The Research That Remembers Last Week: Comparing Competitor Prices Over Time
Every Monday the same question comes back: what did the competitors change? Ten websites, a pricing page each, a blog that sometimes moves. Reading them takes an hour. Noticing what is different from last week takes longer, because the difference is not on any page — it is between two readings.
Firmspace keeps this work in the Robot Office, the automation section of the CRM. A robot is a colleague with a role, a set of skills and a written working instruction; the researcher named Vera has the skills to read the open web, to browse the firm's own records and to create tasks. She works in the background and reports on the office screen.
Two limits are built in from the start. A robot receives only the tools of its own skills, and those tools still pass through the permissions of the person who hired it. Deletion is not among the skills a robot can ever have: it reads, drafts, plans and creates tasks, but removing records stays a human decision.
The task you give her is where the result is decided. Research the market is not a task; a country, a niche and a question are. A usable instruction names the market, the segment and what to compare — plan names, price per user, what the cheapest paid plan includes, what was announced in the last month. Before going out, the robot reads the firm's own data to understand what the company actually sells, and to whom.
Finding the right competitors is a filter, not a search. The robot runs three to five differently worded queries in the local language and country, then keeps only providers that really operate there and serve a similar customer. A global brand that mentions the country in a footer is not a competitor; a local reseller with a pricing page in the same currency probably is.
On each of the eight to ten most relevant sites she opens the pricing page first — /pricing, /preise, or the link from the home page — and then up to three more pages: features, news, blog. What she collects is named and dated: plan names, price per user per month in the local currency, whether there is a free plan or a trial, the target customer, the features that differ, and announcements with the date they were published.
What she does not do matters as much. She does not estimate a price she could not read, and a page behind a login or a paywall is recorded as not readable instead of being reconstructed from an old memory. She cannot see inside a customer's account or check your own margins; those numbers are not on the web. Every claim in the report can be traced back to a source address.
Then comes the step a one-off report cannot do. The assistant reads its earlier research on the same topic and states what changed: a price that moved, a plan that disappeared, a limit that was added, a competitor that was not there last month. The journal keeps the recent reports — up to sixty are read — and the robot asks for the last few, five by default, not for everything ever written, so the comparison stays about this quarter.
The report is saved in two places, and both are useful. It goes into the research journal with its title, text and source addresses, and it becomes a task in the Tasks section with the full text and the links, created on behalf of the person who asked. That task also appears as a post in the team feed, so a colleague sees that the research happened even if nobody forwards the message.
Delivery follows the same pattern. A compact version goes to the firm's Telegram — the notification bot from the integration settings — with the changes first, then one line per competitor, then the analysis. One message carries up to 3,500 characters, and a longer report is split into several messages in order instead of being cut.
The analysis itself is deliberately short: five to eight points about who is cheaper, which features the firm lacks and which the competitor lacks, which segment each one targets, and what that means as a threat or an opportunity. A page of description is not research; the value is in the comparison and in the date on it.
One detail deserves the owner's attention. Everything read on the web is untrusted data: an instruction hidden on a competitor's page is text to be reported, not an order to be followed. The same rule holds in a conversation — after the assistant has read an outside source, a page, an e-mail or a document, write actions ask for confirmation again even if automatic mode is switched on. Deleting always asks.
Where the robot stops is honest and visible. Public pages only; regional prices differ, so the currency and the country of the search are part of the answer; a comparison table published last year is cited with its date, not as today's price. If the owner wants the firm's own position in the comparison, that is a different task with the firm's own numbers — and those come from the CRM, not from the web.
A realistic start: hire the researcher, give her one country, one niche and a list of five to ten competitors, and ask for the first report with sources. Read it, correct the list — a competitor that is not one is worse than a missing one. Next Monday the task can simply say: compare with the last report and tell me what changed. The second run is where the work starts paying back.
A market snapshot ages in a week. A series of dated reports, each citing its sources, is the only version of this work that stays useful — and it is the reason the journal is part of the product rather than a folder on someone's desktop.