THE ECONOMIST : PROPENSITY TO BUY
NOVEMBER 2018
A data product proof of concept looking to tell consumer industry professionals how likely consumers are to buy their products at a city level in hard to reach emerging markets.
01 / Summary
What was delivered and how?
Team
Product team
Product manager
Lead product designer
Lead modelling economist
Cities data modelling economist
Lead engineer
Backend engineer
Front-end engineer
Business analyst
Project manager
QA tester
Steering committee
Head of Marketing
Head of Editorial
Head of Digital
What I did
Customer profiling
Wireframing
Prototyping
Data visualisation
UI design
Handoff
QA
Deliverables
Personas
User journey mapping and flows
Data visualisation
Clickable prototype
Detailed designs
Impact
Over a 6-month period the project team researched the market, conceived of the idea, sourced and modelled the data and designed and tested a proof of concept. The work provoked a step change in the type and way that data is collected for the EIU’s business that has inspired and led the way for a series of innovative new data products.
02 / Finding the value proposition
In April 2018 the Economist Intelligence Unit identified an opportunity to use its data and expertise to create a new product targeting consumer industries giving them the ability to test consumers’ level of interest in their product at a city level in emerging markets.
Initial upfront research had shown the food and retail sector to be of interest and the project PM and myself conducted a series of interviews with companies including Godiva, Kellogg, Mars Wrigley, McDonalds China, PepsiCo, Red Bull and Unilever to further understand their unmet needs and position the EIU’s offering.
We had learned that companies often relied on an imprecise method of combining multiple information sources to reach a vague understanding of the consumer’s propensity to buy their products. In their words “a bad guess is better than no guess”.
Our value proposition, in contrast, would use proprietary survey data on consumer preferences coupled with existing income and demographic data to model demand for a product at the city level in emerging markets. This would give strategists and data analysts a much more accurate, reliable and robust way to identify consumers’ willingness to buy in a single platform.
03/ Customer profiling
Following our value proposition work and using the research interviews as a source, I set about creating a customer profile.
I began with a customer profiling workshop where I worked with a cross-functional team from across the Economist. We referenced our existing research interviews and the individual expertise within the room to map out both the current journey and an aspirational one for a consumer business looking to size markets in new locations.
We then drilled down into the specific roles played within the strategy team of a consumer company and identified two key roles. These were:
Steerers: influence decision-making - especially in meetings; they can inform opinions rather than be involved in number crunching
Informers: responsible for finding the patterns in data that help determine consumers’ buying behaviour; care about the accuracy of data and modelling.
We felt that each of these roles would highlight a different way of using the data, one being a user who is looking for patterns in data and feeds this insight up the chain and then a secondary user who is not a number cruncher, but requires headline information to inform decision-making.
Personas
04 / Starting with data
The solution design phase for me started by working with the data team to create a vision of the service from which we could work backwards and determine the data we needed to collect, the methodology for modelling the data and the means of collecting that data.
From our initial discussions we were expecting to be able to get to 2 key numbers:
the propensity to buy score as a percentage e.g. 34% of consumers will buy chocolate in city X.
and
the market size e.g. (34% propensity to buy is equal to) 2.4 million consumers in city X.
From our research I knew that it was the second figure that was of more interest to consumer companies, “the size of the prize” was how one executive had described this in the research interviews. We also knew that being able to forecast these numbers was a must-have for businesses - the further out the forecast, the better. We decided on a 5-year outlook.
Beyond these initial big numbers, we knew that businesses would also like to be able to build a profile of their potential customers through demographic data i.e. age, gender and income level and also their purchase and consumption behaviours. As a team we converged on 3 data points showing frequency, purpose and channel of purchase to be included.
The final data piece was to represent consumers’ attitudes to health and fitness, technology and lifestyle. We had learned in the research phase that companies will also look for other indicators outside of purchase and consumption behaviours to create the customer profile. This may be things like home ownership, using credit cards and mobile payment apps and exercise routines. We aimed to capture some data relevant to these other indicators and include in our model.
We identified 4 markets to collect data on that we thought would give us a good data sample to work with. These were Bangkok, Ho Chi Minh City, Johor Bahru and Jakarta.
It was also decided to compare two different ways of collecting the data. One method was using a more traditional consumer survey company and the other was through a technology startup company called Streetbees that uses artificial intelligence and geolocation data to run surveys and provide insight by connecting to respondents on the ground through a mobile phone app. The two approaches would give us different outputs with which we would be able to compare and evaluate later before deciding which was more effective and commercially viable.
05 / Design sprint
As the data team began the collection and modelling of the survey data, I started on the page designs.
I ran a design sprint over a week with the product team where we were able to explore the problem space from all angles, brainstorm ideas, converge on the strongest ideas and then prototype and test internally.
We ended up with 4 main views in the tool. These were:
a data summary page giving headline numbers for each city
an individual city data deep dive giving a full breakdown of the data
a data explorer, to allow for an analyst to dig into the data
supporting country analysis to give deeper background and supporting analysis
From this point I was then able to start to think about the tool from a holistic level and map out the service and its key flows.
06 Data visualisation design
Once I had a vision for the overall architecture I began to think more closely about the data design for each view, trying to ensure that headline information could be quickly extracted in places, but also allow for data deep dives where needed.
A late request from the EIU’s CEO was to try and incorporate crowd sourced photography into the product, more as a proof of concept for potential ways to source data in the future, than as a core feature or component for the service.
07/ Live data
Once the surveys and modelling were completed I started to see the shape of the data and I was able to test out the initial designs with the live data.
The data team supplied me with an Excel document containing the data and I applied this to the charts and visualisations in my wireframes.
As expected I found gaps in the data, assumptions I had made that didn’t play out and data that didn’t produce the expected results. I worked closely with the data team to work around some of the gaps and remodel the data where possible. There was some refactoring and rethinking of my original ideas to accommodate the data. For example I had expected to be able to forecast the propensity to buy score, but learned that it was only possible to forecast the market size numbers.
We also had data coming from 2 sources, which gave us slightly different outputs. The Streetbees data included additional cluster analysis and I worked with the Streetbees analysts and one of our own data scientists to understand the data and how best to represent this visually.
After sharing with stakeholders I received further feedback that they wanted to see more overlapping data, so rather than just showing the purpose of consumption and the frequency separately we should try to show the overlap of both. They also wanted to have more ability to compare cities, so I added an additional comparison view. There was also a request to show the same data for non-chocolate confectionary, which we had also collected.
At this point I was able to create higher fidelity wires with real data that we would be able to test both internally and externally. I created a complete click through prototype for one city.
08 / User Testing
With our full data prototype we then began a round of evaluative testing with two different chocolate and confectionary companies (one mass market, one luxury).
I worked with the PM to create a research plan and protocol for the sessions. We considered what we wanted to learn from the session and how we were going to conduct it. We co-created a guideline for the session and I created a click through prototype with one complete city allowing the user to explore the page.
Summary of research:
Our participants sense checked the data and described the overall information design as clear with a logical flow and prioritisation. They found the deeper level city data compelling and not something that our competitors were doing. Participants were able to describe how the service would be used for their strategic planning, giving examples such as “the occasion module would help us to design a product portfolio for a city, and the holiday breakdown would help us to build business on the yearly calendar”.
One participant particularly liked the Explorer module and the ability to play with the data and we were able to get a good idea of the kind of data that he expected to be included and the kind of tasks that he would use this for.
Another participant was also able to describe how the Location Intelligence module would be of value to her role particularly if we could collect real time prices for products in shops. This is an activity that her company currently does, but it is outsourced. She liked the idea of be able to bring supply side data into the service.
It was the forecasting side of the service that our participants thought we could push more. If we could forecast out further and across more indicators, then we would have a strong value proposition. One participant mentioned that “forecasting is something we all struggle with”. He suggested being able to project the purchasing power of income groups and increasing our market size forecast to a 10-year horizon would be something that our competitors are currently not able to do and would provide great value to consumer businesses. We took these ideas back to the data team for them to consider.
Full data city view
09/ Outcomes
Shortly after the final research phase we had an internal go/no-go meeting with stakeholders to decide if we would move into the next phase and build out a full data proof of concept.
The work to date was presented and stakeholders were impressed with what they saw. In particular they liked the focus and granularity of the data, which to a business historically centred on macro-economic country level data was a significant and potentially interesting pivot for the business.
The downside however was the cost, feasibility and expertise needed to get to this more granular data, especially when thinking about scaling this to include other consumer sectors and other products. It was decided to park the project at the time and start to think more strategically about building in-house capability or find partners to run city level surveys and conduct cluster analysis.
It’s disappointing when a project doesn’t mature, but the project team felt when doing a final retrospective that we had learned a lot that would help us with future projects, but more importantly we had provided a valuable steer for senior stakeholders in thinking about the future of data and data collection within the business.
We had shown our capability to get to very granular behavioural and attitudinal data at a city level in parts of the world where few of our competitors are doing so.
This project proved to be a catalyst for the EIU to rethink its approach to data with a move away from statistics bureaus and quarterly country level data to real time alternate data sources such as satellite imagery and location intelligence from mobile devices.
We had gone from initial idea through to final modelled data in less than 2 months. The speed of learning was something that we thought had gone well, especially given that the project team was spread across 4 time zones.
Speaking to prospective users is quite often a challenge. We were targeting quite senior management and executive levels in large organisations. In the early stages of the project we had managed to talk to about 15 organisations, but we felt that we needed to find a better way to keep these people engaged over the project. One suggestion for this was to recruit an editorial board made up of prospects who we would embed more deeply in the project work and feedback to more regularly.
For me one of the big learnings was around how to design for data when the data doesn’t yet exist. This required me to work closely with the data team to create that initial vision that allowed us to begin to think about the shape of the data and what we could get out of it. Wherever we could we used dummy data and modelling to get a better idea of the likely output. I learned that it is important to not become too attached to a single idea and to always think about adjacent possibilities when the data doesn’t match your expected results.