Most SaaS comparison pages begin with the same number: the cheapest advertised plan.
That number is easy to collect and often useless.
Zapier counts tasks. Make counts credits. n8n Cloud counts workflow executions. Email tools may bill by contacts, sends or both. WhatsApp platforms add their own subscription to Meta's usage charges. A $9/month badge does not tell you what the same workload will cost.
I rebuilt StackPick around a small public dataset instead of another collection of generic reviews.
The row is more important than the ranking
Each tool now has a machine-readable record with:
- entry price in USD;
- the billing unit;
- whether a real free plan exists;
- published plan tiers;
- verification date;
- the vendor pricing page used as the source.
The current data is available as CSV and JSON. The human-readable landing page explains the verification method.
The important constraint is provenance. A price without a source and date quickly becomes misinformation, so the dataset treats verification metadata as part of the value, not an editorial footnote.
Entry price still does not answer the decision
A usable recommendation needs at least three additional inputs:
- the job to be done;
- expected monthly volume;
- the operator's priority, such as ease, depth or scale.
That became Find My Stack. It filters only partner routes that are currently active, but commission size never enters the ranking score.
The second interface is Switch & Save. A cheaper list price is not a saving if the migration breaks workflows or requires weeks of rebuilding, so every suggested switch links to a staged migration and rollback path.
The third interface is a set of Plan Finders. They answer a narrower question: which published tier actually covers the volume?
What I learned from modeling pricing units
Comparable workloads need explicit assumptions
Ten thousand automation events are not automatically ten thousand billable units on every platform. A scenario must state how a workflow maps to tasks, credits or executions.
“Free” is a property, not a headline
A free trial, a free plan and a sales demo are different states. The dataset stores a free-plan flag separately and the interface labels trials and paid entry tiers honestly.
A negative saving is still a valid result
The switch calculator can recommend staying put. Hiding that outcome would turn a decision tool into an affiliate funnel.
Dated data is more useful than false precision
Pricing changes. A verified date and source URL let a reader challenge or update a row. An undated $29 presented to two decimal places is not more accurate.
Making the dataset discoverable
The landing page now includes Schema.org Dataset and DataDownload metadata for both formats. The CSV remains a plain file that works in a spreadsheet, while JSON keeps full tier arrays for software consumers.
There is also a free embeddable finder and cost-calculator library. The iframe supplies the interactive tool; a normal attribution link outside it gives readers and crawlers a path back to the source.
The test is not page count
The real funnel is:
decision run → partner click → approved referral
More generated pages are not the goal. If the decision interfaces do not create qualified clicks, the next move is to improve the model or distribution — not create hundreds of query variations.
The dataset is open to download and cite with attribution. If you find a stale row, the vendor source and verification date make the correction auditable instead of subjective.
Top comments (0)