In contextual advertising and performance marketing, the exact same classic corporate thriller plays out year after year. End of the month, CEO's office, reporting meeting. On screen shines a textbook presentation from the Yandex Direct specialist: charts climb steadily upward, the ad campaign in Yandex Direct generated 100 phone calls, and the average nominal cost per lead (CPL) dropped from 1,200 to a pleasant 800 rubles. 80,000 rubles spent, a hundred prospective clients received — the specialist awaits a well-earned bonus.
But next, the head of sales takes the floor, and the atmosphere instantly heats up. It turns out real revenue for the month hasn't grown by a single kopeck, the sales plan is missed, and managers are exhausted and angry. The canonical mutual barrage of accusations begins: the marketer proves they brought "hot targeted traffic," while sales reps just don't know how to close deals. Salespeople fire back that they spent all month picking up the phone only to hear spammers, offers to buy office supplies, and questions from people looking for completely different services.
Internet marketing practitioner Denis Kremnev conducted a fundamental investigation into this problem, listening to hundreds of call recordings and comparing raw ad-account statistics with real CRM data. It turned out that behind the pretty figure of "100 leads," a grandiose analytical illusion almost always hides. Let's dissect the anatomy of phone junk layer by layer and reconfigure the system step by step.
1. Conflict of Interest: Why a Pretty Direct Report Doesn't Feed the Cash Register
The fundamental cause of the crisis lies in the disconnect between incentive systems and key performance indicators (KPIs). An ad contractor or in-house Direct specialist traditionally reports on upper-funnel stages: clicks, CTR, landing page conversion, and the number of recorded goals. If the contract or regulation specifies a goal to "get calls cheaper than 1,000 rubles," algorithms and the specialist do exactly what they were asked: they optimize campaigns for gross volume of incoming rings.
Meanwhile, the business owner cares exclusively about the bottom floor of the funnel: gross margin profit, total payments received, and the real cost of acquiring a paying customer (CAC). When there is no end-to-end link between these two worlds, marketing starts feeding on cheap junk. Yandex's machine learning algorithms see that a certain audience segment eagerly clicks the call button on the site and start aggressively pouring budget into that exact cluster, even if behind those calls are retirees who misdialed a clinic's number.
The contextual ad specialist sincerely considers their job done: the Direct ad account gleams with a hundred green conversions for the goal "Phone number click" or "Incoming call from call tracking." The report displays a pretty average price — 800 rubles. Yet when you walk over to the cash register, there's no money there. Superficial report analysis in such conditions is fatally dangerous: by lowering cost per click and chasing gross leads, the business subsidizes empty work by the sales department with its own hands, burning ad budget and demotivating top managers who turn from closing major deals into operators of a free information desk.
2. Anatomy of Phone Junk: Where 85 Out of 100 Incoming Calls Actually Go
When analysts take a sample of 100 incoming phone inquiries at a typical B2B or service company and begin listening to audio recordings minute by minute, the glossy picture shatters instantly. Denis Kremnev's investigation shows the real breakdown of calls dissected from actual case studies:
- Repeat calls from existing clients (25–35% of volume): A person already signed a contract or placed an order last week. To check delivery status, ask accounting a question, or find out warehouse hours, they don't look up the old contact in their address book. They simply type the company name into search, click the very first ad placement, and call the tracking number. The manager picks up and hears: "Hello, we ordered rolled metal from you yesterday, invoice number 4512, when is the truck arriving?" The system happily logs a new "lead" for 800 rubles, even though the company just paid for a call from its own active buyer.
- Missed calls and technical routing failures (15–20% of volume): A client called at 13:05 when all managers went to lunch, or the phone system dropped the call due to channel overload. The person waited 25 seconds and hung up. For call tracking, that's a completed lead; for the cash register, it's a pure zero, because calling back three hours later is pointless: the client clicked the next link in search results and ordered from a competitor.
- Non-target inquiries and off-profile requests (20–30% of volume): The company does turnkey premium apartment renovations starting from 3 million rubles, while the caller asks to lay two meters of tile at a dacha for 5,000 rubles. An enterprise sells industrial equipment wholesale, while a private individual calls asking: "Can you sell one fitting and ship it via Russian Post?" People clicked a broad ad without reading price filters and minimum order volumes.
- Non-commercial noise and spam (10–15% of volume): Calls from job seekers ("I saw your phone number online, do you need loaders or drivers?"), pushy bank loan offers, automated survey bots, and wrong numbers.
In the bottom line, out of a hundred victorious calls, between 10 and 15 people make it to the Sales Qualified Lead (SQL) category — a real solvent client with an on-profile need. The real unit economics of the funnel turn out to be six to eight times more expensive than the owner thought.
3. Call-Tracking Blind Spots: Fraud, Dynamic Pool Error, and False Triggers
Most marketers blindly trust numbers in their call-tracking dashboard, forgetting the technology's technical limitations. The main vulnerability of dynamic call tracking is an insufficient number pool.
For the system to accurately tie a call to a specific click and keyword phrase, a separate number must be assigned to each unique visitor on the site for at least 15–20 minutes. If 30 people visit the site simultaneously during peak hours while only 6 numbers are purchased in the pool, a swap chaos begins. A number gets reassigned to a new user before the previous one has finished browsing. As a result, the call is attributed to a completely different ad campaign, throwing off analysts and automated strategies.
The second fatal factor is naive call qualification rules. By default, many agencies set a primitive condition: "Any call lasting longer than 30 seconds is considered targeted." Yet if an angry client spends 40 seconds arguing with a receptionist over missed installation deadlines, or a job applicant dictates their resume in detail, call tracking flags that conversation with a green successful-conversion check!
Added to this is the problem of spam bots and click-fraud. Automated botnets click through ads and mimic activity: they linger on the page, move the cursor, and click the phone number, triggering false fires of Metrica goals.
We analyzed in detail how systemic analytics errors distort business economics in the piece on Yandex Metrica traps and hidden ad campaign losses. Without manual or AI-based tagging of the actual content of dialogues, trusting raw counters is strictly off-limits.
4. Where Direct Burns Budget: The Forced Autotargeting Trap and Junk Placements
If you dissect search query reports of ad campaigns, it becomes clear where this entire avalanche of non-target inquiries comes from. The main culprit of recent seasons is forced search autotargeting in Yandex Direct.
The ad system actively pushes machine learning algorithms, claiming the neural network understands user intent better than a human. In practice, the algorithm latches onto broad associations. If you advertise "frame house construction," autotargeting will gladly show your ad for queries like "DIY frame house blueprints free," "cardboard kids playhouse," or "buy used cabin in village." People click the attractive headline, call to clarify, and hit manager refusals. Money for the click is debited, a lead is logged, value is zero.
A similar catastrophe unfolds in the Yandex Advertising Network (YAN), where budgets are frequently drained on mobile apps and gaming sites. Users accidentally tap a banner during a game or while trying to close a pop-up window. Kids playing mobile runners generate thousands of clicks, and anti-fraud systems don't always catch accidental taps. How to clean up incoming flow and protect budget from junk clicks is detailed in the study on why advertisers pay for queries they didn't choose.
5. Technical Campaign Resuscitation: End-to-End SQL Passing and Deep Negative-Keyword Clustering
To stop burning working capital and force contextual ads to bring buyers, a strict technical reconfiguration across three key directions is required:
Infographic cutaway of a CRM end-to-end analytics diagram: filtering out junk calls via negative keywords and passing SQL status to Yandex Direct with a 21-day window.
- Training automated strategies exclusively on qualified leads: Strictly forbid ad algorithms from optimizing on an initial call or form submission. Set up passing offline conversions from CRM to Yandex Metrica. The mechanics work like this: a call enters the company, the manager opens the deal card in CRM (amoCRM, Bitrix24), asks qualifying questions, and if criteria are met, changes the status to "Qualified Lead (SQL)." Via a webhook or regular API export, the status is sent to Metrica tied to the browser identifier
ClientID(_ym_uid). In Direct strategy settings, this exact status is set as the key goal with an attribution window of up to 21 days. The algorithm stops hunting spammers and starts hunting people who resemble real buyers. - Total negative-keyword filtering of informational and junk clusters: Open the search query report wizard for the last 90 days. Filter out all phrases with 0% conversion and spend over 3,000 rubles. Ruthlessly add entire clusters to negative keywords: "DIY," "on your own," "download," "used," "forum," "reviews," "term paper," "jobs," "salary," "GOST," "manual." In search, disable broad autotargeting categories, leaving only exact and targeted matches.
- Cutting off the active customer base at the impression level: Upload a database of phone numbers and emails of all current clients to Yandex Audiences. Add this segment to campaign bid adjustments with a "-100%" decreasing coefficient. Active clients will no longer see paid search ads and won't click away budget just to call support.
- Filtering YAN placements: Export the placement report and blacklist all apps containing the words
game,app,kids,puzzle, as well as domains with abnormally high CTR (above 3–5%) and zero time on site.
6. Recalculating Unit Economics: Calculating Real Customer Acquisition Cost and Sales SLA
The final step is restoring honest business math. Forget the nominal CPL from reports; start calculating Customer Acquisition Cost (CAC) — the full cost of acquiring a paying client.
Let's run a clear recalculation based on our initial figures:
- Campaign budget spend: 80,000 rubles.
- Total calls in the report: 100 (nominal CPL = 800 rubles).
- After listening and scoring, repeat calls, spam, and junk were filtered out: exactly 15 real qualified leads (SQL) remained.
- Real cost of a qualified lead: 80,000 / 15 = 5,333 rubles (6.6 times higher than the figure in the report!).
- Sales department conversion from qualified inquiry to paid contract: 33% (one in three).
- Total real buyers acquired: 5 people.
- Real Customer Acquisition Cost (CAC): 80,000 / 5 = 16,000 rubles.
Comparing the two pictures clearly shows the chasm in economics:
- Per the Direct specialist's report: A lead costs 800 rubles, site conversion looks high, creating the illusion of a hyper-effective campaign requiring immediate budget scaling.
- In business reality: Each real buyer costs 16,000 rubles, the load on managers is colossal, and deal margin is almost completely burned at the advertising stage.
Now the owner sees truthful economics. If the average service ticket is 50,000 rubles and gross margin is 40% (20,000 rubles gross profit), paying 16,000 rubles to acquire a client means operating on the edge of breakeven. Had we relied on the reported CPL of 800 rubles, we would have tripled the budget and driven the business into a deep cash gap. A 21-day attribution window when passing offline conversions lets algorithms account for deals with a long decision cycle and find the most solvent clients.
To keep the funnel running like clockwork, lock in a unified marketing and sales interaction regulation (SLA):
- The 15-minute rule: Any incoming inquiry or missed call must be processed by a manager within 15 minutes during business hours. An hour's delay cuts the probability of closing the deal fourfold, and after 4 hours the client already forgets which exact form they filled out.
- Unified disqualification scale: A manager must record a clear rejection reason in CRM for every non-target call (too expensive, wrong region, spam, job seeker, junk). The marketer reviews this log weekly to clean queries.
- Bonus synchronization: A marketer's bonus should depend not on the number of abstract leads, but on the volume of qualified pipeline in rubles handed over to the sales department.
When marketing and sales look into the same CRM system and operate with honest unit-economics figures, illusions evaporate, and the ad budget transforms from a black hole into a predictable tool for generating net profit.

