--- title: "How to choose an automation platform | JustAutomate" description: "Choose an automation platform by process, not by brand: three processes on paper, five decisive questions, the per-operation billing trap and a real pilot." url: "https://justautomate.ai/en/guides/how-to-choose-automation-platform/" lang: en --- Tools and security # How to choose an automation platform? start with three processes, not with tool names The tool name is the last decision, not the first. Before you open any price list, write down the three processes you want to run first, then check five things: where the data is allowed to sit, how many operations you will run each month, who will maintain it, whether a ready connector exists for the system you already pay for, and what leaving costs. The method is below, including the billing trap that hits loops over records. 9 minutes read - 3 processeswhat we write down in a workshop before any platform gets named; one process runs anywhere and settles nothing - 5 questionswhere the data may sit, how many operations a month, who maintains it, is there a connector, what does leaving cost; each one can strike a candidate off on its own - 1,000 operationswhat the free Make plan allows per month, and one module call is one operation; a loop over order lines multiplies that bill, per the Make price list - 1 to 10 working dayshow long it takes us to get a first cloud scenario running on real data, the pilot that settles the choice better than any comparison table Short answer ## How do you choose an automation platform for a company? Start with processes, not with a price list. Write down the first three processes you want to run: what triggers each one, how many steps it has, which systems it touches and how often it runs in a month. Without those four numbers, every platform comparison is a conversation about taste. Then ask the five decisive questions: **where the data may sit, how many operations you will run each month, who will maintain the platform, whether a ready connector exists for the system you already use, and what leaving costs.** Four of those five answers come out of your own systems in one afternoon. The fifth is in the vendor's documentation. Finish with a pilot of one process on real data, not on the vendor's demo. The choice is not permanent: a first cloud scenario goes live in 1 to 10 working days, and if after a quarter the tool cannot keep up with your volume, you move the scenarios that gain from the move, not the whole company. ## Why does the choice start with three processes, not with tool names? The conversation usually opens with “which tool is best” when it should open with “what exactly has to happen”. A tool is an answer to a process. Until the process is written down, there is nothing to match it against. Three is the minimum, because one process runs anywhere. Differences between platforms only appear when you put a simple process next to one that loops over records and one that touches personal data. At that point some candidates drop out on their own, before anyone opens a price list. Writing them down takes one afternoon with the person who does the work by hand today. A [process workshop](https://justautomate.ai/en/services/process-workshop/) does exactly that, with one addition: we time the process for a week, because what people declare and what the stopwatch shows keep disagreeing. - Trigger: what starts the process and how the team finds out about it today - Steps: the actions in order, including decisions such as “if the customer is outside the EU” - Systems: every application someone logs into along the way, spreadsheets and inboxes included - Volume: how many times the process ran last month, taken from a system rather than from memory - Loops: how many line items an average record holds, because that number multiplies the later bill - Exceptions: the cases handled differently from the rest, and how often they turn up ## Which five questions decide the platform? You could ask forty technical questions. Five of them decide, and each one can strike a candidate off the list by itself. Collect the answers before you talk to a vendor, not during the call, because a vendor will answer them in the direction that suits the vendor. Order matters. The data question goes first: an answer of “customer personal data cannot leave our own infrastructure” closes the conversation about half the market before anyone counts a single operation. | Question | Where the answer comes from | What it settles | | ---------------------------- | ----------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------- | | Where may the data sit? | your security policy, the record of processing activities, contracts with customers | a hard “the data stays with us” leaves self-hosting or an integration written against an API | | How many operations a month? | runs per month from the system times steps, with the loop counted separately | the result against the price list tells you whether per-operation billing is acceptable | | Who will maintain it? | a first and last name, not a department | no such person rules out the self-hosted option, whatever the invoice says | | Is there a ready connector? | the vendor's integration directory and your own system's API documentation | no connector means an API build and weeks of difference in the timeline | | What does leaving cost? | the export format for flows, the number of scenarios, the time to rebuild them | the more scenarios you run, the harder the platform locks you in | | Who will change the flows? | the person who runs the process by hand today | if a non-technical person owns the changes, a visual editor stops being a nice extra | Put the answers side by side in one table, one column per candidate. A candidate with a single “no” in the first or third row is out, even if it wins the other four. ## Per task or per operation: why does a loop change the bill? Price lists look alike until you check what the unit on the invoice actually is. Some platforms count a task, meaning a single step of a flow. Others count an operation, meaning every module call. Others count a whole flow run, whatever happens inside it. And some count nothing at all, because you are paying for a machine. The difference is invisible on a two-step flow and brutal on a loop. An order with eight line items, where the flow checks stock and writes a spreadsheet row for each line, is one run and sixteen module calls. At a thousand orders a month that is sixteen thousand operations out of a single process. So treat the average number of line items per record as its own input. The [Make price list](https://www.make.com/en/pricing) gives 1,000 operations on the free plan and 10,000 on the cheapest paid one, and the vendor moves those thresholds. Volume counted without loops is often several times too low, and that is the most common reason a plan limit runs out mid-month. | Billing model | Unit on the invoice | When the bill jumps | | --------------------------------- | ------------------------------------------------------------- | ----------------------------------------------------------------------------------- | | Per task | a single step executed by the flow | on long flows: every step you add raises the cost of every run | | Per operation | every module call, including calls inside a loop | on loops over order lines, file rows or records pulled from an API | | Per flow run | one run, whatever the number of steps inside | on high run counts, for example webhooks from a busy shop | | Per machine, self-hosted | the server plus the time of whoever keeps it running | when that person does not exist: the cost comes back as downtime, not as an invoice | | Inside a licence you already hold | nothing on a separate invoice, limits written into the bundle | on scenarios that leave the ecosystem: surcharges for premium connectors | The same flow priced in each of these models gives different amounts and a different ranking of candidates. Run your own volume through all of them instead of comparing entry prices. ## What does it cost to leave an automation platform? Exit cost appears on no price list and decides the argument two years in. Scenarios do not move between platforms by import. Modules carry different names, map data differently and handle errors differently, so a flow gets rebuilt from scratch. The scale is easy enough to estimate: a simple flow is hours of work, a branched one with conditions and exception handling is a few days. Multiply that by the number of scenarios you expect to run in two years, not by the three you are starting with. You can lower that cost on day one. Business logic moved out into an [API integration](https://justautomate.ai/en/services/api-integration/) or into a small service of your own stays yours whatever the platform, and the tool is left doing nothing but wiring. The second move is keeping flow exports in a repository, so a migration has somewhere to start. - Rebuilding scenarios: hours for a simple flow, a few days for a branched one - Re-authenticating every connection and repointing webhooks in the source systems - Running the old and the new version side by side on the same data for a few days before you switch the old one off - Run history stays in the old platform, so any audit of the period before the move can only be read there - Team time to learn a new editor, usually 2 to 4 hours per person ## How do you run a pilot before signing an annual plan? A comparison table will not tell you whether the connector to your system handles the one field the whole process stands on. One process running on real data will. Run the pilot on the simplest of the three processes you wrote down, not on the most important one. With us a first cloud scenario stands up 1 to 10 working days after the workshop, and a flow on your own instance usually 14 days after it, because the machine needs a domain, a certificate and backups first. A week of testing is enough to see how the platform behaves when something fails. 1. ### Pick the pilot process The simplest of the three, one that touches at least two systems and runs every day. A process that fires once a month will not show you anything in a week. 2. ### Build it on real data On a copy, but a real one: empty fields, typos and records nobody planned for. A vendor demo runs on data that never fails, so it tests nothing. 3. ### Break it on purpose Switch off one system, pass a bad token, send a record with a required field missing. Check whether the platform retries, whether it tells a human, and whether the log lets you reconstruct what went wrong. 4. ### Count what the pilot consumed Read how many tasks or operations a week of testing ate, then scale it to a month and to a peak month. Only that number is worth putting next to a price list. 5. ### Write the metrics down before go-live Three to five metrics you will read at 30 and 90 days after go-live, each with the place in a system it is read from. Without them there is nothing to compare against a quarter later. ## Is the platform choice permanent? It is not, and that is the most useful thing to tell anyone who has spent six months comparing tools. A platform is a wiring layer. The processes, the data and the logic stay with you, and wiring can be replaced once it stops fitting. The cost of waiting works differently from the cost of a wrong pick. Three months of analysis is three months of the team still retyping data. A wrong platform costs the rebuild of a few scenarios, usually a few days of work. How to put both sides on your own numbers is in our guide to [automation ROI](https://justautomate.ai/en/guides/automation-roi/). Practical rule: choose for the first three processes, not for a hypothetical version of the company five years out. If in two years your volume or a data requirement pushes you elsewhere, you move the scenarios that gain from the move and leave the rest where they are. **Note.** The most expensive choice is no choice. A company waiting for the right tool pays the full cost of manual work for the whole wait, and then decides on the same three processes it could have written down in week one. ## You have a shortlist: what should you read next? The method on this page narrows the market to two or three candidates. After that the details of a specific pairing decide, and each pairing turns on something different, so we cover them separately. If your shortlist looks nothing like the ones below, go back to the table of five questions. A candidate that survived all five has usually just not been priced on your volume yet. - [Zapier vs Make vs n8n](https://justautomate.ai/en/guides/zapier-vs-make-vs-n8n/): three platforms on the same axes, with a straight recommendation for three situations - [n8n](https://justautomate.ai/en/services/n8n-automation/): your own server, per-machine cost, code inside a node when the built-in modules run out - [Make](https://justautomate.ai/en/services/make-automation/): vendor cloud, per-operation billing, routers and iterators without code - [Zapier](https://justautomate.ai/en/services/zapier-automation/): the largest connector directory, per-task billing, linear flows - [Microsoft 365](https://justautomate.ai/en/integrations/microsoft-365/): when the licences you already pay for cover part of the wiring Common mistakes ## Six mistakes people make choosing an automation platform - Picking a tool before the process is written down You automate steps that should not exist. The platform stops mattering, because every one of them will cement the same mess, only faster. - Volume counted without the loops over records The bill comes out several times lower than the real one, and the plan limit runs out mid-month, usually in the peak month. - No named person responsible for maintenance The self-hosted option works until the first update or the first full disk. Then the process stops and nobody knows whose job it is to bring it back. - Testing on the vendor demo instead of your own data A demo has no empty fields and no typos. A missing field in a connector shows up in production, after the annual plan is signed. - Comparing entry prices instead of total cost A zero-cost licence next to a subscription gives an answer that falls apart at the first server invoice and the first Sunday outage. - Delaying the start until the right tool is chosen The cost of manual work runs for the whole analysis, and the decision lands on the same three processes anyone could have written down in week one. Checklist ## Checklist before you choose an automation platform If any line is a no, the decision has not been counted yet. - The first three processes written down: trigger, steps, systems, volume, exceptions - Runs per month taken from a system, not from the team's memory - Average line items per record counted separately, because that is what multiplies operations - The data residency requirement confirmed with whoever owns GDPR compliance - Connectors for every system on the list checked in the vendor's own directory, not in someone's blog post - One person named, first and last name, as responsible for maintenance - A pilot of one process run on real data, including one deliberate failure - Exit cost estimated: number of scenarios times the time to rebuild one Sources ## Sources 1. [n8n documentation: self-hosting and the Community edition for internal company use](https://docs.n8n.io/) 2. [Make pricing: the operation as a billing unit, the free plan and the paid thresholds](https://www.make.com/en/pricing) 3. [Zapier pricing: the task as a billing unit and the limits inside each plan](https://zapier.com/pricing) 4. [Regulation 2016/679 (GDPR), text in the Official Journal of the European Union: the basis for the data residency question](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32016R0679) 5. [JustAutomate, process workshop: scope, how a week runs and what it costs](https://justautomate.ai/en/services/process-workshop/) Let us start with the processes ## The tool name is the last decision, not the first On a free 30-minute consultation we take one of your processes and walk it through the five questions on this page. If a tool you already pay for is enough, we say so on the call. The process workshop costs EUR 3,500 net per week and ends with a process map, a counted volume and a platform recommendation you can act on without us. Implementation is quoted after the workshop. We run at most two implementations at a time, so dates get agreed up front. [Book a free consultation](https://justautomate.ai/en/contact/) ## Frequently asked questions How do you choose an automation platform for a company? Write down the first three processes you want to run and count the volume of each one, loops over records included. Then answer five questions: where the data may sit, how many operations a month, who maintains the platform, whether a ready connector exists, and what leaving costs. Finish with a pilot of one process on real data. What do I do if I do not know how many operations we will run a month? Take last month's event count straight from a system: orders, tickets or closed opportunities. Multiply it by the number of steps in the flow, and separately by the average number of line items per record if the process loops. Add headroom for testing and for a peak month, because a limit runs out exactly when it hurts most. What is the difference between per-task and per-operation billing? A task is usually a single step executed by a flow; an operation is every module call, including calls inside a loop. On a two-step flow you will not see the difference. On an order with eight line items and two calls per line, one run is sixteen units, so a thousand orders is sixteen thousand out of one process. Do we have to pick one platform for the whole company? No. A common setup is one cloud tool for quick integrations between SaaS systems and a second one, self-hosted, for flows touching personal data or carrying large volumes. The price is two places to maintain, so the split only pays off once one of those requirements is genuinely hard. How much does it cost to change platform after a year? Mostly the time to rebuild scenarios, because import between platforms does not work: modules carry different names and map data differently. A simple flow is hours, a branched one a few days. Add re-authenticating connections, repointing webhooks and a few days of both versions running on the same data. Is a platform without a ready connector to our system out of the running? Not immediately. If the system has documented API, the flow connects through an HTTP request, it just takes longer and needs someone who will read that documentation. Check first whether the API exposes the one field the process stands on, and only then count the difference in the timeline. What does the tool cost, and what does the implementation cost? Licences go to the vendor directly and depend on the billing model and your volume; the n8n Community edition carries no licence fee when you host it yourself, and cloud plans start at the thresholds in each vendor's price list. On our side one price is public: the workshop at EUR 3,500 net per week. Implementation is quoted after the workshop. AI model usage, if a flow calls one, runs EUR 0.005 to 0.07 per query paid to the provider directly. Is it worth waiting for a better tool before automating? No, because the cost of manual work runs for the whole wait and the decision lands on the same three processes anyway. A platform is a wiring layer you can replace: the processes, the data and the logic stay with you. Start with one process and revisit the choice after a quarter, with your own numbers instead of someone else's table. ## Related - [Zapier vs Make vs n8n](https://justautomate.ai/en/guides/zapier-vs-make-vs-n8n/) - [No-code and low-code for business](https://justautomate.ai/en/guides/no-code-low-code-for-business/) - [Automation ROI](https://justautomate.ai/en/guides/automation-roi/) - [Process automation step by step](https://justautomate.ai/en/guides/process-automation-step-by-step/) - [API integration](https://justautomate.ai/en/services/api-integration/) - [Process workshop](https://justautomate.ai/en/services/process-workshop/) --- **JustAutomate** · contact: contact@justautomate.ai · form: https://justautomate.ai/en/contact/