Yandex Metrika: which sources bring inquiries that become sales

Form submissions measure activity, but cannot tell you which sources bring inquiries that become sales. You must link inquiries to CRM outcomes, allow equal processing time and measure missing links separately. Otherwise, a source with faster deals or more complete data gains an advantage without necessarily delivering better inquiries.

This intermediate guide is for advertisers analysing their own product in Russia. The Russian Federation remains the market in every language version; a Russian interface or inquiry does not establish a customer's location. First check eligibility for the specific product, territory, channel and data processing. A country selector or offers listed on a platform do not establish availability for you.

The deliverable is a reproducible source-quality worksheet covering the inquiry population, linkage rate, qualification, confirmed sales and a decision with a named owner. Official pages were checked on October 10, 2026. No private Metrika counter or client CRM was accessed; both tables are educational examples, not screenshots. This worksheet is an editorial recommendation, not a built-in AffiliateCore report or a promise of a ready-made integration.

1. Define the decision the report must support

Start with one question: “Which sources produce more confirmed sales per comparable inquiry after equal observation time?” Choose one product and one inquiry type, such as a service demonstration request. Keep consultation requests separate from direct purchases: their intent and handling differ. Analyse products separately before combining results.

Define an inquiry as a unique accepted request with a stable internal identifier. Exclude agreed tests and proven duplicates. Unsuitable inquiries, refusals and unanswered requests stay in the original population; removing them artificially improves sources with many problematic inquiries. Apply identical exclusion rules to every source and retain the reasons separately.

Assign ownership of definitions to sales and reconciliation to analytics. If teams still disagree about statuses, first use the inquiry review guide. Here, the task is to take agreed outcomes and establish whether they can be compared fairly by source, without redesigning the review queue.

2. Prepare access and check data suitability

You need authorised access to the correct counter, an anonymised CRM export, status-change history and someone familiar with the current matching method. Record the domain, counter ID, time zone, export date and responsible staff in an internal working note. Keep contact details, correspondence and access secrets out of public spreadsheets and partner reports.

Before analysis, confirm that your organisation can use the selected service in the intended way for its product and territory. Separately verify report permissions and the agreed data-handling procedure. Successful login does not replace these checks. Without an authorised CRM-to-analytics link, start with a CRM worksheet and disclose the gap: a precise Metrika source ranking is unavailable.

CRM data transfer and matching have technical conditions. Review the official description with the integration owner. Distinguish technical matching deadlines from your quality-assessment horizon. A long sales cycle does not justify delaying the initial data transfer until a deal closes.

To prepare terms for your own proposal, create an AffiliateCore advertiser profile. Registration does not connect Metrika or confirm eligibility for a product, source or geography. The prepared proposal undergoes the required review and moderation.

3. Separate goals, inquiries and sales

Goal completions, visits with goal completions and unique inquiries are different units. A goal can be completed several times in one visit, as the Metrika definitions explicitly explain. One hundred goal triggers therefore do not automatically mean one hundred prospective customers. Verify the actual goal condition and whether an inquiry was accepted.

Use an authorised test to establish what is recorded: a button click, a successful form response or another step. In the worksheet, confirm a sale through a predefined CRM event, such as payment that remains valid at the checkpoint. An agreed meeting or issued invoice does not become a sale because someone labelled the stage “successful”.

Educational table 1. Worksheet definitions; this is not the Metrika interface.

Express qualification criteria through observable evidence: product fit, a confirmed need and the team's ability to fulfil it. Do not substitute a manager's subjective impression for quality. Store historical milestones separately from current status: an inquiry may qualify and later end in refusal. Consequently, totals for all stages “ever reached” need not equal the inquiry count.

4. Record the Metrika report settings

Open the correct counter and the “CRM order sources” report under “End-to-end analytics”, if data transfer is configured. Consult the report documentation for conditions and definitions. If expected data is absent, check access, the counter and transfer status. An empty report does not prove there were no sales. The documented Russian menu path is «Сквозная аналитика → Источники заказов из CRM»; the English names above explain those labels.

Record the period, source grouping, filters, selected metrics and attribution model. Do not rely on an old saved-report name. The report settings guide explains grouping and metrics. Start with one source level, then add detail after checking totals: splitting immediately across many dimensions obscures small groups and complicates reconciliation.

Recheck attribution especially carefully: some earlier models have been disabled since May 20, 2026. Use the current models, not an old screenshot. Record the exact selected name. If the model changes, treat the calculations as different views of one dataset, not a sudden change in source quality.

Copy each metric's denominator from its definition into your working note. Do not assume every “conversion rate” means sales divided by inquiries. Our external worksheet defines its denominator separately. A difference from the interface percentage is acceptable when explained by units, period or data coverage, rather than concealed through rounding.

5. Reconcile the population before calculating rates

Keep the original anonymised export separate from the processed version. Each inquiry needs an internal key, creation date, product, country established by an agreed business criterion, source with its provenance, status history and a successful-match indicator. Leave unreliable sources as “unknown”. Allocating unknown inquiries proportionally across known sources creates false precision.

Count the entire CRM population, then tests, duplicates and remaining inquiries. Split eligible rows into linked and unlinked. Reconcile both equations: original rows minus documented exclusions equal inquiries analysed; inquiries analysed equal linked plus unlinked inquiries. Check each product and period separately so errors in different groups cannot cancel out.

Do not join tables solely on date and amount: matching values do not establish that records describe the same inquiry. With multiple orders per inquiry, define a rule in advance: count at most one successful outcome per inquiry in the sales rate. Order counts and revenue remain separate analyses. A many-to-many join error can double deals without adding a single buyer.

Use the official goal report guide to check goals. Matching aggregate totals alone does not validate every link. Inspect an authorised anonymised sample of linked records and several exclusions. Flag cases where CRM source provenance differs from Metrika attribution. The report reconciliation article covers common discrepancies.

6. Give every inquiry equal observation time

A cohort groups inquiries created within a defined period. Choose the observation horizon from your own distribution of time to decision, not an attractive round number. Examine which outcomes appear quickly and which arrive late. Assess status-transfer delays separately: a completed sale may not yet appear in the export. No single horizon suits every product.

Our educational example uses inquiries created on September 1–7, 2026, evaluating each exactly 14 days after creation. Export after every window has closed and data has updated. This requires event history: current records on September 21 cannot by themselves reconstruct each inquiry's state at day 14. Without history, disclose the limitation and avoid claiming a strictly comparable cohort.

Convert creation, qualification, payment and cancellation times to one agreed time zone, retaining original timestamps. A sale on day 15 falls outside the 14-day result but belongs in a separate later assessment. It is neither a lost sale nor grounds to rewrite the inquiry's creation date. If a refund arrives later, produce another result version with a new horizon.

Keep this inquiry-creation-date calculation distinct from a calendar selection of visits in the interface. Rows and dates need a consistent relationship; identical date filters do not guarantee it. Nor does segmentation automatically convert one set of units into another. The segmentation guide explains limitations on transferring segments between reports.

7. Calculate the example cohort and check the arithmetic

These figures are fictional, not AffiliateCore, Yandex or channel results. A and B are groups whose source membership is independently stored in CRM; linked rows were reconciled with the selected analytics view. This is necessary to know group size despite missing Metrika links. Without independently recorded membership, unlinked rows remain one unknown group.

Educational table 2. Outcomes at day 14; the unit is a unique inquiry.

Qualification, payment and cancellation here concern linked inquiries only. Cancellations are a subset of previously paid inquiries, and every event falls within the window. Sales still valid equal 27 − 3 = 24 for A and 18 − 0 = 18 for B. Stages overlap; do not add the table rows together.

Linkage rates are 108 / 120 = 90% and 90 / 100 = 90%. Qualification rates among linked inquiries are 54 / 108 = 50% and 63 / 90 = 70%. Valid-sale rates among linked inquiries are 24 / 108 ≈ 22.2% and 18 / 90 = 20%. B's higher qualification rate does not establish better final outcomes. A's higher observed sales rate alone does not demonstrate a lasting advantage.

Consider fictional group C without expanding the table: 80 inquiries, 40 linked, 28 qualified, 12 previously paid and two cancelled. That leaves 10 valid sales, 50% linkage, 70% qualification and 25% sales among linked inquiries. The last figure looks strongest, but excludes half the inquiries. Do not declare C the winner or apply its 25% to the unknown half.

To illustrate bounds, temporarily treat unlinked outcomes as unknown. A's overall sales rate could range from 24 / 120 = 20% to (24 + 12) / 120 = 30%; C's from 10 / 80 = 12.5% to (10 + 40) / 80 = 62.5%. These are arithmetic extremes assuming at most one sale per inquiry, not confidence intervals or forecasts. If CRM already knows unlinked outcomes, use them for the overall CRM result; the limitation concerns allocation to a verified Metrika source.

8. Decide with incomplete data in mind

Start with measurement quality. When linkage is low or uneven, investigate differences in forms, intake methods, integration versions or source-field completion. Avoid universal thresholds such as “90% is always enough”. Both the size of the gap and its relationship to outcomes matter: even a small systematic omission can change the conclusion.

Then consider qualification, valid sales, pending outcomes and handling time together. High qualification with weak final results calls for investigating later stages and comparable criteria. Many pending inquiries suggest waiting or reconsidering the horizon. Manager workloads, product availability and response speed may explain effects mistakenly attributed to a channel.

Assign one action per group: fix measurement, wait for windows to close, check handling or continue observing until the next comparable cohort. Record the rationale, owner and review date. With very few sales, decimal precision does not make a reliable ranking. Causal conclusions require a separate, properly designed experiment; this report describes observed associations.

Add costs only when period, source and cost coverage are complete and compatible. A cheap form submission does not equal a cheap confirmed sale, but budget calculations cannot repair poor data linkage. The initial decision concerns comparison reliability, rather than traffic purchasing or bid setting.

9. Correct common errors before sharing

Do not delete refusals to improve quality, classify unfinished inquiries as refusals because the week ended, or move late sales into new inquiries. Distinguish repeat customers from first-time customers with a separate flag. Never retrospectively change qualification criteria for just one source. Each operation changes the question the percentage answers.

Check provenance before merging identically named sources. A manager-entered field, a visit tag and calculated attribution may mean different things. Retain original values and a separate, agreed mapping dictionary. Version any changes, recalculate comparison periods consistently and preserve earlier results so changes can be explained.

Behaviour recordings and heatmaps can reveal page difficulties; the separate Microsoft Clarity guide addresses that work. Technical offline-conversion transfer also requires a separate task. This article provides no upload code or advertising-audience configuration and makes no claim that AffiliateCore has built-in CRM transfer.

10. Finish with an auditable worksheet

Before handing over the decision, check that:

• The product, Russia as the market, channel eligibility and access rights were verified separately.
• Inquiries, qualification and valid sales are defined; tests and duplicates are explained.
• CRM and Metrika sources are reconciled, or differences explicitly recorded.
• Time zone, creation dates, status history and observation horizon are retained.
• Original rows reconcile with exclusions and the final population.
• Unlinked and pending inquiries appear alongside percentages.
• The actual attribution model is recorded, rather than copied from old instructions.
• Figures, rounding, limitations, owner and next review date are stated.

Save the export version, filter definitions and a short note explaining what is known, what remains unknown and which action the data supports. A colleague should be able to repeat the calculation without accessing customers' private correspondence. Repeat after the agreed interval, comparing groups of equal age and noting product and handling changes.

If you already have a profile, log in to the advertiser account to draft terms for your own proposal for review. Check current requirements and moderation separately. This worksheet helps define quality criteria; it does not guarantee offer approval, geography availability, a Metrika connection or future sales.