About Workflow Test Lab: who runs it, how we test, and how the site makes money
Updated 2026-08-23. This page describes the lab itself. Every other page on the site is about a workflow we tested; this one is about how those pages come to exist.
What the site is
Workflow Test Lab publishes small-business automation workflows that we have built, run, and deliberately broken on real accounts before recommending anything. The first subject is client intake for solo consultants and 1-5 person agencies: a form submission that becomes a CRM record and an acknowledgement email, on the free plans of Tally, Make, HubSpot and Brevo.
Each published workflow comes with the same set of pages: the tested workflow itself, what broke and which fixes we tested, the exported scenario you can import, and who should not use the stack at all. The cost calculator is built from the same measurements.
Who runs it
The site is run by Tuomas Talka, an independent operator based in Finland. It is a small, one-person operation, not an agency or a software vendor, and it has no commercial relationship with any of the tools it tests beyond the affiliate programs listed on the disclosure page.
We use AI tooling in the lab. Claude (Anthropic) builds and runs test scenarios, collects the API evidence, and drafts pages under Tuomas's direction. Before a page is queued for approval, a second AI model from a different vendor (OpenAI GPT) audits the draft against the evidence files and flags unsupported statements, which are then fixed; this step is written into our publishing rules. Tuomas reads the rendered page and approves it before it is published. The build system publishes only pages that have an approval file pinned to the approved version of the text; an edited page drops back to draft until it is approved again. We say this here because a site about testing things honestly should be honest about how it is made.
How we test
- Real accounts, real plans. Every workflow runs on accounts we created for the lab, on the plan named on the page (free tiers so far). We do not test in sandboxes or from vendor demos.
- Scripted cases, every observation counted. Each workflow gets a written test matrix (new lead, duplicate, special characters, concurrent submissions, forced failures). The first workflow was run in two full rounds on 2026-08-16; later changes were tested on 2026-08-20 and 2026-08-21. When something was observed only once, the page says "observed once".
- Broken on purpose. We provoke failures (a wrong API path, a duplicate race, an invalid field) to see what the tools do, and publish the failure modes before the fixes.
- Machine-readable evidence. Results are verified through the tools' own APIs where possible: execution IDs and operation counts from Make, records from HubSpot, message IDs from the receiving inbox. Screenshots are kept for the form-submission steps.
- Claim ledgers. Every material statement on a workflow, template, decision or tool page maps to a ledger entry that names its support: observed in our test, taken from a dated vendor document, or an inference labelled as such. Vendor facts carry a "checked on" date and a refresh date.
- Limits stated on the page. What we did not test is listed on each page, not hidden in a footnote.
What we do not do
We do not accept payment for placement, sponsored reviews, or vendor-supplied test results. We do not publish a recommendation before the workflow has passed its test matrix. We do not remove negative findings at a vendor's request; if a vendor disputes a finding, we re-test and publish the re-test.
How the site makes money
Through affiliate links, and only inside content that tested the product in question. Each page that carries such links says so at the top and marks each link next to the link. As of 2026-08-23 we have one active affiliate partnership, with Make (joined 2026-08-22), and its links appear on two pages, each marked next to the link. Every other product link is a plain link, and where an application to a program is pending no links are placed until it is approved. Our published findings about Make, including the failure modes, predate the partnership and have not been changed since it began. The disclosure and privacy page has the details.
Contact
Operator: Tuomas Talka, Finland. Email: hello@workflowtestlab.com. Corrections and re-test requests are welcome; tell us which page and which statement, and we will check it against the evidence and answer.
Site built with a static generator from Markdown files; no cookies, no analytics, no third-party scripts. Hosted on GitHub Pages.