Contents
In short. Automate first the process with the shortest payback, not the most annoying one. Cost the manual work per month: runs × (minutes ÷ 60 × cost of an hour + error rate × cost of an error). Then divide the build cost by the saving minus the running cost. In the illustrative example, moving leads into the CRM pays back in 2.1 months, the weekly report in 8.8, and the supplier check in 72.7. This is a method, not a norm: use your own numbers.
Which process to automate first: cost the manual work
Start with the process whose manual work costs most and pays back soonest. Monthly manual cost combines frequency, time and the cost of an error: runs per month × (minutes per run ÷ 60 × cost of an hour + share of runs with an error × cost of one error).
A run is one pass through the process: one lead, one invoice, one report. Cost an hour with taxes and contributions. Put an error in money: for a lost lead, a customer’s expected contribution times the share of leads that reach a sale (the affordable lead cost article shows how).
- List 5-10 repeating processes where people copy, reconcile or forward data.
- Measure each for two weeks: runs, minutes, errors and their cost.
- Work out the monthly manual cost and keep the 3-5 most expensive.
- Drop those with warning signs (next section), then calculate payback for the rest.
What not to automate, even if the process is irritating
Automation locks a process in as it is
Automation locks a process in as it is: it speeds up a clear one and makes a chaotic one chaotic faster. Check five warning signs before calculating.
- No agreed order: everyone works their own way. Write it on a page first.
- Exceptions exceed half of the runs: only the automated share saves money.
- Rare process, cheap error: its cost is close to the running cost of the solution.
- A person decides and the error is expensive: automate preparing data, not the decision.
- Personal data flows through it and it is unclear where it goes: read the data section below first.
Illustrative example: what manual work costs in four candidates
Conditions of the illustrative example; every figure is invented. Staff time costs $30 an hour; whoever builds the solution costs $60. A lost lead costs $22.50: a customer’s expected contribution of $150 times the 15% of leads that reach a sale (figures from the affordable lead cost example).
Row A: 300 × (4 ÷ 60 × $30 + 4% × $22.50) = $870, of which $600 is time and $270 is lost leads.
| Item | Time | Errors |
|---|---|---|
| A. Leads to CRM | $600 | $270 |
| B. Invoices | $320 | $64 |
| C. Weekly report | $480 | $80 |
| D. Supplier check | $180 | $75 |
Illustrative example: the two columns add up to the monthly manual cost
Scroll right to see all columns
| Process | Runs per month | Minutes per run | Error rate | Cost of an error | Manual cost per month |
|---|---|---|---|---|---|
| A. Leads from forms and messengers into the CRM | 300 | 4 | 4% | $22.50 | $870 |
| B. Invoices and contracts | 80 | 8 | 2% | $40 | $384 |
| C. Weekly ad and sales report | 4 | 240 | 10% | $200 | $560 |
| D. Monthly supplier stock check | 1 | 360 | 25% | $300 | $255 |
Will the build pay back: payback, stress test and threshold share
A build estimate is a forecast and it can be too low
The monthly saving is the manual cost times the share automation removes (the rest is exceptions and human checking). The net effect is the saving minus the running cost: subscription plus an hour of support a month. Payback is the build cost divided by the net effect.
Assumptions for A-D: automation share 90%, 80%, 70%, 60%; build 24, 20, 40, 40 hours; subscription $40, $40, $60, $60 a month. Not price lists.
A build estimate is a forecast and it can be too low. Stress test: double the build at the same effect; if payback still fits the horizon (here 12 months), the process passes. For A the build ceiling (12 × net monthly effect) is $8,196 and the $1,440 estimate is under half of it; doubled, payback is 4.2 months. B takes 11.6 months and just passes. C takes 17.6; to pass, its build must fit into 27 hours instead of 40.
Threshold share: the share of manual work automation must remove to pay back in 12 months is (build ÷ 12 + running cost) ÷ manual cost: 25.3% for A, 52.1% for B, 57.1% for C, 125.5% for D. Above 100% is impossible: the supplier check cannot pay back in 12 months however well it is automated.
| Item | As planned | Build twice as expensive |
|---|---|---|
| A. Leads to CRM | 2.1 months | 4.2 months |
| B. Invoices | 5.8 months | 11.6 months |
| C. Weekly report | 8.8 months | 17.6 months |
The supplier check (D) takes 72.7 months and is not shown. The 12-month horizon is chosen for the example
Scroll right to see all columns
| Process | Net effect per month | Build | Payback, months | If the build costs twice as much, months | Verdict |
|---|---|---|---|---|---|
| A. Leads from forms and messengers into the CRM | $683 | $1,440 | 2.1 | 4.2 | Pilot first |
| B. Invoices and contracts | $207.20 | $1,200 | 5.8 | 11.6 | Second in line |
| C. Weekly ad and sales report | $272 | $2,400 | 8.8 | 17.6 | Simplify and recalculate |
| D. Monthly supplier stock check | $33 | $2,400 | 72.7 | 145.5 | Do not automate |
Which type of solution: CRM settings, a no-code service, a script or an AI agent
Take the simplest type that covers the process. Anthropic’s guide to agents of December 19, 2024 says the same: find the simplest solution possible and increase complexity only when needed.
For process A, first check your CRM’s built-in workflows: if leads arrive from one place, they suffice. With several sources you need a service such as n8n. An AI agent suits free text; copying fields has known steps.
Scroll right to see all columns
| Type | What it is | What to check |
|---|---|---|
| CRM settings | For a process that lives inside the CRM. HubSpot: workflows automate your processes; enrollment triggers automatically enroll records that meet criteria | Professional or Enterprise tiers; some features need credits or seats |
| No-code service (n8n, Make, Zapier) | Chain of steps linking a form, a messenger, a CRM and a spreadsheet without development: workflow (n8n), scenario (Make), Zap (Zapier) | Billing unit, data storage, who fixes failures |
| Script | For a task inside Google Sheets and Gmail. Google Apps Script: JavaScript saved to Drive, runs on Google’s servers | Trigger quota 90 minutes a day (consumer) or 6 hours (Workspace); someone must read the code |
| AI agent | For steps that cannot be listed in advance, such as free text. The model directs its own steps and tool use (Anthropic) | Higher latency and cost; human check when errors are expensive |
n8n: what it is, its licence, where to run it and how it differs from Make and Zapier
The n8n documentation calls it a fair-code licensed workflow automation tool that combines AI features with business process automation. A workflow is a set of connected nodes started by a trigger. Fair-code, per faircode.io, is a model rather than a licence: free to use, open source, commercially restricted by its authors.
Scroll right to see all columns
| Service | What is counted | Units for 300 runs |
|---|---|---|
| n8n | One execution of the whole workflow; steps inside and manual test runs are not counted | 300 |
| Make | Credits: by default 1 module operation equals 1 credit | 1,500 (5 modules) |
| Zapier | Tasks: each successful step; triggers, Filter, Paths and Formatter are not counted; the free plan gives 100 tasks a month and two-step Zaps only | 1,200 (4 actions) |
- Sustainable Use License: use or modify the code for internal business, personal or non-commercial purposes; distribute only free of charge and non-commercially. Files with .ee. in the name fall under a separate Enterprise licence.
- Per the FAQ you can run your business on the free version, build automations for clients and charge for setup, but not host n8n as a service where clients build workflows or use features needing an Enterprise key without one. In doubt: license@n8n.io.
- Hosting: your own server (install script, Docker Compose, cloud providers) or n8n Cloud. The free Community edition needs no licence key but lacks, for example, SSO and Git version control.
- Docs warning: n8n recommends self-hosting for expert users; mistakes can mean data loss, security issues and downtime; without server experience, use Cloud.
Where customer data goes: what the rules say and what to check in a service
Leads and chats contain personal data. In the EU, the European Data Protection Board says a processor acts only on the controller’s instructions under a contract, and the controller is liable for its processor’s compliance too. GDPR Article 28 says to use only processors providing sufficient guarantees; Article 44 allows transfers outside the EU only on that chapter’s conditions.
Other countries have their own rules; which apply to you is a question for a lawyer, the one place in this article where you need one.
Find out where the service runs and what it stores. A self-hosted n8n instance keeps execution data for 336 hours (14 days) by default, so your customers’ leads stay in the log. Zapier’s terms (section 4, posted April 8, 2026) ask you to warrant that you are not located in, under the control of, or a national or resident of a country subject to United States sanctions, and will not violate US export laws.
We found no such clause in n8n’s self-serve terms (updated October 26, 2026) or on Make’s terms page, but n8n refers to other documents and we did not read Make’s master agreement PDF. Both publish a data processing agreement. We found no official list of countries where these services take payment: check before building. We do not recommend workarounds.
A 2-4 week pilot: how to run it and when to stop
A pilot tests two assumptions of the calculation: the share handled automatically and the build cost. Take one stream of one process, for example leads from the website only. The 2-4 weeks is the journal’s guideline.
- Measure the manual process for two weeks; for leads, use the 50-enquiry audit from the article on lead handling.
- Write the order of actions on one page, with the exceptions automation will not take.
- Build one stream and switch on error alerts at once: n8n has error workflows, for example an email or a Slack message.
- Run the automation alongside a person for one or two weeks, then switch over and leave a person to check a sample.
- Recalculate the formula on facts: share handled automatically, build hours, running cost.
- Decide: extend, fix or stop.
The candidate card: what to write down first, and what the calculation leaves out
What the calculation leaves out. Saved hours become money only if the freed time goes into work or replaces a hire; otherwise only the saving on errors is real. Speed and quality enter only through the cost of an error. The numbers are invented.
This journal is published with the involvement of WEBNOSTRA, which provides AI automation services. The calculation works without a contractor; check any offer against the same ceiling and threshold share.
Copy the card into a spreadsheet for each process; the example is process A.
Scroll right to see all columns
| What to write down | Where to get it | Example (process A) |
|---|---|---|
| Process and boundaries | First and last action, owner | From a lead arriving to a CRM record |
| Runs per month and minutes per run | Two-week measurement | 300 and 4 |
| Error rate and cost of an error | Edit log; cost in money | 4% and $22.50 |
| Monthly manual cost | The formula from the first section | $870 |
| Build, running cost, automation share | Contractor, service price, pilot | $1,440, $100 a month, 90% |
| Payback and stress test | Build ÷ net effect, also with a doubled build | 2.1 and 4.2 months |
| Data and owner | Which personal data, where stored, who gets alerts | Name, phone, chat; lawyer has looked; head of sales owns it |
Key decisions
- cost the manual work in money and take the process with the shortest payback
- stress-test the calculation with a build twice the plan; skip broken, rare or cheap-to-get-wrong processes
- pick the simplest type of solution, find out where customer data goes, and pilot for 2-4 weeks
What to discuss with your team
- Runs, minutes and errors of every candidate come from a two-week measurement, and an error is priced in money
- Payback, the doubled-build stress test and the threshold share are calculated
- Warning signs are checked, and customer data and the service’s hosting country are accounted for
- The billing unit is in the running cost, and the pilot has stop criteria and a recipient for error alerts
Terms in plain language
- Manual cost
- Monthly cost of a manual process: staff time plus losses from errors
- Automation payback
- Time in which the net effect (saving minus running cost) covers the build cost
- Trigger
- An event that starts the automation: a new lead, a deal stage change, a scheduled time
Sources
- n8n docs: home page, what n8n isdocs.n8n.io · checked 03/11/2026
- n8n docs: understand workflowsdocs.n8n.io · checked 03/11/2026
- n8n docs: Sustainable Use Licensedocs.n8n.io · checked 03/11/2026
- n8n docs: license FAQdocs.n8n.io · checked 03/11/2026
- n8n docs: hosting options and the free Community editiondocs.n8n.io · checked 03/11/2026
- n8n docs: compare editionsdocs.n8n.io · checked 03/11/2026
- n8n docs: self-hosting knowledge prerequisitesdocs.n8n.io · checked 03/11/2026
- n8n docs: understand executions and the quotadocs.n8n.io · checked 03/11/2026
- n8n docs: error handling and error workflowsdocs.n8n.io · checked 03/11/2026
- n8n docs: execution environment variables, data retentiondocs.n8n.io · checked 03/11/2026
- n8n: pricing page, billing by executionsn8n.io · checked 03/11/2026
- n8n: Self-Serve Termsn8n.io · checked 03/11/2026
- n8n: Data Processing Agreementn8n.io · checked 03/11/2026
- faircode.io: what fair-code isfaircode.io · checked 03/11/2026
- Make Help Center: creditshelp.make.com · checked 03/11/2026
- Make: terms and conditionsmake.com · checked 03/11/2026
- Zapier Help: how task usage is measuredhelp.zapier.com · checked 03/11/2026
- Zapier: pricing and the free planzapier.com · checked 03/11/2026
- Zapier: terms of service, section 4 on export controlszapier.com · checked 03/11/2026
- Google Apps Script: overviewdevelopers.google.com · checked 03/11/2026
- Google Apps Script: quotasdevelopers.google.com · checked 03/11/2026
- Anthropic: Building effective agents, December 19, 2024anthropic.com · checked 03/11/2026
- HubSpot Knowledge Base: create workflowsknowledge.hubspot.com · checked 03/11/2026
- amoCRM help: Digital Pipeline (in Russian)amocrm.ru · checked 03/11/2026
- EDPB: data controller or data processor (guide for small business)edpb.europa.eu · checked 03/11/2026
- GDPR Article 28: processor (text of the Regulation)gdpr-info.eu · checked 03/11/2026
- GDPR Article 44: general principle for transfers (text of the Regulation)gdpr-info.eu · checked 03/11/2026