Automate the work around the work.

Most teams don't lose time to their actual job. They lose it to the copying, chasing, formatting and re-entering that sits either side of it.

AI automation should remove work, not add a dashboard

The test for any automation is whether someone's day gets shorter. If the result is a new interface to check, a new alert channel to mute and a new system to maintain, the automation has moved the work rather than removed it.

In practice that means most good automation is invisible. The follow-up email is already drafted in the CRM when the salesperson opens it. The support ticket is already categorised and routed. The monthly report is already assembled and waiting for a comment. Nobody logs into anything new.

It also means the honest scope is usually narrower than the brief. A request to "automate customer service" almost always resolves to two specific things: classify and route incoming messages, and draft replies to the six question types that make up most of the volume. Those two are worth building. The rest of the brief usually isn't, yet.

Workflows worth automating

The ones that come up most, and what "done" looks like for each.

CRM

Enrichment, deduplication, stage hygiene and drafted follow-ups, so the record reflects reality without anyone maintaining it manually.

Email

Triage, classification, drafted responses and extraction of the details that would otherwise be retyped into another system.

Routing, structured capture and clean handover — often the highest-value automation in this region, and the most commonly skipped.

Customer support

Classification, first-draft replies, and surfacing the small number of tickets that genuinely need a person today.

Lead routing

Read, enrich, score and assign an inbound enquiry before anyone opens it, with the reasoning attached.

Content operations

Briefing, drafting, bilingual variants, and the approval trail — with a human owning the final pass.

Reporting

Pulled from the systems the numbers actually live in, assembled on schedule, with drafted commentary and verifiable figures.

eCommerce

Product data cleanup, bilingual descriptions, returns triage and post-purchase messaging that doesn't read like a template.

Internal knowledge

Answering staff questions from your own documents, with citations, instead of the same person being asked every time.

Notion & internal systems

Keeping the workspace people actually plan in synchronised with the systems where the work happens.

Document processing

Extraction and classification from contracts, invoices and forms, including Arabic and mixed-language documents.

Admin tasks

Approvals, onboarding, scheduling and the small recurring work that quietly consumes a team's week.

When not to automate

Cases where I'd recommend against a build, from having watched them fail:

The process changes every few weeks. You'll spend more maintaining the automation than the automation saves.

It runs a handful of times a month. Automation has a fixed cost; low-frequency work rarely clears it.

The process is undocumented and contested. Automating a disputed process just makes the dispute faster.

The real bottleneck is a decision someone is avoiding. Software will not resolve that, and will be blamed for not resolving it.

There's no owner. Automations without an owner degrade silently and are discovered broken months later.

Saying no to these is part of the service. An automation that gets switched off after two months costs more than the manual process it replaced.

The operations work that comes before automating

Half of what gets asked for as automation is really this. It is worth doing on its own, and it is what makes the automation afterwards cheap instead of fragile.

Process audit and mapping

Write down what actually happens, not what the org chart says happens — who touches it, where it waits, and which step everyone quietly works around. Most of the saving is visible before any software is involved.

Tool stack consolidation

Teams accumulate tools faster than they retire them, and end up paying twice for overlapping ones nobody fully adopted. Working out what to keep, merge or drop usually pays for itself before anything is built.

SOPs and documentation

A process that only exists in one person's head cannot be automated, delegated or safely handed over. Writing it down is also the cheapest way to find out that two people were doing it differently.