Data Ingestion

The insight generated by Corvidae has three pillars; events, stitching, and modelling.

Events are the individual touchpoints in a customers’ buying cycle. Stitching is a process we go through to knit together these touchpoints – without the need for cookies – to generate a clean thread of each customers’ journey to purchase. Modelling is the goal-scorer of this trifecta, and is a neural network responsible for analysing this matrix of paths-to-purchase which weights the impact of every touchpoint.

The most valuable quality of the data provided by Corvidae is that it is unbiased, and agnostic of traffic source. Our pixel simply records the referrer and passes this information anonymously though our patented data science pipeline. In short; every referrer is considered equally, and no preferential adjustments are made, as you might expect to see when looking at platform data.

Using up to two years of existing analytics data we can train our models to understand the buying patterns of your historic customers for greater predictive accuracy. Google Analytics, Analytics 360 and Adobe Analytics are all viable candidates for backdating.

For other providers, where clickstream data feeds are not supported, we are also able to set up a Corvidae specific analytics platform. Corvidae can utilise existing tagging setups, or a new setup can be created by QueryClick, to feed the required data to our platform. This is supported by an open source tool. We then enhance this offering to replicate the required feeds available in the previously mentioned products.

Beyond clickstream data, we can work with a client to understand the different tracking on their channels, what data is available, and will be valuable in the processing. This allows us to then ingest social media platform data, impression data for different platforms, offline data, CMS data and one-off data loads.

Offline data, if available, can be combined with the online journey to give a much fuller view of the customer journey for certain client businesses. This allows Corvidae to attribute to offline as a channel further improving tracking of the customer journey.


Impression data can come from several different sources, without being as explicit as clicks leading to conversions.

This is where ‘out-of-the-box’ solutions can struggle to give an accurate picture of value. To accurately derive insight from all potential types of impression data, detailed segmentation and cleansing of the existing data needs to be performed before building a Machine Learning model.

TV and Radio Data

Corvidae can split all output analysis very granularly. This is key to accurate attribution analysis for TV & Radio. This channel analysis is also very dependent on us being supplied with ad campaign data regarding times and types of ads. From ingesting and analysing the known campaign data, this can be aligned to the insight we wish to generate from the onsite click data we are evaluating.

The foundation of attributing to these offline channels is to build accurate estimations of visits and conversions at varying time slices. Anything from specific minutes of the day, to specific days of the month. We can also benefit from including at least two years’ historical data, to accurately incorporate seasonality into our insights.

Once this is known and, compared with the ad campaign schedules we receive from the client, the model can begin to learn a lot about your customer’s behaviours in relation to the TV and Radio ads being used. This allows accurate levels of visits and conversions to be attributed to individual instances of ad exposure, including down to the detail of regional ads if appropriate.

Beyond this, we can extrapolate on customer behaviour to give several valuable insights. Common information garnered from this analysis are, response time from customers being exposed to an ad until a notable uplift in relevant factors is observed, and re-attribution of ‘lost’ customers who may be attributed incorrectly to be pulled into the effect of a TV or radio ad.

Social Media Data

Corvidae will analyse different interaction types within social media. In clickstream data, we can specify which conversions have been impacted by social media at a certain point in their journey, by using the referrer to tell the particular social media service used and its effectiveness in driving customers to the site.

This can help to observe the effectiveness of individual post types, timings, and content when attributing value more granularly. Many social media services offer ad impression data, so we may also consider customers who have been exposed to a social media post of ad but have come onto the site through another channel. The effect of this exposure must be included in the attribution process. If there is the available data to marry users of these social media services, with users on a site (which can be performed in several ways) Corvidae can attribute value to any social media interaction.

We can report on effectiveness as granularly as we analyse it, including performing insight generation on the value of several different user behaviour factors. For instance, the value of users seeing the ad and coming directly to the site against those coming from a different channel sometime after. Likelihood to purchase depending on behaviour and then attributing that value very specifically.

Offline Data

It is possible to include any offline data in our model building and include it as a channel in attribution reporting. The way in which this is performed can differ between individual clients.

For instance, we can use customer identifiers which are used offline to consolidate this as part of an overall customer journey, adding to the online information we’ve gathered. This could be a loyalty card scanned at checkout, email address collection by store assistants at checkout, or anything where the data can create a connection between online and offline.

We may also use statistically derived assumptions based on footfall information. For example, we can derive from the online data and the relative footfall in different locations, that ‘X%’ of offline are likely to converge with ‘Y%’ of online users.

In some instances, Corvidae can predict patterns which will help identify different sessions as the same user both online and offline.

This relies on the ability to train our models with sufficient data from both online and offline, so behaviours and patterns can be discovered by the model and used when reporting channel value and generating insight.