3 First Steps For Better Integrated Analytics

The negative effects of having loosely connected analytics teams who do not work closely together are felt in two areas:

-  Consumers of analytics waste time in understanding and therefore acting on information, arguably making worse decisions. 

- Creators of analytics waste time in re-work, re-explaining and general circular grinding of gears having work understood by decision makers. 

Despite this, it is not uncommon to find multiple analytics teams working in parallel. The classic trifecta is; a financial reporting team, a HR/people analytics team, and various types of operational teams. Often with good intentions, these teams work together in a loosely connected way, each overcoming various challenges related to disparate data sources, tools and stakeholder priorities. Unsurprisingly, they can have seemingly little interest in or influence on how the whole picture comes together for those interested - typically senior leadership. In the worst situations, the level of trust in insights from data is reduced because ‘the numbers don’t match’.

The best solution to these problems is a coherent data strategy, coupled with focused attention and action towards the democratisation of enterprise data, but achieving this is no small task.

In ideal transformation cases, there is executive buy-in, a consistent and co-ordinated change management program with well-defined requirements derived from value stream mapping, and appropriate expertise shipped in or preferably co-created with existing teams to create an integrated analytics practice. In reality, this convergence of ingredients is rare or happens on an elongated timescale, with many twists and turns. 

So, what is the inspired analytical leader or team member to do to drive change organically and avoid having their skills and work becoming redundant or even counter-productive?

Here are three first steps which will improve the analytical output of multiple teams for the organisation, while also making each team and individual more well-rounded and successful. These do not require the purchasing of new tools or participation in a formal transformation program. Some leadership sponsorship will help, but by adopting these methods we will lay the foundations for greater buy-in and a faster progression towards analytics maturity. 

1) Create a Data Dictionary

A data dictionary can start as a Google doc or Confluence page, anything where multiple people can contribute easily. At a fundamental level it is as easy and difficult as writing things down. It requires discipline of thought and clarity of communication.

A basic data dictionary should contain two broad categories of information about your data;

- Objective factual information such as what system is used, snapshot time, type of variable (integer, string, logical …)

- Subjective contextual information such as how a particular dataset or value is understood by your team, how certain fields are used to create key metrics

The tools to manage data governance are improving, but all require a good amount of human supervision to set up - do not underestimate the amount of grit and effort required upfront to catalogue and understand the assets you have.

If you can write it down, you have made a good start. If you can get consensus of understanding, or at least reference to a central set of documentation, you are 80% there and ready for technology to help you the rest of the way.

2) Practice a reproducible workflow

If a piece of data analysis cannot be objectively replicated it may be considered opinion, and therefore have reduced credibility depending on the audience. This doesn’t mean it is worthless, there is nothing wrong with qualified and expert opinions, especially if they can be proven to provide insight against reality but this is a topic for another post. 

Also, a reproducible workflow will reduce errors through ease of peer-review and dramatically improve the development and deployment cycle for iterations and improvements.

What is the absolute minimum example of a reproducible workflow? I say it is a README file with instructions inside a folder that also contains or points to the output. 

We can start with one better; a project folder structure which contains three sub folders;

Code - for all scripts

Data - for data and/or instructions on how to connect to sources

Docs - notebooks, reports, presentations

The best examples of reproducible workflows and project structures are found in software engineering, but they also require the most work to set up and maintain. There are lots of good examples to be found which are simple to implement and will make a big difference - here is one of my favourites. 

Remember, the point of doing this is to explain how you arrived at your output - this is valuable for others and the ever-important Future You. 

3) Adhere to a Style Guide

A Style Guide is documentation around how your outputs should look to the end consumer. How are fields named, what colours are used, what fonts, what publishing platforms? Small efforts here can result in huge gains in understanding and removal of circular discussions with decision makers.

I will give you a hot tip - if you have a marketing department most of your work has been done for you. Take logos, fonts, colours, everything directly from your branding guidelines. If there are editorial content guidelines, use those too.

Analytics-specific information to consider in our Style Guide includes display names for metrics, ordering of factor levels, filtering groups for dimensions, use of scientific notation (or not) and basic standards of data visualisation (no 3D pie charts, y axis set to 0 …).

What next?

A simple framework

When each team is operating in this framework, the stage is set to create more effective strategic analytics, empowered by technology and transformation processes. Even if this never occurs, each team will be functioning at a higher level, with faster and better outputs, easier collaboration for deeper insights and happier team members.

In summary: Choose and agree where you will store information about your data, conduct your work in a reproducible way, choose and agree how you will publish your work.