# Call tracking for real estate developers and AI speech analytics for calls

Author: Ruslan Matveev · Date: 2026-07-01 · Source: https://matveo.com/en/blog/kolltreking-i-rechevaya-analitika · Russian original: https://matveo.com/blog/kolltreking-i-rechevaya-analitika

Call tracking for developers: how dynamic call tracking works, what it costs in 2026, and how AI speech analytics reviews 100% of sales team calls.

## Why a developer's analytics is blind without call tracking

Without call tracking, a developer's analytics is blind, because 60 to 70% of first enquiries in real estate come by phone, and to the system all of those enquiries exist as one faceless stream with no source. A purchase worth millions of roubles is the kind where a person wants to hear a live voice: to check the price, availability and mortgage terms. They will fill in a form later, if they fill one in at all. The share of calls in new-build sales has held in this range for years: the more expensive the purchase, the more often the first touch happens by voice. As a result, the marketer sees ten form submissions in the report and confidently optimises campaigns around them, while thirty calls came in alongside them and the report says nothing about where they came from.

Then the arithmetic kicks in. If more than half of enquiries are not tied to a source, analytics rests on the smaller and less typical part of demand: forms are more often left by people browsing from a distance, while a hot buyer prefers to call. Budget decisions made on forms are skewed: the channel that brings calling buyers looks unprofitable and loses money, while the channel with cheap forms gets it. After a quarter of such optimisation the funnel is full of cheap contacts, only a handful reach a viewing, and without call data the cause cannot be found. We covered how to assemble the whole chain from impression to contract in the article on [end-to-end analytics](https://matveo.com/en/blog/skvoznaya-analitika-v-nedvizhimosti) in real estate. Here we look at its most underrated layer: calls.

## How dynamic call tracking works in plain terms

Dynamic call tracking swaps the phone number on the website separately for each visitor: a person from advertising sees one number, a person from organic search sees another, and when a call comes in, the system uses the dialled number to reconstruct the source right down to the campaign, the ad and the search query. A call becomes just as measurable an event as a submitted form. A simple example: two people are looking at the same [residential complex website](https://matveo.com/en/blog/sayt-zhilogo-kompleksa) at the same time, the first came from the query “two-bedroom near the metro”, the second clicked a banner in the Yandex Advertising Network. Each sees their own number, and both calls land in the report with their own sources. The link between session and number is direct, so it is hard to argue with: the source of every call is determined technically, without guesswork.

Technically it is simple. The call tracking platform rents a pool of numbers, and a script on the website shows each visitor their own number from that pool. While the person is on the site, and for a few minutes after they leave, the number is reserved for their session. If they call, the system links the call to the session, passes the source to Yandex Metrica and creates a deal in the CRM with the campaign tag. The buyer notices nothing: they see an ordinary landline or mobile number and speak to the same sales team. Mature platforms have ready-made integrations with Metrica and the common CRMs, so no custom code is needed. The call recording is stored in the deal card, and later it will be useful for speech analytics.

## Static versus dynamic: which to choose

Static call tracking assigns a separate number to a whole channel, permanently: one number is printed on a billboard along the motorway, a second goes into the project listing on a property portal, a third into Yandex Maps. It answers the question of which channel brought the call and stops there. Dynamic tracking distinguishes sources within a channel right down to the keyword, which is why it is the choice for the website and paid traffic. Choosing between them is essentially choosing depth: how far down you want to see the source of a call, and at what level you intend to make budget decisions.

The working setup is almost always a combination. Dynamic tracking goes on all website traffic: search ads, paid social, organic, programmatic. Static numbers cover offline and the platforms where a script cannot be installed: outdoor advertising, print, property portals, maps, directory profiles. In practice a developer ends up with 10 to 20 static numbers on external platforms plus a dynamic pool on the website, and all calls flow into one report and one CRM. The typical trap here is a static number on the website to save money: the “website” channel will appear in the report, but the [search advertising](https://matveo.com/en/blog/kontekstnaya-reklama-novostroyki) inside it stays a black box, and optimising campaigns by calls becomes impossible.

## How many numbers you need and what it costs in 2026

The number of swap numbers is calculated from the site's peak concurrent traffic. A practical guide: a single-project website with up to a thousand visits a day usually needs 7 to 10 numbers, and at three to five thousand visits you need 25 to 40. The operators' formula is a little more complex, it factors in session length and how long a number is held after a visitor leaves, but the order of magnitude is this. If a campaign brings a spike in traffic and the free numbers run out, the system starts showing some visitors the main number, and those calls lose their source. That is why the pool size is reviewed after every noticeable budget increase.

In money terms, in 2026 renting one number costs 300 to 800 ₽ a month depending on the region and number type, and the platforms' own tariffs start from 5,000 to 7,000 ₽. A developer with one project usually fits into 15,000 to 40,000 ₽ a month including numbers and call recording, while a developer with several projects and noticeable traffic pays 60,000 to 120,000 ₽. The ranges depend on the segment and the feature set. Against a project's advertising budget this is a few percent, and treating this line as a place to save is dangerous: without it the remaining ninety-plus percent of the budget is allocated on incomplete data. One misjudged channel costs more in a month than a year's call tracking subscription.

## Typical call tracking implementation mistakes

The most common implementation mistake is skimping on the number pool. When there are more visitors on the site at once than numbers in the pool, the system shows one number to two people at the same time and attributes the call to one of them at random. The statistics look complete and convincing, but the sources in them are mixed up. On the surface everything works, so nobody notices this mistake for months. A pool sized just enough for a quiet month breaks in the first week of a campaign push. Paying for analytics like that is worse than doing without: it creates confidence in wrong numbers.

Then come smaller mistakes with the same effect. The number on the site is embedded as an image or in a third-party widget, and the script does not swap it. After a redesign or a site update the swap stops working, and there is nobody to check: nobody makes a test call from an ad. Forgotten platforms such as maps and aggregators keep the main number, and calls from them fall into direct traffic. And finally, calls are counted but not qualified: in the report a call from a buyer with a budget weighs the same as a call from an estate agent or spam from service vendors. This last problem is solved by speech analytics.

## Speech analytics: AI reviews 100% of calls

New-generation speech analytics for real estate calls transcribes every conversation and hands it to a language model for review. Previously the head of sales physically managed to listen to 5, at most 10% of calls, picking them at random. An LLM reviews 100% and answers specific questions about each conversation: is the client a fit, what is their budget and request, which objections came up, did the manager offer a viewing. The tagging is standardised: the model fills in the same card for every call, and the figures become comparable across managers, campaigns and months. The cost of processing has fallen by an order of magnitude in recent years, transcribing and reviewing one call costs a few roubles, so the “we only listen to a sample” limitation is gone.

AI analysis of sales conversations finds what used to stay invisible. Dropped leads: a real buyer called, named their budget, and was never offered a viewing. Systematic deviations from the script by a particular manager. Recurring objections that marketing cannot see from the ad account: for example, a third of callers ask about the finish, which the landing page says nothing about. Every call gets tags: qualified or not, request class, stage of readiness. In our experience the most valuable finds of the first weeks are precisely the dropped leads: they are put back to work with a follow-up call, and some of them reach a viewing. Sales call analysis stops being a spot check and becomes a full picture, useful both for training managers and for assessing traffic quality.

## The closed loop: the call tag goes back into advertising

A [closed loop](https://matveo.com/en/blog/zamknutyy-kontur-reklama-analitika) is when the tag AI gave a call goes back into advertising as an optimisation signal. The campaign that brings qualified calls gets more budget, the campaign with junk calls loses it, and Yandex Direct's automated strategies learn from the “qualified call” event instead of the “any enquiry” event. The logic is simple: algorithms need an honest signal, and a call tagged by AI is the most honest of the fast signals, because in real estate months pass before a contract, while a call happens on day one. In our case with an international project across four markets, betting on qualified enquiries instead of cheap leads gave a conversion of around ten percent to sale and a fivefold return on traffic.

At Matveo we test this loop in a two-week marketing audit, and calls are a mandatory block in it: we make test calls from different sources, check whether the number is swapped, whether the source reaches the CRM and what happens to the call recording. Most often the test calls alone show where the source is lost: the swap did not fire, the call did not reach the CRM or the recording was not saved. The cost of the audit is credited towards the pilot, so for those who go further the check is effectively free. The output shows what share of enquiries is currently being lost and what effect AI call review would have on your flow. We covered in detail how a signal from sales changes campaigns daily in the article on the closed loop between advertising and analytics.

## Frequently asked questions

Question: Does a developer with one small residential complex need call tracking? Answer: Yes, if at least a few hundred thousand roubles a month go on advertising. With 60 to 70% of enquiries coming by phone, without call tracking you are allocating budget on the smaller part of the data. The cost is 15,000 to 40,000 ₽ a month, and it pays for itself with the first budget shift from an empty channel to a working one. Switching the service off later is easy, but recovering six months of lost statistics is not possible.

Question: Does a swapped number put buyers off? Answer: No. The visitor sees an ordinary landline or mobile number and reaches the same sales team, the swap happens in the site code and is not visible in any way. The only requirement: the numbers must be local to the project's region, people in Yekaterinburg genuinely do not like calling a Moscow number.

Question: How does call review by a language model differ from classic keyword-based speech analytics? Answer: A classic system searches for set words such as “expensive” or “I'll think about it” and often gets the context wrong. A language model understands the meaning of the whole conversation: it tells a polite refusal from real interest, notices that the manager did not offer a viewing, and explains why the lead was lost. The tagging is at the level of a strong supervisor, only across all calls at once.

Question: Can telecom operator reports replace call tracking? Answer: An operator's report shows the number of calls and their duration, but it does not link a call to an advertising source and a website session. Managing advertising needs exactly that link. Operator reports are useful as a control figure: if call tracking sees noticeably fewer calls than the operator, part of the enquiries are bypassing the swap numbers.

Question: How long does it take to implement call tracking and speech analytics? Answer: Basic number swapping on the website goes live in a few days. The full loop with a CRM link, goals in Yandex Metrica and AI call review usually takes two to four weeks to assemble. Discipline takes the longest: managers must work in the CRM, and call tags must genuinely take part in campaign optimisation.