Revenue and Conversion Harm

The gap between what happened and what caused it

Revenue fell after the article appeared. That sentence carries two facts and one assumption. The facts are that revenue fell and that the article appeared. The assumption is the word after quietly doing the work of the word because.

A decline that starts the week content shows up is a reason to investigate. It is not a finding. And it says nothing about the number the owner is actually asking for, which is what the business would have earned if the content had never been posted. That is a counterfactual — a claim about a world that did not happen — and nothing measures it. It can only be modeled, from assumptions about growth, margin, retention and how long an effect persists, and each assumption is a place the figure can be pulled apart.

So it is worth being blunt about what this page is for. Working out what an attack cost does not take the content down, does not move a review, and does not change a word on the page that upset you. It produces a record of what happened to the business — worth a great deal on the day counsel asks what the exposure was, and worth nothing at all if what you came here for was to make the page disappear. Whether a statement is legally defamatory is a question for a defamation attorney; what follows is about your own books.

A comparison period chosen after the fact is not a control

The standard damages exhibit in this market is a chart with a vertical line drawn on it. Ninety days before, ninety days after, a percentage in bold at the top. It is persuasive, and it is close to worthless, for reasons that have nothing to do with arithmetic.

A window selected once you already know where the dip sits is not a control. It is the dip, chosen. Pick the span that makes the decline largest and what you have measured is your own selection. Move the same window four weeks in either direction and the figure usually changes; if two defensible windows give two different answers, neither one is a measurement.

The second failure is the event date, and it ruins more analyses than anything else. Content that appeared in March may not have ranked until June, may have been syndicated in August, and may have reached one buying segment months before another. Splitting the data on the publication date, when the content was invisible for eleven weeks, compares two periods that have nothing to do with each other. An analysis built on the wrong event date is worse than none at all, because it looks rigorous and it travels — into a board pack, into an insurance notification, into a letter somebody sends.

The honest form of a before-and-after is a stated date with a stated reason for it: the day the content began appearing on the query, or the day the first inbound question arrived, and a note of how that day was fixed.

Everything else that moved the same week

Businesses are not laboratories. In any given month a dozen things change at once, and every one of these moves the same numbers a reputation event moves. Each has to be checked and written down before the event gets credit for anything:

• Seasonality. The same weeks last year, and the year before that. Not last quarter.
• Promotions, pricing and packaging. Yours, and the competitor's — a rival's discount is invisible in your analytics and obvious in your close rate.
• Paid media. Budget shifts, bid strategy changes, creative refreshes, a channel switched off. Paid changes leak into organic and direct numbers constantly.
• A site change made the same week. A migration, a redesign, a template edit, a consent banner, a tag that stopped firing. Tracking discontinuities look exactly like demand collapses and are far more common.
• Sales and operational changes — a departure, a capacity limit, a supply problem, a service failure — and customer concentration, where one large account leaving reads as a market shift.
• The category and the macro picture. Was the whole sector down that quarter?
• Other reputation events — including ones nobody in the building knows about yet.

Then the question that has to be asked first even though it is the least welcome: is the content a cause or a symptom? A wave of bad reviews after a genuine service failure is a record of a decline, not the reason for it. If that is what happened, everything downstream is wrong, and finding out late costs far more than finding out now.

What makes an analysis harder to argue with

Three things separate work that survives scrutiny from work that impresses only the person who commissioned it.

Data that predates the event. Not a report generated afterward covering the earlier period — the numbers as they were recorded at the time, exported to files with dates on them. This is the only ingredient that cannot be obtained later at any price, and it is why the argument for monitoring is really an argument about the record.

Something comparable that was not affected. A second location, a second product line, a segment the content never reached, a brand you also own. The comparison then runs on the difference between the two changes rather than on one line falling, which strips out everything that hit both — the season, the category, the economy, the ad market. The formal methods for this all share one requirement: a series that is genuinely comparable and genuinely unexposed, plus enough history to know what normal looked like. Most businesses find they have one or the other and not both.

A mechanism, not a coincidence. This is the part almost nobody collects and it is worth more than every chart combined. A prospect's email saying they read the piece. A recorded call where the buyer raises it. An RFP where the disqualification note names it. A cancellation reason typed into the CRM. Each one connects the content to a decision by a named human being, which is a different category of evidence from two lines moving in the same month — and they exist only if somebody sets up a way to catch them in the first week.

What the published research on ratings actually found

Two peer-reviewed studies are worth naming, because they are the strongest causal evidence in this area and because knowing what they cover tells you how far a figure can honestly be stretched.

“(1) a one-star increase in Yelp rating leads to a 5-9 percent increase in revenue, (2) this effect is driven by independent restaurants; ratings do not affect restaurants with chain affiliation.”

— Michael Luca, “Reviews, Reputation, and Revenue: The Case of Yelp.com”, Harvard Business School Working Paper 12-016, 2011, revised 2016, read 12 August 2026

The method is why it counts: a regression discontinuity exploiting the rounding of displayed star ratings, with Yelp data merged against restaurant revenue reported to the Washington State Department of Revenue, covering Seattle restaurants from 2003 to 2009. Businesses either side of a rounding boundary are nearly identical except for the number displayed, which is as close to a natural experiment as this subject offers.

“An extra half-star rating causes restaurants to sell out 19 percentage points (49%) more frequently, with larger impacts when alternate information is more scarce.”

— Michael Anderson and Jeremy Magruder, “Learning from the Crowd: Regression Discontinuity Estimates of the Effects of an Online Review Database”, The Economic Journal 122(563), pages 957–989, September 2012, read 12 August 2026

Same rounding trick, different outcome: reservation availability for San Francisco restaurants, July to October 2010.

Now the limits, which are severe. Both are about restaurants, each in one city, and both are more than fifteen years old — the platforms, the result pages and buying behavior have all moved since. Both measure a star rating, not an article, a complaint page or a false statement. One found no effect at all for chain-affiliated businesses, which is a finding about who is exposed rather than a footnote. And nothing in either paper supports running the sign backwards: a study of what a rating increase did is not a study of what a decrease does, and quoting it that way is a misuse a careful opponent will spot at once.

The interest worth disclosing is unusually mild, and stating it is part of the point: these are academic papers, published through a business school and an economics journal with their data sources and method set out for anyone to check, rather than analyses produced for a party trying to establish a loss. Which is exactly why the honest thing is to name what they cover and stop, instead of borrowing their authority for a number about your business.

The measurements you own outright

The material that requires no model and no assumption sits on property you control. It is observation rather than inference, and it is where the work should start:

• Conversion rate on the affected pages, before and after, holding the traffic source constant. If the same channel sends the same visitors and fewer of them buy, that is a fact about your site rather than about the market.
• Win rate and sales-cycle length by segment. Reputation damage often shows up as deals taking three weeks longer, not as deals disappearing.
• Assisted conversions and direct traffic patterns, which pick up the people who searched, read something, and never arrived.
• Repeat and retention rates among customers who already knew you, usually far less affected — and that contrast is itself informative about who saw what.
• Objection tracking. One required field in the CRM asking whether the buyer raised it. Add it today and in ninety days you hold the only dataset in this subject that is both yours and about the mechanism.

Cost belongs on the same ledger and is the part businesses forget to record: staff hours spent on the matter, extra advertising bought to compensate, discounting authorized to close deals that would have closed anyway, recruitment that took longer. Those are actual outflows with invoices and timesheets behind them, documented rather than modeled — which makes them sturdier than any projection of what might have been.

Where I stop, and why the stopping point matters

A firm that quotes a lost-revenue figure off a traffic chart is quoting a number it invented. The arithmetic behind those figures is an assumed click rate multiplied by an assumed search volume multiplied by an assumed conversion rate multiplied by an assumed order value — four uncertainties compounded and presented as one confident dollar amount.

What can be established rigorously is almost always narrower than what the business feels, and saying so is the whole difference between an analysis and a sales pitch. I will describe what moved, what else could have moved it, what was ruled out and how, and what the record shows about people connecting the content to a decision. Where a formal loss figure is needed for a legal proceeding, that is a separate engagement with a written methodology behind it, and it is not what this page offers.

A conclusion here also does not travel back to the content. The review, the article or the thread sits exactly where it was before the analysis started, unchanged by anything discovered about its effects. What changes is that the business now knows what happened to it, in a form somebody else can check — either the most valuable thing on this site or entirely beside the point, depending on what you came looking for.

Frequently Asked Questions

Published 2026-08-12