How advanced end-to-end analytics helps to boost apartment sales

Advanced end-to-end analytics based on GCP and GMP helped to get 12.5% more apartment buyers at a fixed marketing cost for a leading apartment developer with MAU of 1 million and annual revenue of €1.75 billion

Now we know where the customer is coming from. We tell you how we rock end-to-end analytics in real estate development.

The specialists who came to outline the problems and create a completely new analytics system at the company of the developer were not previously connected to the development. Problems that were faced:

• In real estate, the sales funnel is very long and complex, comparable to B2B sales. The customer journey is 3 to 6 months.

• Today's real estate industry is an omni-channel customer journey where they can learn about a company from anywhere and either go to the website, call the call center or come to the sales office.

• Weak attribution models on the long funnel: too many promotional touches. Single-touch models don't account for a number of channels at all, and many multiple-touch models distort the real value of intermediate campaigns.

• We need an attribution model that can decompose the value of all customer interactions from the first communication to the point of purchase, or show at what point they left and why.

• A lot of customer interactions happen offline, and they need to be found somehow and combined with other digital footprints. With low percentages of activity linking across touchpoints, it's hard to create a single customer profile across all communications.

The main mistake is analyzing customer behavior only late in the process. Many companies do not pay attention to the customer's early interactions with the brand - he saw a billboard in the street or watched a commercial on TV. They start their story only from the moment of "maturation", when the person is already thinking about buying and considering options. This results in a distorted picture and creates the illusion of a short sales funnel, which the client passes through in a month or less.

No one wanted to solve this problem in the real estate industry comprehensively, and some were indignant that we were infringing on "industry standards". At the same time, many critics’ attribution systems simply will not be able to determine when the same lead will resell in another channel and get the same money for it. Or they will simply artificially tweak the effectiveness of certain methods of traffic acquisition.

But that's not all. Even if a set-up that looks like full end-to-end analytics is put together, it can have such a complex data structure that it would be extremely difficult to explore individual aspects of the customer journey.

A full Customer Journey Map includes more stages of a customer's interaction with a company, so the biggest challenge is to go through the entire customer journey, find all the touchpoints, analyze that data, and put it all together. What touchpoints did the customer go through? Did this interaction provide answers to customers’ questions? How did it go, what were the impressions?

It sounds simple, but it's very difficult to do for four reasons:

• There are a lot of digital touchpoint devices.

• The desire of a customer to keep his anonymity in the early stages of the funnel: a person googling "buy an apartment in Toronto" does not want to be bombarded with advertising messages and calls, he does not want to leave his digital footprint.

• Bringing together data from online and offline customer interactions.

• High requirements for data purity and lack of sampling.

The first three points on a long funnel alone can baffle experienced customer data analysts. It takes a nontrivial and complex stack of technologies to properly address these issues.

But most importantly, you need high-level specialists with a broad set of competencies and experience implementing solutions in different business models to solve these problems.

In the past, each project had its own website. Having analyzed the pros and cons, we decided to unite all the projects on one site together with a client-focused CRM system.

One of the problems with the data was that we didn't see the whole customer journey from visiting the site to closing the deal, we didn't analyze variations in the customer journey. To find all the paths, we had to collect information...well, from everywhere.

We had an end-to-end analytics system for performance advertising purposes that could attribute ad campaigns on a last-non-direct-click model to calls. We had Excel reports for marketing on the number of calls, meetings and deals, which our only web analyst (!) spent half a day every day unloading from the CRM.

And there was a mountain of different philosophical questions, but if we decided to build a huge system of analytics where no one had done it before, we had to go all the way with our decision.

Important
It often happens that when employees get access to hi-tech marketing tools, they start experimenting and end up with something horrible, unworkable, and expensive. But we were lucky: our experience with the systems of client data analytics and end-to-end analytics systems was enough to build a working system, avoiding all the problems.

Now let's move on to what we got.

Let's lift the veil of secrecy and outline about a quarter of the real customer data rotation scheme:

This schema shows a little part of our system. And yes, it's very simplified. Technicians wouldn’t approve it, but this is a general explanatory diagram

The new schema, besides the legacy Microsoft Dynamics, consists of two parallel end-to-end analytics systems that make up a single system.

The first is a combination of Google BigQuery and iDataFusion Apps and some self-written solutions. The second is CDXP to create a single customer profile and uses that data in real-time. We say the following without false modesty: such an integrated solution in development does not exist yet. We took the most up-to-date stuff from other markets with proven efficiency. The funny thing is that if you look at the same set-up in the half with iDataFusion Apps, the same scheme can look completely different and much more interesting

Getting and linking client data from CRM, expense and attribution data from iDataFusion Apps by iDataFusion, and call tracking along with data cleansing allowed for detailed funnels to be built for each project as well as for each cohort of clients. All of this formed the basis of the predictive performance analytics system. A separate and challenging piece of work is to properly visualize this amount of data, a task we're still tackling today.

The distribution of all the client data into time cohorts made it possible to more accurately forecast conversions for individual projects based on previous and current advertising activity. With a history of client interactions for ongoing advertising campaigns in our hands, we can tell which portion of clients will most likely sign a contract, and we can also intensify advertising on those properties for which we forecast insufficient numbers of deals.

CDP, by acquiring and aggregating customer interaction data, allowed us to narrowly segment customers and automate customer interactions to improve conversions. The high purity of the data, even for a small number of customers, ensures that the automated system works well enough and that the numbers can be trusted.

We’re learning how to work with all the data we've received and collected. beginning to apply multichannel, multi-funnel attribution data, customer records, and status control data. starting to shape a new understanding of the sales funnel. Specialized "bottlenecks" that we didn't notice before became transparent. Now that we can see them and can automatically register a customer at a critical stage, we can affect them automatically, by means of CRM marketing and retargeting tools.

To date, we have implemented a full range of approaches and tools that have the right to be called end-to-end analytics. We have connected not individual pieces of the funnel, but the entire customer journey with its variability and numerous ROPO transitions. For example, we can understand that the person who went to the sales office for a meeting, even though they don't call us or pick up the phone from a manager, is still active in the funnel because they keep looking at the availability and price of the apartment they're interested in on a regular basis.

The client profile, which used to contain an average of 6 events, now contains 72 events.

Our current analytics track about 170 event types. All the nuances of using the site are covered: scrolls, tabs, displays of significant areas in the screen, and even mouse hovers - effects that appear when you only point your mouse at them. A single profile also collects about twenty typical events for Web personalization, automated communication scripts, and customer reactions to them. All of our MS Dynamics events, of course. Honestly, this is the sore spot we've been debugging so far.

A few months ago our client lifecycle management system started working in debug mode -  we take a great deal of pride in it. To put it very simply, it's something like RFM control, but on a single purchase and with the eventfulness of moving through the segments.

Our database contains several hundred million events, and this is not raw, but specialized, deeply structured data. In order to structure them and "glue" them together correctly, we have a system of seven types of identifiers. For example, the average customer who reaches the end of the funnel has 3.5 cookies and about 15-20 coltracking cookies. And it is not uncommon for a card to have two or more mobile numbers.

We have the most difficult stage of transformation ahead of us - in people's heads. We have just started to use innovative, nonstandard technologies for real estate, and we are already facing the fact that some concepts need to be introduced and justified for several months. It will be especially difficult to explain new metrics to people who have been working in development for a long time and are not used to this.

In addition, there are some new things that can multiply the complexity of even the simplest reports. For example, you can gather analytics on "demand buckets" and build a predictive system on the "flow" of customers from project advertising campaigns to late lead stages, but to other, "demand buckets" related projects.

Cohort conversions, lead reactivation accounting, and A/B testing methodology are all fairly new things to people immersed in the daily routine of development, which consists of calls, meetings and deals.

But thanks to the support of company management, who understands the need for such qualitative transformations, we can move steadily down the road of a true digital revolution in development.

After the first application of the data we have collected and systematized, we will manually begin to make the first approaches in terms of creating smart predictive models based on machine learning. These models will provide additional insights into potential growth points in our funnel.

And, of course, we plan to immerse all of our colleagues in customer analytics - to increase their awareness and improve their interactions with related departments.

"Where are the detailed numbers?" - you ask, and you'd be right. The client lifecycle control system, the mechanics of ROPO-linking and identification in different touch points and systems, the data visualization set-up, the practical experience of using the system by product managers, and many other things were not included in this article.

The fact is that there is no universal recipe for building an end-to-end customer-centric analytics system, especially for development. There is no perfect solution, which will cover 100% of your needs. In our case, however, we can take some other components and use them to build a sufficiently powerful system.

So why did we choose these components? How exactly are they used? Why do we need this? Why this? No, let’s paraphrase it.
What were we working on when we designed the architecture? What goals did we set ourselves and what solutions did we see? These are the most interesting questions. About which, alas, we have not said anything.

My colleagues and I would like to write a series of articles on working with client data in real estate. And if you, the readers, find it useful and interesting - write to us about it. Instead of describing specific know-how in places, we will describe some logical chains, by which you can also correctly identify your problems, and solutions and perhaps come up with your own solutions and push the development of the market, and maybe related areas even further. So please feel free to comment and ask anything of interest.