Corvina Home: Cutting 12,000 URLs to 4,000 in Eight Weeks
Corvina Home is a home-goods retailer with $24M in annual revenue, 35% of which comes from organic search across 12,000 indexable URLs. In eight weeks it migrates to a new CMS that supports only 4,000 URLs. The SEO lead plans to keep the 500 best-performing pages, 301 the next 1,500 into category pages, and let the remaining 10,000 return 404. Separately, the paid team bids on brand terms at a $0.34 average CPC, and analytics shows 62% of brand ad clicks are followed by an organic click in the same session. Analytics runs consent-mode modeling and reports 18% of conversions as modeled or unattributed. The CMO has one week and wants a single defensible number for organic revenue at risk before she signs off on the migration.
1. Produce the revenue-at-risk number and say what you would change about the migration plan.
Read the worked answer
Never reason from URL counts; reason from revenue deciles. Organic revenue is 0.35 x $24M = $8.4M. A typical retail distribution gives roughly: the 500 kept pages ≈ 78% ($6.55M), the 1,500 redirected pages ≈ 14% ($1.18M), and the 10,000 tail ≈ 8% ($672k).
Build the risk number from three components.
- Kept pages: long-run value intact, but migration mechanics (template changes, broken internal links, render issues, redirect chains) typically cost 10-20% of organic revenue for 2-8 weeks. Call it $650k-$1.3M of temporary loss.
- Redirected pages: a 301 to a category page preserves most link equity but loses topical relevance. Assume 30% long-term value loss on $1.18M = $354k.
- 404s: essentially all of $672k lost, though some demand is recoverable by new tail pages later.
Total: roughly $1.0M-$1.35M at risk, or 12-16% of organic revenue and 4-6% of total revenue. Present it as a range with the assumptions stated, and label the temporary and permanent portions separately.
Now attack the premise. The 4,000-URL limit is a content-model choice, not a physical constraint. Most of the 10,000 tail is almost certainly filter, sort and variant URLs; consolidate them behind a parameterized category template with canonicals, and the real page count may fall under 4,000 without deleting anything. Never blanket-redirect to the homepage - that is treated as a soft 404. Map every redirect one-to-one or one-to-nearest-relevant-category, and keep the old URLs in the sitemap for 60 days after launch while monitoring Search Console coverage.
2. Should Corvina keep bidding on brand at $0.34 CPC given that 62% of brand clicks get an organic click in the same session?
Read the worked answer
The 62% overlap statistic is a trap as usually read. It shows that many brand ad clickers also click organic, not that the paid click was wasted. A share of those users click whichever result appears first; if the ad were off, they would click organic and Corvina would still win the visit. The genuine risk is cannibalization, and the genuine benefit is defense against competitors conquesting the brand term, message control, and incremental clicks from users who never scroll past the ad.
Do the unit economics. At $0.34 CPC, if the brand landing page converts at 6%, cost per order is about $5.67. On a $90 average order with 52% contribution, that is trivially profitable per order if the order is incremental - and a pure loss if it is not.
Test rather than argue. Run a geo-split brand holdout for four weeks: pause brand ads in half of matched markets, then compare total orders (paid plus organic) per market. If total orders fall by less than about 10% of prior paid volume, most paid brand traffic was cannibalizing organic and you can retreat to a defensive floor - top two positions on the three highest-value brand terms only. If total orders fall materially, keep bidding and consider bidding more aggressively on competitor-conquest terms.
Keep the cheap insurance while you test. At this CPC the spend is small relative to the risk of a competitor owning the branded SERP for eight weeks. Do not pause brand cold on the strength of an overlap metric.
3. How do you present the 18% modeled or unattributed conversions to the CMO without overstating precision?
Read the worked answer
Name the mechanism first. Consent-mode modeling estimates conversions from users who declined tracking, using aggregate patterns from consenting users. It is a modeled estimate with its own confidence interval, not an observed count. The 18% is not missing data to be recovered; it is a modeled layer on top of observed data.
Report a range and a comparison, not a point estimate. Present organic revenue as observed plus modeled, with the modeled portion flagged. For the migration decision, run the revenue-at-risk analysis on observed organic revenue only - that produces the conservative, defensible floor. Then state separately what the modeled uplift would add if correct. A CMO can act on $1.0M at risk; she cannot act on a number that silently embeds an 18% modeled component.
Give her a stability check. Compare the modeled share month over month. If it swings from 12% to 24% without a corresponding change in traffic sources, the model is unstable and should not be used in any board-level figure.
Set up the pre/post comparison before migration. Export 90 days of URL-level organic sessions, conversions and revenue now, and define the primary success metric as organic revenue at 8, 30 and 90 days post-launch against forecast. Do not benchmark week one against week one of last year - seasonality plus the migration itself will make that unreadable.