Enhancing a flagship online scientific database to best-in-class

Discovery

Recruiting users

One of the first hurdles we faced was the fact that the Product Owners had no contact with users, and so were unable to provide them for research. I therefore had to look up companies and institutions on LinkedIn, write to Heads of Departments and key individuals to request participation.

While this was a time-consuming process, I was able to get a good understanding of the types of user involved, and put together a group of users with whom I could conduct research and testing, as well as add and remove users depending upon availability. This ensured that I woudn’t have to repeat the recruitment process again, and also had a set of varying opinions and perspectives.

Planning the research

As I am not a materials Scientist, I worked with my Subject Matter Expert colleagues to understand the current state of the platform, and how it was used. This helped me to refine my research into four key stages to examine:

Involving the team

As well as improving my own knowledge, it was important to help my product team understand the context behind decisions, and so in order to include them in research sessions, I would have them listen in to the call, and provide questions suggestions via Slack. This helped them to take part and foster empathy with the users, without interrupting the flow of the interview.

Major discoveries from the research

1. What brings the user to the site?
   • Users were often provided access to Springer Materials by their company or institution, and used it to look up data for their own research and experiments because of Springer’s reputation within scientific publishing. 
   • However, they had found their experience in locating the data they needed to be frustrating and unclear, which they reported back to the people who bought the subscriptions, leading to a decline in renewals.
2. How do they search for the information?
   • The most common search terms were for a material and a property, such as ”steel melting point” or “mercury surface tension”.
   • However, users expressed frustration with the results they received, with searches not giving desired results. when they found what they wanted, they would have to click on multiple pages to actually locate the information they needed.
3. How do they work with the information they need?
   • As the site had been based upon scientific journals, when the users found the relevant page, they would then have to download a PDF with image scanned pages, which they would then search through by hand to obtain the data they need.
4. How do they take the information away?
   • As the data was image scanned, there was no way to digitally export it, and so users would have to print or write it out by hand, or type it into another application themselves.

Analysis: co-location

Our product team and stakeholders gathered at our Pune office for a week together to analyse the findings:

• Presenting my research findings, I defined the key discoveries and provided personas to identify the needs of different user types.
• I then organised sessions building empathy maps and user journeys to further explore the problems and identify what we knew against assumptions to be explored later.
• We split into small groups, coming up with ideas that leveraged different expertise and viewpoints, which were presented back to the wider group for criticism and discussion.
• This then helped us come up a with a series of concepts which we could build into prototypes to test with users and challenge assumptions.
• Working with the Product owners and Business Analysts, we could then prioritise solutions on an Impact vs Effort matrix, which helped us to build a plan for the work, ensuring that we could demonstrate value quickly while also keeping larger pieces of work for later impact.

Solutions

Improving user journeys

The quickest way of demonstrating value to users was to make the process of finding a piece of information more easy, by reducing the number of steps involved,

• We started by identifying “red routes”; key journeys that users took through the site to find information, such as finding a specific material such as “benzene”, and looking up a property such as “boiling point”
• We identified in our research that the user journey had bloated due to “experience rot” (continual addition of extra pages and features outside the original remit that confuse the original objectives and impair user success).
• Taking those key routes, we quickly mocked up clickable prototypes for users to try, ensuring that our assumptions met with their mental models, and gaining feedback before revising the actual database.
• During this testing, we also identified opportunities such as digitising the data within the PDFs, which I could then advocate to the business as opportunities for future product development

Making pages responsive

During our research, we identified not only the different requirements for each persona, but also discussed with them the different scenarios in which they would access the database. For example:

• People accessing the site on a mobile device would often want to check a single, specific piece of data, and appreciated the ability to access the information quickly, without having functionality that they did not require at the time get in the way.
  • This also worked for people who would view the database on a smaller window on a desktop, ensuring that their view focussed upon the information that was key to their search
• People who accessed the site while sat down, with a larger device such as a laptop or desktop, would want to spend more time investigating, and would want more supporting information

By returning to our “red routes”, we could then examine them with an extra dimension, ensuring that the layout and display for different device sizes responded to those requirements

Improving search

A larger piece of work, we recognised that improving the results given in search would be more effort, but highly important in ensuring that users could find the information they required.

• We asked users to take us through their search processes, and highlight where they saw results that they felt were not suitable to their queries.
• We then reviewed with the developers, and discovered that it was the markLogic text-based search that was the problem, which only searched for text strings, and didn’t understand context.
• We identified the key problem was that the simple case-agnostic text-based search was not suited to the nuance of certain search terms, for example:
  • “tin” refers to the metal; tin
  • “TiN” refers to the material Titanium Nitrate
  • As the search did not understand the context, typing either of those terms into the search would yield the same results.
• We therefore explored a number of solutions, including:
  • case-sensitive searching, to understand specific chemical terms.
  • contextual drop-downs, to allow the user to specify which terms they were searching for
  • bringing in search database experts to implement a graph search, which created a contextual model between inputs (such as recognising tin as a metal, which would therefore have a link to steel, which is also a metal, but not tincture, which is a chemical process, but has the letters “tin” in it)
• We built a solution that could analyse the different cases and chemical terms, as well as building a new graph search database, which would understand the relationships between different materials, such as metals, providing even richer search results.
• This was then placed on a separate testing link, allowing us to test with users remotely, as well as in person, including taking it to various Chemistry Conferences in the USA, where we could conduct guerrilla testing with attendees, for a wider set of results.

“This is great. It really will save me a lot of time searching in the future.”

Corporate Researcher during testing at Materials Research Society Boston Conference 2016

Developing the homepage

As well as catering to the requirements of the scientists and researchers who used the platform, by examining the needs of the Buyer, we recognised that the homepage had three important roles:

1. To provide a starting point for users to explore the content
2. To update returning users with new developments
3. To demonstrate value for non-technical users such as Buyers

We therefore redesigned the homepage to include the following features:

• A summary of the different types of content for new users to explore,
• A timeline of latest additions and improvements to update returning users
• Details of the depth of information and sources to demonstrate value for Buyers

These changes were added over time, to accommodate the work around supporting these features, and other efforts on the database.

Making data digital

One of the most fundamental solutions that we identified with the product, with the highest level of impact but also the greatest level of effort, was the fact that the data needed to be digitised. At the start of the project, data existed solely within the database as scanned-in pages from scientific books and journals, which frustrated users, and led to users feign that the product was not worth the subscription price.

it was this impact on subscription revenues that allowed me to petition Product Owners to organise a way of tackling this problem. Using the company’s relationship with client institutions, they were able to find a group of post-doctorate scientists to work through the scanned pages, performing specialist data entry to extract and annotate data for the graph search database, converting the scanned data into a fully digitised format. This work took about a year to complete, but this had a profound effect on the how we could make the product more valuable to users.

Surfacing results early

The digitised data meant that we could include it within search results, meaning that simple questions could be answered sooner, and demonstrating value more quickly. These could be displayed as search snippets, filled with simple answers, that led on to pages with more complex insights.

Working with data

We recognised that the majority users searched with the same pattern, namely a material (such as iron, benzene or carbon) and a property (such as boiling or melting point, band gap, or similar). By understanding this, we could use our newly digitised data to improve our offering:

• When a user first searches for a material and property, then we can surface the simplest answer in the search snippet, as described above.
• If the user wants more in-depth information, then they can click through to a page providing a dynamic graph, which uses the findings from scientific papers to plot data in a visual away, demonstrating the behaviour of the material and the property against a scale – for example, demonstrating how the boiling point of steel changes when submitted to different atmospheric pressures.
• Alongside this graph, the results are also provided in a tabular format, which allows users to work with the data displayed in the graph. They can change criteria, expand or limit the dataset, all to ensure that they get the information they require for their purposes.
• As the last part of the model we discovered, this data can then be exported into a range of format, such as visual images of the graph, or spreadsheets of the results, facilitating use in the user’s work.

“This is awesome. I don’t think anyone else is doing this. Where do I sign up to get it?”

Student Researcher during testing at American Chemistry Conference, Philadelphia 2016

Outcomes

• Our work reversed the customer attrition that 
the product was experiencing, and brought a 32% subscription increase from academic, corporate 
and other clients
• Within two years, that extra revenue paid for all of our salaries and the money that the company had paid into the project.
• Developing the digital data pages won accolades 
for the product as a “best in class” tool from the 
American Chemistry Society conference in 2016
• In 2017, the Indian Government bought licences to provide the product in all libraries across their country
• A site survey, run at the end of my tenure as UX design lead, showed that 69.2% classified the product as “great”, a 22% improvement on when I started.