.

Sources: OWASP API Security Top 10, 2023 (opens in a new tab), ICO, contracts between controllers and processors (opens in a new tab)

How we work

Map, build, break, run.

  1. 01 · Map

    Draw the flow

    Which system sends what, when, and what must never happen twice. The vendors’ limits and version dates go on the same page.

  2. 02 · Build

    Write it in code

    Receive, verify, store, process. Built in code by default, on a platform when that is the better fit.

  3. 03 · Break

    Test the failures

    Send it twice. Send it late. Take the other side down. An integration is tested on what goes wrong, not on the happy path.

  4. 04 · Run

    Hand over, or keep us on

    You get the documentation, and the code on full payment. Watching the vendors’ changes is part of the optional monthly sprint.

One integration

€500 – €5,000

Fixed and agreed before we start. Where it lands depends on the systems involved and on what a failure would cost. For UK clients we quote in pounds on request. See all pricing

FAQ

Questions about System Integrations

Reviewed by Chris de Gruijter,

Tell us which two systems.

Start with the free AI Scan, or tell us what should be connected and ask for a quote.