It started with a desire to "outperform every agency." A 24/7 Moscow vet clinic whose mobile teams visit homes with medical equipment had previously worked with two agencies. Those teams had narrowed the keyword list down to 180 high-intent broad phrases like "vet home visit." Then the clinic owner decided to "vacuum up" all search traffic himself.
On paper, the plan looked bulletproof: take the same 180 phrases, switch to a "maximize clicks" strategy with a weekly budget of 1,000,000 ₽, and duplicate the campaign five times to grab five spots in search results. On top of that, he turned off Yandex Metrica based on a stubborn myth that clicks cost less without a tracking counter.
The exact opposite happened. Cost per lead shot up from 6,854 ₽ to 26,878 ₽, while leads crashed from an average of 350 down to 79. With an average visit bill of 20,000–25,000 ₽, every phone call burned through all revenue before the van even started its engine.
Why Five Duplicates and a Blind Algorithm Break Unit Economics
The clinic brought in contextual ad agency MAKO (the case was published on vc.ru) to investigate the crash. Two classic traps surfaced right away. First, you can't capture five search spots for one website using duplicate campaigns. Instead, you trigger campaign learning cannibalization.
Here is why: Yandex Direct's algorithm needs a dense stream of real phone calls to learn what a good customer looks like. When you split that stream across five identical campaigns, each one gets only crumbs of data. The AI decides the campaigns are weak, cuts reach, and drives up click prices. In response, you raise bids and overheat the auction yourself.
Second, turning off Metrica blindfolded the algorithm. Without call data, it simply chased the fastest clickers. Even when the market later cooled down, lowering bids alone couldn't fix the damage.
The owner kept the main ad account for himself and handed MAKO a secondary account with a separate budget of 1,000,000 ₽ per month. The goal was straightforward: build a second steady stream of orders — at least 10 qualified leads per day at up to 4,000 ₽ each.
1,362 Symptom Keywords Instead of 'Vet Home Visit' and a Service Feed
Bidding on "call an emergency vet" is expensive because every competitor fights there. So the team took medical dispatch stats, ran them through AI, and pulled out the exact phrases panicked pet owners type into their phones: "dog breathing heavily / choking," "cat vomiting foam with blood," "dog swallowed foreign object."
Instead of 180 overheated keywords, they gathered 1,362 specific symptom queries. Competition there is much lower, and the owner's intent to call a doctor is 100%. They launched a Unified Performance Campaign (Search and Yandex Advertising Network / RSYa), smart banners, and a Product Campaign. When search volume dipped due to the clinic's main account, they shifted the main focus to the Product Campaign.
The name is misleading, but the format works great for services too via an XML service feed — a catalog file where each procedure is listed like a product card (title, description, price, link, photo). In Yandex Direct's feed documentation, clinics even have a dedicated "Medicine" YML feed, while other industries can use a universal CSV. Direct automatically matches the right service to each search query.
Then came a technical snag. To avoid typing hundreds of cards by hand, the team generated them with AI in a Google Sheet and served the feed from there. Soon the spreadsheet slowed down. Under Yandex rules, a feed URL must respond directly without delays or redirects, so the hiccups caused Direct to cut impressions. Moving the catalog to a static XML file on the server restored impressions immediately.
In the first week, cost per lead dropped to 4,604 ₽, and by the end of the month it hit 3,597 ₽. The account reached 8–10 qualified leads per day. Right at that moment, competitors struck.
100 Spam Leads an Hour: Why They Deleted Every Web Form
One day, 70% of incoming traffic turned into junk: website forms were getting 50–100 fake submissions every hour. This was click fraud and conversion spam — a bot attack from competitors. The real damage isn't just wasted clicks, it's poisoning the auto-bidding strategy: the algorithm sees dozens of "successful" form fills and starts hunting for more bots just like them.
The first instinct was to add SMS phone verification. Fortunately, the team put themselves in the customer's shoes: if your pet is dying in your arms, would you wait for an SMS code? No, you'd close the tab and call someone else.
Instead, they deleted every form on the site, leaving only one contact option: a direct phone call to a doctor. Bots lost their target. To keep signals clean for Direct, they set up call tracking and call tagging via Calltouch. The service swaps phone numbers to log traffic sources, and each call gets tagged as "Qualified lead" (questions about visits and symptoms) or "Spam/Irrelevant" (bots, loan pitches, ads, wrong numbers). Call-center operators handled the tagging to avoid manual bottlenecks.
Using Calltouch's call-to-Metrica integration, the system matches each call to a site visit via ClientID and sends conversion goals only for approved tags (which can be sent within a 20-day window after tagging). Spam tags were grouped into exclusion segments, and junk mobile apps and fraudulent placements were blocked.
In month two, this cleanup produced 150 unique leads, 80% of which were clean qualified leads. Around 400 spam attacks were filtered out without hurting the budget, and cost per lead fell to 2,500 ₽.
A One-Day Landing Page, Honest '30–90 Minutes,' and Calls From Search
By month three, ads hit a wall with the old website: dated design, vague pricing, and a risky promise to "Arrive in 30 minutes." In Moscow traffic, doctors couldn't make it that fast, so angry clients canceled visits and left negative reviews.
The client didn't want to pay for a full redesign, but gave the team free rein. In a single day, they built a new landing page on Tilda:
- put a direct offer on the first screen: "24/7 Emergency Vet Care at Home" and "we'll save your pet";
- replaced "30 minutes" with an honest "30–90 minutes" arrival window;
- published a real price list, moved reviews into the main menu, and added promos for routine services (like spaying discounts);
- generated emotional AI visuals for the site and ads instead of stock photos.
They paired the landing page with a fresh ad account and a clean Metrica counter, free of old bot history. On comparable traffic, call conversion jumped 1.8x, from 2.99% to 5.25%.
Next, they tested a simple idea: in a panic, people will call straight from search results without visiting the site. They upgraded the ads with promo extensions, a Yandex Business profile, and ad contact blocks featuring a call button. According to Yandex's docs, tapping "Call" on a smartphone costs the same as a regular click. During a test check, the site logged 7 phone-number clicks while the call center received 12 real calls — nearly half came straight from search results.
They trained the new ad account in two stages. First, they used micro-conversions — early signs of interest while call volume was still low (viewing several pages, spending over two minutes on site, clicking the phone number). Once Calltouch gathered enough data, they switched to paying for real leads. They also captured warm traffic with retargeting and a helpful article (for example, on tick bites) with clinic contacts.
The Full Picture
Keep the scope of this case in mind: the clinic's name isn't disclosed, some figures are averages, and the work ran in a secondary ad account alongside the main one. It's fair to compare results against both baselines: before the owner's experiment and at the peak of the crisis.
How cost per qualified lead changed in Yandex Direct
The chart shows where the biggest gains lived. Dropping duplicate campaigns, restoring Metrica, and moving out of head-on bidding wars into symptom keywords and a product feed drove the sharpest price drop. Eliminating web forms to stop bots delivered the second wave of savings, and an honest landing page plus direct search calls sealed the final push.
The secondary account reached 10 to 19 qualified leads per day — averaging 15 orders daily without cannibalizing the main account. For any urgent local service — tow trucks, plumbers, emergency locksmiths — this playbook turns into a clear checklist.
Crisis checklist for urgent local services
- Remove duplicate campaigns and verify Metrica is sending data so you don't fragment algorithm learning.
- Break broad keywords down into specific symptoms, breakdowns, and emergencies customers search for.
- Package services into a YML/XML or CSV product feed and serve it as a static server file, not from Google Sheets.
- During a bot attack on urgent demand, remove web forms instead of adding SMS CAPTCHAs; connect call tracking with tagging and exclude spam segments.
- Link Yandex Business for direct calls from search results and remove unrealistic arrival-time promises from your site.
When your main ad account hits a ceiling, you don't always need to tear down a working setup. Sometimes it's smarter to carve out a test budget, build an honest landing page in a day around specific pain points, and train a fresh counter on clean phone calls.

