Gmail automation: labels, filters and the point where an inbox stops coping

For most small businesses the inbox is the real operating system: enquiries arrive there, quotes go out from there, and whether something got done is determined by whether it is still bold. That makes Gmail the most obvious place to start automating and one of the easiest places to make things worse. The difference comes down to a single decision — whether email stays the trigger or becomes the database.

Labels are the automation surface

Almost every useful Gmail automation is built on labels. A label is a piece of state you can attach to a message, and every automation platform can watch for one being applied.

That gives you two distinct patterns, and they behave very differently.

Filter-applied labels fire automatically based on rules — sender, subject, content. Gmail applies the label the moment the message arrives, and your automation reacts. This is how "an enquiry from the website form arrives, create a lead" gets built. Entirely hands-off, and entirely dependent on the filter rule being right.

Manually applied labels are a human saying "this one, do the thing." Someone reads a message, applies a label called quote-requested, and the automation takes over from there. This is dramatically underrated. It puts human judgement exactly where judgement is needed and automates everything after it, which is usually the correct division of labour.

The manual pattern is also the safest way to start. Build the automation, trigger it by hand for two weeks, watch what it does, and only then consider making the trigger automatic.

What Gmail costs in a business context

Google Workspace prices are served in local currency and change with promotional periods, so we list the tier structure and storage rather than a dollar figure that would be wrong for half our readers. Open the pricing page in your own country for the current rate. Source: the vendor’s own pricing page, read August 5, 2026. Prices change — verify before you commit.

Five Gmail automations worth building

• Enquiry to CRM, with the original preserved. A labelled email creates or updates a contact record and attaches the message body. The email remains the source; the CRM becomes searchable. Nobody has to copy anything.
• Attachment extraction. Supplier invoices, signed documents and photos automatically saved to the right folder in Drive, named consistently, and logged. This single automation removes a job nobody enjoys and everybody does badly.
• Structured extraction from unstructured email. Customers write enquiries in prose. An LLM step pulls out name, service, location and preferred date into fields the rest of the workflow can use. This is the highest-leverage AI step in most email workflows.
• Unanswered-message escalation. A scheduled job that finds messages in a label older than a set number of hours with no reply, and posts them to Slack. Catches the thing that falls through the cracks precisely because everyone assumed someone else had it.
• Send-side automation from a template. A record reaching a stage generates a personalised email as a draft for review. Draft, not send — see below.

Draft, do not send

This is the rule that separates email automation people keep from email automation people switch off after an incident.

The temptation is to close the loop entirely: the trigger fires, the email goes out, nobody touches it. It works beautifully in testing. In production, eventually, the automation sends a quote follow-up to someone who already paid, or greets a customer by the merge field rather than their name, or emails the same person four times because a webhook fired four times.

Generating a draft costs the recipient nothing and costs your team about four seconds. In exchange you get a human check on every outbound message, which is worth considerably more than four seconds when the alternative reaches a customer.

The shared inbox ceiling

Most small businesses run a shared address that several people watch. This works until it does not, and the failure is always the same set of symptoms.

• Two people reply to the same customer because there is no way to claim a message.
• Nobody replies because everyone assumed someone else had it — the inverse of the same missing feature.
• You cannot answer "how long do we take to respond?" because an inbox has no concept of first response time.
• When someone leaves, their context leaves. The history of a customer relationship sits in individual heads and individual folders.
• Automations start fighting each other as labels multiply and rules overlap in ways nobody has documented.

You can push the ceiling higher with discipline — a clear label taxonomy, one owner per label, an automation that flags anything untouched. But when a business is consistently missing enquiries, the honest answer is that an inbox is the wrong tool for tracking work. Move the work to a system built for it: a CRM for sales conversations, or a field service platform for jobs. Keep Gmail as the transport it was designed to be.

Practical constraints to know

A few limits that reliably surprise people mid-build.

Sending limits exist and differ between consumer accounts and Workspace accounts. Gmail is not a bulk sending platform, and using it as one risks the deliverability of your ordinary business email — which is a far more expensive problem than the one you were solving. For anything resembling a campaign, use a dedicated email platform.

Filters run on receipt, not retroactively. Creating a filter does not label existing mail unless you explicitly apply it to matching conversations.

Threading can confuse automations. A trigger on "new message" can fire on a reply within a thread it already processed. Deduplicate on message ID rather than assuming.

Apps Script is free and right here. If you are inside Google already, Apps Script can read and label mail on a schedule at no cost, which handles a good deal of this without a paid connector.