Viewing entries tagged
data visualisation


Context brokering - understand the hype.

If you haven’t heard the term ‘context brokering’ - you will do soon.  The 2016 Gartner ‘Hype cycle’ highlighted it as a smart machine technology that will begin to mature over the next 10 years.  

Gartner Hype Cycle 2016 - from

Gartner Hype Cycle 2016 - from

To understand this development a little more it’s probably worth defining what we think ‘context brokering’ could mean and its different levels of application, as at present it seems to covers a range of different problems and timescales, which can cause confusion.  So, let’s look at the first part.


At the time of writing (19/07/2017) - if you google ‘context’ you get the following:

In life (and business generally) we regularly make plans to deliver a particular outcome or to get to a particular point.  To form such plans, we generally need some kind of understanding of what’s going on.  This is where the context part comes it.

Having a picture of the ‘context’ you’re currently facing (to use the google definition - the circumstances that form the setting for an event, statement or idea), or likely to face in the future allows you to both understand what decisions you might need to make and any assumptions that could be implicit in them.  To understand this a little more, have a look at our post from January 2017

Once you understand what context means, you start to appreciate the issues of ‘timeliness’ (when do you want to do something) and 'relevance' (what sort of data is there to help you understand a context, and how regularly is this data produced) - and these two qualities relate to the ‘brokerage’ side of things.  


A ‘broker’ is someone or even a section of computer code that offers a particular service and this is generally to find or provide some kind of data to help you make a better decision.  In the application of context brokering, the broker is (in theory) a type of service that finds and returns data relevant to the particular context required to form a plan or make a decision.  

However, the decisions that people need to make vary greatly, both in terms of complexity and timeliness and this is what’s crucial to understand when considering how context brokering could work.  What’s crucial about ‘context’ in the context of context brokering (I won't do that again, I promise) is that it currently has different meanings - and these generally relate to the timeliness of the data being used.  By exploring what 'context brokering' could mean, and how it could be applied we've deduced that there are two key applications for how context can be useful in the real world:

  1. Immediate prediction of consumer behaviour
  2. Understanding of strategic insights

To understand these two different applications, we've unpacked them a little here.

1. Immediate prediction of consumer behaviour

The immediate predictive benefit of context brokering is probably in the form of brand and consumer insight generation.  With advances in big data, many organisations generate and/or have access to a large amount of data across a wide range of sources.  By managing and aggregating all these different data sources organisations can start to generate particular contexts, perhaps for how a brand is performing or how consumer spending happens.  For example, an online sales platform could use different cookie data to track what associated web-pages a shopper has viewed before and around a purchase.  This could yield interesting predictive data, for example, do people look for certain products during certain climatic conditions for example, in a heat wave - do sales of hats and fans go up? Or would a person purchasing sun cream, mosquito repellent and beach towels, be interested in new sunglasses?  As such, does such an understanding of unique contexts derive a commercial advantage?

To fully get value from immediate context brokering, there are a number of research questions to consider.  For example, how does an organisation bring together and model data?  For a large organisation with an established infrastructure, this isn’t likely to be an issue - the sources and data collection mechanisms already exist - it’s mostly a question of making sense of the data and being able to translate it into some form of action or insight.  It is perhaps a bigger challenge for an organisation trying to understand the utility and applicability of this type of service to their business, for example, if you don’t have access (or the need) for big data and you’re not generating masses of data, what value will it have to your business?

Additionally, if context brokering is likely to add value to your business, how wide ranging are you data sources?  How accurate a context can you produce just using data from facebook? Can you get a more accurate context from looking across as wide a variety of social media channels as possible?  If so, what are the cost implications for drawing data from such a wide variety of sources?

Another question to consider is what technologies exist to make context brokering a reality? This is currently a hot area.  Predictive techniques are improving as people use more sophisticated statistical models.  Machine learning can be applied to train algorithms to detect patterns and find specific terms in larger and larger datasets, ditto for machine vision algorithms and visual data.  At the same time, databasing and data storage continues to rapidly increase as does processing power.  All these technology trends make the immediate predictive benefits of context brokering increasingly enticing.  However, it’s still worth reflecting on the fact that however pure the model and the maths, at some point, context has to equate to action for it be of value and this is something that can often be forgotten.  Essentially, the final challenge is making sure the right kind of predictions are linked to the right kind of behaviours!

2. Context brokering for strategic insights

Another application for context brokering relates to less immediate decision making and relates more to research and development.  Timeliness is not such an issue in this application - its the scale and breadth of data covered to inform a decision that’s important here.  At present, it’s the kind of activity that kicks off many large projects, especially in research, policy and academia. The common element to immediate prediction is that, such projects are undertaken to determine what we believe the state ‘truth’ is around a particular issue/idea or event.  

This form of 'strategic analysis' is less time pressured than immediate prediction, but the sources and ranges of the data used to inform our actions, plans and decisions are still important.  In the past, organisations have generally done some kind of early activity like a literature review, or assigned an intern or student to summarise the research around a particular issue.  From this, an assessment is produced on why we want to do something or follow a particular course of action. Context is really important here, often we base our first principles for a course of action on our belief around a certain event, paradoxically, this often the point where we do the least amount of research and it can be subject to a high degree of bias and poor research.  For example, if too small a starting dataset is used our assumptions and lack of research can be quickly exposed as the work is shared more widely.  This is where context brokering can offer a decent alternative to such traditional techniques (which we see today in the form of literature searching and workshops).

Using context brokering for strategic insights improves how we gather, store and map knowledge; enabling us to have greater confidence in our initial assumptions or understanding of complex problems.  As a technique it can also allow us to produce ‘ontologies’ for particular problems that can allow more specialist data gathering and improved research gathering and network understanding and this can allow us to learn and gather more data more efficiently.

However, as with immediate prediction, there are challenges.  Selection bias can have a large impact for such a technique, especially if a small data set is used and if particular terms favoured knowingly or unknowingly, the process can simply yield more things to confirm a particular view of the world.  Additionally, the issue of perfection versus relevance still applies greatly - however good our model is, it still needs to be communicated with decision makers who need to be able to understand and interact quickly with the main findings of the model, while at the same time trust that the model relates to ‘real’, trusted data.

Additionally, strategic insight generation is probably based on a more limited format of data. Where there are considerable conversion issues for immediate prediction, strategic context brokering tends to rely on text-based analytics (this could reflect the longer lead time in data used for research and development planning?)  This means it is, in some ways, a simpler area of study, one that can benefit greatly by further research of the applications of machine learning for speeding up how the data can be processed and used.  However, its still worth coming back to the potential bias a human can apply in such analysis - but does the intelligence and insight that such human input provides outweigh the downsides? This is a key issue for further research - one that data scientists and analysts continually grapple with. How do you configure the optimum balance of machine-based learning to improve the efficiency and scale of human analysis?  What role does the human analyst have in the analysis process, when at least for the next 20 years, they are likely to remain the best predictor of context and its translation into specific insights, actions and implications?

With all these points on board, and to offer some kind of conclusion to this post, it’s probably worth defining and thinking about what 'context brokering' could mean in the future as we start to understand its applications a little more.

An updated definition of context brokering

Context brokering is a service that enables actions and insights to be generated from broad sources of data and information.  It can be applied with different levels of timeliness - from the immediate to the strategic.  Immediate context brokering as a service applies advanced forms of computer science to provide actions and insights either to another system or a human.  Strategic context brokering, applies the same principles to wide ranging problems that have considerable published literature (often from a scientific or research basis) to map and better inform decisions and insights to be formed around the dataset.  

Additionally, another thing to reflect on is how context brokering works as a process - which whatever the timeliness of the data, tends to rely on the following process.  

  1. Definition of problem and sources

  2. Data gathering

  3. Data storage

  4. Mapping

  5. Action/insight generation

  6. Feedback to 1 (as required)

Final thoughts

'Context brokering' is a newly emerging area and it’s exciting to be in it.  Our own insights have come from the smaller scale applications of strategic context brokering, but what’s interesting is how applicable many of the techniques are to different sources and timescales. However, it may still be worth reflecting that certain principles for analysis still hold true, and are perhaps more important than ever when applied to the era of 'big data'.  As well as issue of timeliness and relevance, trust is still key.  How much do you value and rely on your sources? It is your sources that will ultimately still drive and sustain the validity and quality of whatever context you produce.

What do you think?  If you have any thoughts you’d like to share on context brokering, please either add them here or drop us a line at!