What should you automate first — the workflow-automation ROI map
About half your work is technically automatable — and almost none of it is worth automating first. The order is decided by two axes, and the glue-vs-build call is decided by one number: volume.
The pitch for automation is everywhere and mostly useless, because it skips the only question that matters: what do you automate first. McKinsey's standing estimate is that roughly half of all current work activities are technically automatable with already-demonstrated technology [1]. But "technically automatable" is not the same as "worth automating this quarter." Point your first project at the wrong task and you spend real money making a rare chore slightly faster while the thing bleeding an hour a day stays manual.
Done right, the upside is not marginal. Forrester's economic study of one major automation platform put a composite organisation's three-year return at 248 percent, with on the order of 200 hours a year reclaimed per employee on the high-impact use cases [2]. The market is pricing that in: intelligent process automation is on track to roughly triple, from about 14.5 billion dollars in 2024 to nearly 45 billion by 2030 [3].
This piece is the map operators actually use. Two questions decide every automation — how often does this task run, and how much does each run cost you in time and errors. Plot your work on those two axes and the queue orders itself. Then one more decision settles how to build it: glue it together with no-code tools, or build it properly. That call comes down almost entirely to volume.
Start every automation conversation by ranking tasks on two axes, not by what would look most impressive in a demo. The horizontal axis is frequency — how many times a week the task actually runs. The vertical axis is pain — the time each run eats plus the cost of getting it wrong. The top-right quadrant, high-frequency and high-pain, is where you start. Everything else waits its turn or stays manual.
The quick wins are almost always the boring, high-frequency plumbing: moving the same data between two systems, routing a new lead, sending the third follow-up, generating the identical report every Monday. None of it is glamorous; all of it runs dozens of times a week and quietly fails when a human is tired. McKinsey's own breakdown puts data collection and data processing among the most automatable activities of all [1] — which is exactly the work most teams still do by hand.
For operators in this region there are two accelerants the global guides miss. The first is compliance: Egypt's e-invoicing mandate from the Tax Authority [4] turned invoicing from a back-office chore into a structured, machine-readable obligation — which is to say, a process that should be automated end-to-end rather than retyped into a portal. The second is sheer volume: a MENA e-commerce market on track to pass 57 billion dollars by 2029 [5] means order, fulfilment, and support flows that no team can keep handling by hand without drowning.
Once you know what to automate, the next decision is how: glue or build. No-code glue tools are metered per task — cheap and fast at low volume, which makes them the correct first move for a workflow that runs a few dozen times a month. Their cost climbs with every run, and they break silently when a connected app changes a field. A custom automation you own inverts that shape: a higher one-time cost, then a thin maintenance slope and no per-task meter — the same near-flat curve Forrester measured when efficiency gains landed around 200 hours per employee per year [2].
The honest filter is: do not build what you can glue, and do not glue what you should build. Below the crossover — low volume, simple logic, nothing proprietary — glue it and move on; building custom there is over-engineering you will regret maintaining. The build case appears when volume is high, when the logic is genuinely yours, when the data has to stay inside your own systems for AI work like scoring or summarisation, or when a glue chain has already started failing at 2 a.m. and nobody can see why.
There is one anti-pattern that wastes more automation budget than any wrong tool choice: automating a broken process. Automation makes a process faster, not better — point it at a messy workflow and you get the mess faster, with the errors now invisible inside a script. Define and clean the workflow first, then automate it. And resist automating the rare, judgement-heavy task just because it would look clever; that work lives in the bottom-left of the map for a reason.
If half of all work activities are technically automatable, why leave any of it on the table? Every manual task is a recurring tax on payroll and a recurring source of error. Be aggressive: map the whole operation, automate top-down, and treat anything a human does twice the same way as a candidate. The compounding time savings dwarf the build cost.
Automation is brittle and the maintenance debt is hidden. Every glue chain is a thing that breaks when an app updates, every script is something only one person understands. The reported ROI numbers come from vendors selling the tools. Keep humans in the loop, document the process well, and you avoid a graveyard of half-working automations nobody trusts.
Never build what you can glue. No-code platforms ship in an afternoon, need no engineers, and let the operator who knows the workflow own it directly. Custom builds are slow, expensive, and over-engineered for problems a connector solves in ten minutes. Stay on glue until it genuinely cannot do the job.
Camp A is right about the prize and wrong about the order — map frequency against pain and automate the top-right first, not everything at once. Camp C is right at low volume and Camp B is right that an un-owned, undocumented automation is a liability. So: glue the low-volume workflows, build the high-volume ones and anything where the data must stay yours, and never — in either case — automate a process you have not cleaned first.
- 01How do you score "pain" in a way finance will accept — is reclaimed-hours-times-loaded-cost enough, or do you need to price the error risk separately?
- 02At what monthly task volume does the glue-vs-build crossover actually land for your operation, and how much does fragile-connector maintenance move it?
- 03When AI agents can read and act on your systems directly, does the no-code glue layer disappear entirely, or does it become the orchestration layer above the agents?
- 04For a MENA operator, how much of an automation backlog is genuinely bespoke versus compliance plumbing (e-invoicing, VAT, payroll) that should be a reusable regional module?
- 05How do you keep an owned automation from rotting — who maintains it, how is it documented, and what is the honest annual cost of keeping it alive?
- [1]McKinsey Global Institute — A future that works: automation, employment, and productivity (about 50% of work activities technically automatable).
- [2]Forrester — The Total Economic Impact of Microsoft Power Automate (248% three-year ROI, ~200 hours/employee/year reclaimed for a composite organisation).
- [3]Grand View Research — Intelligent Process Automation market (USD 14.55B in 2024 to USD 44.74B by 2030, 22.6% CAGR).
- [4]Egyptian Tax Authority — VAT and e-invoicing filing requirements.
- [5]Digital Commerce 360 — MENA e-commerce market projected to reach 57.8 billion dollars by 2029.
We map your workflows, then automate the ones that actually pay.
We start with the frequency-vs-pain map, glue the low-volume wins, and build the high-volume ones you should own. Fifteen minutes to find the hour-a-day your team is losing to manual work.
Book a free 15-min consultation