Home / Guides / Automation patterns
Free guide · 22 patterns

The names for what you want automated

Most people know exactly what is going wrong between their tools. What holds them back is not knowing what the fix is called. These are the 22 patterns we use most, in plain English: what each one is, the sign you need it, and a sentence you could send to us or any developer.

Moving data between tools

How a change in one system reaches the others.

Webhook

Also called: event push

The tool calls your automation the moment something happens, instead of your automation asking it over and over.

You need it when
Updates arrive minutes or hours late, or someone hits refresh to see if anything changed.
What to ask for
When a deal is marked won in the CRM, send a webhook that creates the customer and invoice in accounting within a minute.

Incremental sync

Also called: polling with a watermark

On a schedule, ask each system only for what changed since the last run, and remember where you stopped.

You need it when
The tool has no webhooks, exports time out, or reports cap out before you see everything.
What to ask for
Every 15 minutes, pull tickets updated since the last run and keep a local copy current, even past the API's page limit.
Where we've used it
A helpdesk updating about 3,750 tickets a day kept in sync this way, past a 300-page API cap.

Two-way sync

Also called: bidirectional sync

Changes in either system flow to the other, with a written rule for which system wins when both change the same field.

You need it when
The same customer is edited in two places and the copies drift apart.
What to ask for
Keep contacts in sync both ways. The CRM wins on email and phone; billing wins on address and payment terms.

External ID mapping

Each record stores the other system's ID for the same customer, job or ticket, so matching never depends on names being typed the same.

You need it when
Duplicates appear, or "Acme Inc" and "ACME, Inc." are treated as two customers.
What to ask for
Store the billing account ID on every CRM company, and match on it before anything else.

Queue and replay

When a system is down or slow, changes wait in a queue and are applied in order once it is back, instead of being lost.

You need it when
A vendor outage means someone re-keys a day of changes, or nobody notices they were lost.
What to ask for
If billing is unavailable, queue every change and replay them in order when it returns, then report what was replayed.

Retry with backoff

Also called: 429 handling

When an API says "too many requests", wait and try again, waiting longer each time, instead of failing.

You need it when
Syncs fail at busy times with rate-limit errors, or a script works on ten records and breaks on ten thousand.
What to ask for
On a 429 or timeout, retry with exponential backoff up to five times, then alert a named owner.
Where we've used it
Built into the API clients of a real-time data platform we built and run.

Rate-limit budget

Also called: API quota budgeting

Decide in advance how many API calls each job may spend, so the important work never runs out of quota.

You need it when
A nightly report uses up the quota and the daytime sync starts failing.
What to ask for
Give the customer-facing sync first claim on the API quota; reports run only in what is left.
Where we've used it
Quota budgets per call and a priority queue on a data platform handling about 1.9 million events a day.

Keeping data right

How you know the numbers agree, and how changes stay safe.

Reconciliation

Compare two systems record by record, list every place they disagree, and put a price on each gap.

You need it when
What you delivered and what you billed do not match, and nobody can say by how much.
What to ask for
Match every active seat against the invoice lines for each account, both directions, and price each gap per month.
Where we've used it
A six-figure yearly under-billing gap found on one account; each audit cut from two to four hours to under five minutes.

Dry run

Run the change without saving it and show exactly what would be added, changed and deleted.

You need it when
A bulk update or new sync is about to touch live customer records.
What to ask for
Before any bulk change, show me a dry-run summary of every add, change and delete, and wait for my approval.
Where we've used it
Every write we ship runs as a dry run first.

Snapshot and rollback

Save the current state before a change, with a list of exactly how to undo it.

You need it when
A mistake would mean rebuilding settings or records by hand.
What to ask for
Snapshot the affected records before each run and keep a rollback list for 30 days.

Read-back verification

After writing a change, read it back from the system it went into to prove it landed as intended.

You need it when
The tool said "success" but the setting is wrong, or nobody checks until a customer complains.
What to ask for
After the build, read every setting back from the platform and report anything that differs from the plan.
Where we've used it
A 14-site, 166-seat phone system build verified with zero problems on read-back.

Idempotency

Running the same job twice has the same result as running it once: no duplicate invoices, tickets or emails.

You need it when
A rerun after a failure creates duplicates, so people are afraid to rerun anything.
What to ask for
If the job runs twice for the same order, it must not create a second invoice.

Audit log

Every automated change is recorded with what changed, when, why, and which run did it.

You need it when
Someone asks "who changed this?" and the answer is "the integration".
What to ask for
Log every write with the record, the old and new values, the time and the run ID, and keep it for a year.

Chasing people

How follow-up happens without one person sending every reminder.

Escalation ladder

A reminder goes to the owner first, then their lead, then a manager, only while the item keeps slipping.

You need it when
Reminders get ignored, or managers are copied on everything and stop reading.
What to ask for
Remind the ticket owner at 3 days, copy their lead at 7, and their manager at 14, and stop the moment it moves.
Where we've used it
A 12-stage ladder watching about 7,000 projects and 17,000 locations.

One-click ignore

Also called: acknowledge with a reason

Each reminder has a link to say "not this one" with a reason, and the system tracks ignored against acted.

You need it when
People mute the reminders because some are wrong, and the useful ones get lost with them.
What to ask for
Put an ignore link with a reason on every reminder, and report weekly which reminders were acted on.
Where we've used it
Per-ticket ignore links with ignored-against-acted tracking in production.

Approve-or-dispute link

A secure link in an email turns the customer's click into a record, such as an approval or a tracked dispute ticket.

You need it when
Answers arrive as email replies that someone has to read, interpret and type in.
What to ask for
Email each customer their corrected line items with approve and dispute links; disputes open a ticket with the details.
Where we've used it
Token approve-or-dispute links tracked as helpdesk tickets.

Digest

One summary at a set time, sorted by what matters, instead of a stream of separate alerts.

You need it when
Forty notifications a day, so the important one is missed.
What to ask for
Send each owner one 8 am digest of their items, most at stake first, instead of individual alerts.

Seeing and deciding

How the right person sees the right thing, including what AI is allowed to do.

Exception report

Show only what is wrong or at risk, sorted by what is at stake, instead of every record.

You need it when
The dashboard shows totals, but not the twelve accounts that need a call today.
What to ask for
List only open items that break a rule, sorted by monthly value, with a link to each record.
Where we've used it
A queue engine sorting every ticket into respond now, parked, aged or stale.

Health score

A 0 to 100 score per customer, built from several signals and weighted toward recent activity.

You need it when
You find out a customer was unhappy when they cancel.
What to ask for
Score every account nightly from tickets, reply quality and billing disputes, and flag any that drop 15 points in a week.
Where we've used it
1,700+ companies churn-scored nightly, with a calibration check that catches drift.

Watchdog

Also called: heartbeat check

A small check that notices when a sync or feed goes quiet and tells a named person the same day.

You need it when
A connector stopped weeks ago and you found out at month-end.
What to ask for
Alert me if the billing sync has not run in 24 hours or processed fewer records than usual.

Human in the loop

AI drafts the answer or the change; a person approves it before anything is sent or saved.

You need it when
You want AI's speed but cannot risk it emailing a customer or changing a record on its own.
What to ask for
Let the assistant draft corrections and replies, but nothing is sent or changed until a named person approves it.

Grounded answers

Also called: retrieval, RAG

AI answers only from your own documents and data, and shows where each answer came from.

You need it when
Staff ask the same questions all day, and a general chatbot makes up answers.
What to ask for
Answer questions only from our SOPs and help centre, cite the page used, and say "I don't know" otherwise.

Not sure which ones you need?

Tick the tools your business runs on and see the builds we would start with, or tell us what keeps slipping and we'll name the pattern on a 30-minute call.

See ideas for your tools Schedule a 30-minute assessment