Competitor price tracking
It reads their pricing pages. No AI invents a number.
Watches what your rivals charge, shows where you sit in that market, and tells you when one of them moves, starts a discount, or drops below you.
How does it read the prices?
With no AI at all. It takes the currency, the plan names, the features and the billing period straight off the page. An AI is only consulted after three failed attempts, and even then its answer has to survive the checks below.
- It throws out things that look like prices but are not. A funding announcement is not a price — "raised", "ARR" and "Series B" are not what you charge.
- It throws out things that look like plan names but are not. "Get started" is not a plan.
- It knows monthly from yearly, so a €99 annual plan is not filed next to a €99 monthly one.
Can it make a price up?
No. Every number has to appear on the page before it is stored. If the figure cannot be found in the text it was read from, it is discarded.
This is the one path where being wrong costs you money directly. You would be changing what you charge on the strength of it, so a plausible-sounding number is worse than no number.
How does it decide whether I should raise my price?
By arithmetic. There is no AI opinion in the recommendation at all, which means the same evidence always produces the same answer and you can check the working.
The reasoning, from a comment in the code:
“A pricing recommendation has to be defensible and reproducible; ‘the model said raise it’ is neither, and the user is being asked to change what they charge.”
| It will tell you one of five things |
|---|
| Add your own price, so it has something to compare against |
| There are not enough public prices to advise on |
| Consider raising |
| You are in the top quarter of the market — check that it is earned |
| You sit mid-market, and nothing indicates a change |
What if there isn't enough data?
It says so. Below three comparable plans it declines to advise rather than producing a confident answer from two data points.
Its confidence follows how much evidence there is, not how big the gap looks. A huge gap measured against one rival is not a strong signal, and it will not present it as one.
What about different currencies?
It never quietly converts them. A rival priced in another currency is shown in that currency, flagged rather than folded into the comparison.
“a ‘cheaper’ rival priced in a different currency is not cheaper, it is unconverted.”
Will the alerts nag me?
Five kinds, each built not to cry wolf: a price drop, a price rise, a new plan, a removed plan, a promotion, and an undercut.
- A promotion alerts when it starts, not every day it continues.
- An undercut alerts only when a rival moves below you having not been below you before. A rival who has always been cheaper would otherwise alert on every run forever, which is exactly how you learn to ignore a channel.
- Suggested competitors go into an approve-or-reject queue, and a rejection is remembered so it stops offering them.
What does it cost?
€0.10 per competitor, per check. Each rival is checked hourly at most and each page read at most once a day, so ten competitors watched daily is about a euro. Unlimited competitors is €129 a month.
Questions
How does it find my competitors?
It suggests them and you approve or reject each one. A rejection sticks, so it learns who you do not consider a rival.
What if a rival hides pricing behind a sales call?
Then there is no public price to read, and it says so rather than estimating one.
Does it work for products, not just software?
Yes. It reads pricing pages, whatever is on them.
Will they know I am watching?
It reads public pages the way any visitor does, at most once a day per page.