top of page

The value of a bespoke customer data platform – just how ‘honest’ is your ‘single source of truth’?

Updated: Mar 19

Making customer-focused decisions vital to value creation can be an intimidating task when you feel blind to what is going on within your business.

In my time at Coppett Hill, I’ve become used to hearing the most common concerns of Chief Executives, Chief Marketing Officers and Private Equity value creation leads. Frequent questions include the measurement of customer lifetime value, which prospective customers to prioritise, and how to improve the customer journey to increase conversion. From the experience of my colleagues, I’ve learned that the most common (and arguably most distressing) answer to these questions in a boardroom is ‘we don’t have the data’.

If you happen to be in one of those roles I’ve mentioned above and a question has made its way on to your desk, there is a good chance it is of high strategic importance – which may well require some of the missing data mentioned above.

Why do we ‘not have the data’?

First of all, it is rarely the case that the data you need is not collected at all – almost all aspects of your customer journey, product and transactional relationship with customers and suppliers is digitised in some way. Even if you are after customer feedback or competitor intelligence you can gather with periodic qualitative research or mystery shopping.

Should we clear the first hurdle and have a potential source for our required data, we might be faced with data which we not have confidence in, or even more perplexingly, several different data sources offering contradictory estimates of the same metrics.

Sometimes the problem can be as simple as different systems holding data for different geographies, or different products/service offerings. This confusion can be compounded by a lack of analytical skills within the organisation to delve deeper into the data and search for a rational explanation for these differences.

The question is, when you have several data sources all claiming to be a ‘single source of truth’, which do you choose to place your confidence in? In particular, when each might have its own group of loyal users within the business using it on a daily basis?

From my experience, often the answer has been none of them. We often start our work with clients by building a bespoke ‘customer data platform’ that is a foundation for insights, recommendations and informed decision making.

How can I build a customer data platform I can trust?

1)      Decide what is important.

The first step in this process is to find relevant input data sources that we are going to combine into a centralised database.

The input data sources differ from business to business; however, we would generally expect there to be some versions of ecommerce/sales software (perhaps even transactional data from a finance system), marketing tools such as Google Analytics, marketing automation/email software and customer experience software in most growing businesses. You might also include product usage data (if you sell a software product) or timesheet data (if you sell professional services).

You will want to also compile more ‘fixed’ data or assumptions relevant to your business – think financial inputs like cost of goods sold (COGS), direct staff costs, payment processing costs or perhaps financial forecasts.

Any input data source that you think could conceivably contain some data relevant to metrics you would like to generate, should be included. It’s okay if some things are assumptions where accessing actual data is not justified by the risk-reward, for example allocating payment processing costs at a transactional level.

2)      What do customer interactions with your business look like?

It helps at this point to consider all the different interactions it is possible for a customer to have with your business.

Mapping each of your data sources into a series of ‘events’ in the customer journey, such as an enquiry, purchase, or customer success contact allows for more powerful analysis of your customer data platform. It helps to expand the scope of questions it is possible to answer to include customer-centric issues like conversion and retention, as well as segmenting top level metrics like customer lifetime value (LTV) by real-life behaviours, such as whether a customer has interacted with your customer success team (more on this later).

3)      Putting it all together.

We can then start the process of joining these data sources together. You will typically need to use an ETL tool (Extract-Transform-Load – such as Fivetran or Daton) to extract the data from your source systems, and a data warehouse to store it (we use Google BigQuery, as we have found it to be fast and cost effective to maintain). Once you have all your data sources successfully imported into your data warehouse, you can start the task of cleaning and structuring your data.

The goal is to end up with a comprehensive dataset of customer interactions you are interested in, allowing you to understand the complete journey of a customer from the point they first interacted with your business, to the present day (and to set up key metrics on top of that). To keep this data up to date, you will need to utilise the scheduling capabilities of your data warehouse – using BigQuery, you can configure the queries which underpin your dataset to run as frequently as every 15 minutes.

4)      Getting the insights.

This where you see the benefits of your hard work! You can connect a powerful visualisation tool such as Tableau to your data platform, and start to answer your strategic questions. It’s one of the most satisfying parts of my role when I can share an insight with a client that they have never seen before. If your data is updating routinely, you can also build KPI dashboard to allow you to monitor leading indicators of success.

Why can creating a single view of customer behaviour be difficult?

Locating the data for event types you are interested in isn’t always straightforward. Untangling the data structures in the backend of your existing software tools to get a view of event type, details and customer attributes can prove a challenging task, as visually demonstrated by the anxiety-inducing web we extracted from a Coppett Hill client’s Salesforce system.

Accommodating for this involves a lot of careful inspection, common sense checking, and often quite complex logic to evaluate for nuances between source data. Something which may at first seem like a dead certainty to appear in a straightforward, easy to understand fashion within a data export, given its significance, may in fact be much harder to find.

There can often be some digging involved to discover what corresponds in the backend to the metrics that you and your team use every day in the web interface. Labelling can also be an issue – when you have 50 different date columns within one table, figuring out which of these to use isn’t straightforward.

Amalgamating data from various sources also relies upon the existence of a ‘common key’. To explain (without getting too deep into the weeds), if you would like to combine data from multiple sources on the same event, you need a column which is present in both data sources to ‘match’ on, for example a unique Customer ID. This may sound simple enough, but when these columns are formatted even slightly differently, you run the risk of losing heaps of valuable data – which of course has trickle-down effects. Metrics drawn from the top layer of a bespoke customer data platform can be wildly thrown off by just one of these rogue ‘matches’.

Difficulty can also come when datasets or assumptions change. Establishing how to handle historical data is vitally important – should a financial input change in the future, how do you enact this going forward, but ensure that the accuracy of your historical profit data isn’t compromised? We have found that defining a set of validation tests, as well as piece-by-piece implementation and a constant feedback loop, have been helpful when navigating these issues.

We are sometimes asked 'can't we just buy a piece of software to do this for us'. In our experience, the sheer variety of datasources required, their changing nature over time, and the ability to perform wide-ranging analysis leads us to recommend building your own customer data platform - using in best-in-class software components for collating, storing and visualising data.


How can I take this to the next level?

Each data source typically comes with some data on each individual customer – but not all of it. When you compile the data locked within several sources into one centralised dataset, you begin to get a much clearer picture of who your customers really are.

The real value of a bespoke customer data platform comes from when you start to segment customers against a measure of their worth (such as LTV – we would recommend focusing on up to three key measures at first). Example segmentations include:

  1. Customer attributes (e.g. where they are based);

  2. First purchase characteristics (e.g. initial value or products selected); and

  3. In-life behaviours (e.g. if they’ve contacted you), including delving into the results of experimental marketing schemes and much more.

Using ‘flags’ when working with our clients’ data can make the segmentation of customers much more straightforward when conducting analysis (an example flag could be whether a customer has received a particular type of communication).

Alongside a reliable top-level estimate of the key metrics we described at the start of this article, a customer data platform gives you the opportunity to view these metrics at a segmented level – allowing you to make decisions which reflect key differences between these segments, as we set out in our guide to increasing LTV. For example, if you were to see that your customer buying Product A had a much higher LTV than those buying Product B, you might want to prioritise marketing channels that attract customers interested in Product B.

One other way of getting the most from your customer data platform is making the output visualisations/dashboards accessible to the whole team. The need for back and forth with an in-house data science team is eliminated. The whole team can see the whole picture the whole time, effectively removing any potential variance between the board level view, and the operational view of the health of a business.

If you’d like to discuss how you can join your customer data sources to understand these relationships for your business, please Contact Us.



Subscribe to get my ideas as they are published

Thanks for subscribing - I normally publish ideas at the start of each week.

bottom of page