Every integration you add is a thing that can fail without telling you.
That is the property that makes connected systems different from the tools they connect. A broken tool announces itself — you open it and something is wrong. A broken integration produces silence, which is indistinguishable from a quiet week.
And the failure is usually discovered by a customer. The form submission that never reached the CRM, the order that never triggered the fulfilment email, the unsubscribe that never propagated and produced a complaint.
This is about connecting a small business stack so that when something breaks — and it will — you find out before your customers do.
The three kinds of connection
They carry different risks and deserve different care.
One-way, read only. A dashboard pulling numbers. Lowest risk — a failure shows a stale figure, and the worst case is a wrong decision made on old data.
One-way, write. A form creating a CRM record, a purchase adding a tag. Medium risk — a failure loses data silently, and the record simply never appears.
Two-way sync. Two systems each updating the other. Highest risk by a wide margin, and worth avoiding wherever a one-way flow will do.
Why two-way is worse: when two systems disagree about a value, something has to decide which wins. A sync that resolves conflicts by "most recently updated" will happily overwrite a correction with the stale value that was touched later, and it does so quietly. Two-way syncs also produce loops, where each system's update triggers the other's.
A useful default: one system owns each piece of data, and everything else reads it.
Deciding what to connect
Three questions before adding any integration.
1. What happens if this silently stops? If the answer is "nothing much for a while", it is a low-risk connection. If the answer is "orders stop being fulfilled", it needs monitoring, not just setting up.
2. Does this replace manual work that actually happens? Automating something nobody does is not a saving — it is a new thing to maintain in exchange for a task that was not being performed.
3. Could a scheduled export do this instead? A weekly CSV is unglamorous, robust, and visible when it fails. For plenty of small-business needs it is the better answer, and it does not sit in the critical path.
Where integrations break
Six mechanisms, and none produces an error message.
1. An API changes. The provider updates their interface and the field you relied on is renamed or removed. The integration keeps running and stops carrying that value.
2. A credential expires or is revoked. Somebody rotates a key, an OAuth grant is withdrawn, an employee leaves and their account — which authorised the connection — is deleted. Connections authorised by a personal account die with the account.
3. A field changes shape. A dropdown gains an option, a required field is added, a date format changes. Records start failing validation and are dropped.
4. Rate limits. A busy day exceeds the allowed call volume and calls are rejected. The failure is proportional to your success, which is the worst possible timing.
5. A queue backs up. The connection works and lags, so data arrives hours late. Nothing is broken and everything is wrong — a delayed purchase event fires the cart recovery email after the customer bought.
6. Somebody changed something. A workflow renamed, a tag retired, a form field removed by a person who did not know it fed something else.
All six share the property that matters: the system reports success while carrying nothing.
Making failures visible
Four measures, in order of value per effort.
1. Alert on failure, never on success. Success notifications get filtered within a fortnight and stop being read, which is how a two-month-old break goes unnoticed. Failure alerts arrive rarely and mean something.
2. Alert on absence, not only on error. This is the one almost nobody sets up and it is what catches silent failure. An integration that should fire daily and has fired nothing in three days deserves a message — that is a break, and no error occurred.
3. Keep a register. One page: every integration, what it connects, what triggers it, who owns it, when it was last confirmed working. Stacks accumulate connections built by people who have left, and nobody removes them because nobody knows what they do.
4. Test quarterly, from the outside. Submit the form, place the order, trigger the condition. Twenty minutes, and it is the only method that catches a break the dashboards report as fine. The same discipline for marketing flows.
The data problem underneath
Connected tools multiply the same person.
A customer exists in your ecommerce platform, your email tool, your CRM and your helpdesk. Four records, four versions of their address, and no agreement about which is right.
Three consequences, and the third is expensive:
They get contacted more than you think. One human, three records, three sends. Your per-contact frequency is not what your reports say, because your reports count records.
Corrections do not propagate. They update their address in one place and the other three keep the old one.
Suppression does not propagate, and this is the serious one. Your email platform knows an address hard bounced or complained. Your CRM does not — so the next import re-adds it, and you mail an address that already bounced or already complained.
That is a self-inflicted deliverability wound, and it is one of the most common. Repeatedly mailing hard-bounced addresses damages sender reputation directly, and re-mailing somebody who complained is how a complaint becomes a pattern. What bounces signal · why the reputation cost compounds.
The fix is structural: suppression lives in one place and every import checks it. Keeping data clean across tools.
What you grant when you connect
Most people click through the permission screen, and the permissions are the security model.
A tool connected to your mail account with sending scope can send as you. A tool connected to your store can read customer data. A tool connected to your site can often write files.
Which means your security perimeter includes every vendor you have ever connected, including the ones you stopped using and never disconnected.
The consequence that matters here: a compromised third-party tool with mail-sending access sends spam from your domain, which lands your domain on a blocklist, after which your order confirmations and password resets stop being delivered. Cleaning up the tool does not clear the listing. The same chain, arriving via a hacked site · what removal involves.
Three habits: review connected apps quarterly, revoke what you no longer use, and never authorise a connection from a personal account that might one day be deleted. The full picture.
A stack that holds up
For a small business, the principles rather than the products.
- One system owns each piece of data. Everything else reads it
- One-way flows wherever possible. Two-way sync only where genuinely required
- Suppression in one place, checked by every import
- Failure and absence alerts, never success alerts
- A register, with an owner per connection
- Quarterly testing from the outside, and quarterly review of connected apps
- Scheduled exports where a real-time connection is not needed
- Nothing irreversible without a person, on any automated path
And keep the count down. Each connection is a permanent maintenance obligation, and a stack of six well-understood connections outperforms twenty nobody can explain.
Frequently asked questions
Why do automations break without any error?
Because the systems report on what they attempted rather than what arrived. An API field gets renamed, a credential is revoked, a record fails validation, a queue lags — and the integration continues running while carrying nothing. Alerting on absence rather than only on error is what catches this.
Should I use two-way sync between my tools?
Avoid it where a one-way flow will do. When two systems disagree about a value, something must decide which wins, and conflict resolution by most-recent-update will overwrite a correction with a stale value that happened to be touched later. Have one system own each piece of data.
How do I know if an integration has stopped working?
You mostly do not, unless you set up an absence alert — a message when something that should fire daily has fired nothing in three days. That plus quarterly testing from the outside, by actually submitting the form or placing the order, catches what dashboards report as fine.
Why do customers get emailed too often?
Usually because one person exists as several records across connected tools, so your per-contact frequency is higher than your reports show. Reports count records, not humans, and each duplicate is another send.
What happens when suppression lists do not sync?
Your email platform knows an address bounced or complained, your CRM does not, and the next import re-adds it. Repeatedly mailing hard-bounced addresses damages sender reputation directly, and re-mailing somebody who complained turns one complaint into a pattern.
What am I granting when I connect a tool?
Usually more than the task needs — a tool connected to your mail account with sending scope can send as you. Your security perimeter includes every vendor you have connected, including ones you stopped using and never revoked.
How many integrations should a small business have?
As few as do real work. Each connection is a permanent maintenance obligation, and six well-understood connections outperform twenty nobody can explain. Where a scheduled export would serve, it is more robust and it fails visibly.
Chapters in this Guide
Free: The 60-Minute Email Authentication Fix
A no-fluff checklist to set up SPF, DKIM & DMARC correctly and pass Gmail & Yahoo's sender requirements.
Muhammad Basim has worked in digital marketing since 2013, focused on email deliverability and AI-assisted content production. He is the author of The Email Deliverability Playbook and The Email Copywriting Playbook.
Related Articles
Does Send Time Optimisation Work?
Send time optimisation models are trained on when recipients opened messages, and open timestamps now include machine activity that has nothing to do with when anyone was reading. Privacy protection pre-fetches message content on its own schedule, regardless of whether the recipient opened it. Security gateways fetch content during pre-delivery scanning, which happens before the […]
Prompting for Marketers: The Part That Actually Matters
Output quality is set mostly by the material you supply, not the phrasing you use. That is the finding people take longest to accept, because it is less interesting than the alternative. A carefully worded request with no context produces a competent generic answer. A plainly worded request with your actual customer language, your positioning […]
What You Grant When You Connect a Tool
Your security perimeter includes every vendor you have ever connected, including the ones you stopped using and never disconnected. A connection is a standing grant. It does not expire because you stopped logging in, it does not lapse because the trial ended, and it does not disappear when you delete the app from your phone. […]