The subscription is the smallest number in this decision. Platform fees for a small Dubai operation land somewhere between AED 40 and AED 400 a month. The cost of owning that same automation through its first year lands between AED 15,000 and AED 60,000 once you add the build, the rebuild after the first vendor change, the WhatsApp message spend, and the hours somebody burns restarting a scenario that stopped on a Friday night. Comparison pages rank for the small number. The big one decides whether the automation was worth building.
Prices below were read off the vendor pages in August 2026. Every one of these companies has changed its billing model at least once in the past two years, so treat the figures as orders of magnitude and check the current slider before you commit a budget.
Why comparing plan prices tells you almost nothing
Zapier's paid entry sits around USD 30 a month billed monthly for 750 tasks, roughly USD 20 on annual billing, with a Team tier near USD 103 a month for 2,000 tasks. Make's Core plan sits near USD 16 a month for 10,000 credits, Pro near USD 28, Teams near USD 51 at the same credit tier. n8n Cloud starts around EUR 24 a month for 2,500 executions.
Divide those out and Zapier's entry plan buys 25 units per dollar while Make's buys 625, so Make looks twenty-five times cheaper. It is not. The three numbers describe three different meters counting three different things, and the same business process produces wildly different readings on each.
Tasks, operations and credits are three different meters
A Zapier task is one action step that finishes successfully. Triggers are free. Filters and paths are free. Steps that never ran because a filter stopped the Zap are free, and Zapier does not bill a step that errored. A Zap with one trigger and eight actions that survives to the end therefore costs eight tasks per record.
Make counts a credit for every module run. It converted operations to credits one for one on 27 August 2025 and kept most actions at a single credit, with AI modules now charging variable amounts based on tokens, pages or file size. The important part is the word "every". A fourteen-module scenario costs fourteen credits per record, whether or not those modules did anything useful.
n8n counts one execution per workflow run, no matter how many nodes are inside it. Forty nodes or four, the meter moves by one.
The meters diverge most sharply on filtering, and almost no comparison page mentions it. Suppose you push 5,000 raw leads a month through a qualification flow and 70 percent get discarded early because they are duplicates, wrong city or missing a phone number.
On Zapier the filter costs nothing and stops the run. Only the 1,500 survivors consume actions, so at six actions each you are at 9,000 tasks.
On Make the filter itself is free, but every module sitting upstream of it bills on all 5,000 records. Trigger, CRM lookup, data store check: that is 15,000 credits before a single lead has been qualified, plus another 13,500 for the survivors. Same logic, roughly 28,500 credits against 9,000 tasks.
Flip the shape and the result flips with it. A workflow that discards nothing and does twenty things to every record is punishing on Zapier's meter and cheap on n8n's. Your process shape decides which meter is kind to you.
What the bill actually looks like at 1,000, 10,000 and 100,000 runs
Take one real Dubai flow: enquiry arrives, deduplicate against the CRM, create or update the deal, send a WhatsApp confirmation, log the row, issue a payment link, schedule a follow-up check. Call it six billable actions on Zapier, fourteen module runs on Make, one execution on n8n.
At 1,000 enquiries a month you need 6,000 tasks, 14,000 credits or 1,000 executions. Every paid Make tier starts at 10,000 credits, so you are already buying up. n8n Starter's 2,500 executions covers you with room to spare. This is the volume where the monthly difference between the three is smaller than one hour of anyone's time, and picking on price is a waste of a decision.
At 10,000 enquiries you need 60,000 tasks, 140,000 credits or 10,000 executions. The gap stops being academic. Zapier's slider goes to two million tasks, so it will happily sell you the capacity, and you will feel it. n8n's per-execution meter is now dramatically cheaper per unit of work than either competitor, because the complexity inside the workflow is free.
At 100,000 enquiries you need 600,000 tasks, 1.4 million credits or 100,000 executions. No-code stopped being the cheap option somewhere on the way here. At this volume the platform fee has overtaken what a maintained piece of code costs to run on a small server, and it keeps climbing while the code does not.
The arithmetic above is optimistic, because failures and replays add to it. Zapier will not bill you for the step that errored, but when you fix the problem and replay, the steps that already succeeded bill again. Make bills every module the failed run touched before it died, then bills the full replay. At a three percent failure rate that is a few percent on top, which is invisible until you read the usage log.
One thing nobody can answer for you: Make's own documentation and its community threads disagree about whether a scheduled scenario that polls and finds nothing consumes an operation. Do not resolve that from a blog post, mine included. Run it for a week and read your own operations log.
What no connector covers in the UAE
WhatsApp is the awkward one, because a connector does exist and it is not enough. Zapier has a WhatsApp Business app, Make has WhatsApp Business Cloud modules, and both send approved template messages perfectly well. What they do not carry is conversation state. Since 1 July 2025 Meta bills per message rather than per 24-hour conversation. UAE marketing templates run near AED 0.15 to 0.20 each, utility templates near AED 0.04 to 0.06, and replies inside an open 24-hour service window are free (rates checked against Meta's own rate card in August 2026; they move without notice, so confirm the current slider before you budget). Getting that economics right means knowing, at the moment you send, whether this contact's window is open, what you already said, and what they answered. A linear scenario does not know any of that. It fires, it charges you, and it forgets. Holding a thread with memory across days is the job of a WhatsApp agent that qualifies and closes in the chat rather than a chain of modules.
Payment gateways are the second gap. I checked both directories in August 2026 and found no first-party app for PayTabs, Telr or Network International in either. Check again before you plan a build, because these things appear without announcement. Until one does, you connect through raw webhooks and HTTP modules, which means signature verification, retry handling and idempotency are your problem. Get idempotency wrong and one gateway retry becomes two deals in the CRM and two confirmation messages to a customer who paid once.
Then the small local details that quietly eat a weekend. UAE phone numbers arrive in at least four shapes: 0501234567 from a web form, 971501234567 from the WhatsApp payload, +971 50 123 4567 from a business card scan, 00971501234567 from an older system. No connector normalises those. Your dedupe step compares strings, decides they are four different people, and you now have four records for one buyer. That fix is eight lines of code, and there is no no-code equivalent that is not a stack of nested text functions nobody will ever be able to edit. Arabic fields bring their own version of the same problem, where trailing whitespace and right-to-left marks break exact matching. And plenty of local CRM and accounting installs sit on-premise or on a regional host with no public endpoint at all, which no cloud platform can reach without a tunnel or a small middleware service in between.
n8n and the price of owning a server
Self-hosted n8n Community Edition is free under its fair-code licence, with unlimited workflows, users and executions. The software is genuinely free. The operation is not.
You are now the one who patches the host, renews certificates, runs Postgres backups and tests that they restore, watches disk fill up with execution logs, and reads release notes before upgrading. Infrastructure for a small workload starts around USD 5 to 10 a month, which is the number everybody quotes. The number nobody quotes is the hour or two a month someone competent spends keeping it healthy, and the bad afternoon when an upgrade breaks a node and every workflow stops. If that person exists on your team, self-hosting is the cheapest serious option available. If the answer is "we will figure it out," pay for cloud.
Who is on call when a scenario dies at 2am
None of these platforms have a duty engineer for your business. They will email you when a run fails. That is the whole service.
What this costs is measured in time to notice, not in subscription. On our own stack we found two sync bugs between payment confirmation and CRM stage, and until proper alerting existed, roughly AED 15,000 of paid bookings had been sitting in the CRM marked unpaid for weeks. The dashboard was green the entire time. The slow-lead maths is similar and better documented: for a typical Dubai SMB, roughly AED 4,100 a month walks out through late and missed responses.
That is why we sell the monitoring rather than the diagram. Putting one process on autopilot with us starts at AED 6,000 to build and AED 1,200 a month to keep alive, and the monthly part is where the failure alerts, the replays and the vendor changes live. Whatever you choose, price that role. If nobody owns it, you have bought a system that works until the first time it does not.
When it is time to write code
A few conditions push a process off no-code, and any one of them is enough on its own.
Volume: past roughly 50,000 runs a month, platform fees have overtaken the cost of a maintained service on a small server.
State: anything that has to remember a conversation, hold a queue, back off and retry on a schedule, or make a decision based on what happened three days ago. Scenario builders are stateless by design and every workaround is a data store bolted on the side.
Reach: a gateway, an on-premise CRM or a bank file format that no connector speaks. You are writing the integration either way, so write it somewhere you can test it.
For that work we quote a fixed scope at a fixed price. An internal tool or a working prototype starts at AED 9,000 and takes about two weeks; an MVP starts at AED 18,000. A live engineer reviews every line the AI writes before it touches customer data, which is the part that separates it from generated code nobody read.
How to pick for your actual situation
Choose Zapier if your workflow is under ten steps, throws away a lot of records early, runs under about 5,000 tasks a month, and the person maintaining it does not want to think about JSON. Its free tier of 100 tasks a month is enough to prove one idea.
Choose Make if you need branching, arrays and real data reshaping, if your scenarios run more than ten modules, and if somebody on the team can read a webhook payload without flinching. Count your pre-filter modules before you believe any budget estimate.
n8n Cloud fits workflows that are heavy per run at moderate run counts. One execution per run regardless of node count is the best deal on the market for that shape of work.
Self-hosted n8n belongs to teams where a named human owns backups and upgrades. Not a role. A name.
Write code when you hit state, volume or reach, and accept that you are buying something you will maintain rather than something you will rent.
Most Dubai businesses I audit end up with two layers rather than one: a no-code platform for the twenty small connections that change every month, and a small piece of custom code for the two or three things that actually carry money. That combination is usually cheaper at year two than either purist position.
