Re-architecting from the bottom-up

Posted by

In December 2001 PeopleSoft, a large enterprise software company, relaunched its public website, and customer and partner extranets, Customer Connection and Alliance Connection. It took 11 months and more than 60 people to redesign and build the information architecture and graphic identity, build the technical infrastructure, migrate and rewrite existing content for the new content management system, test it, and finally publish the new site live.

All information architectures have a top-down and a bottom-up component. Top-down IA incorporates the business needs and user needs into the design, determining a strategy that supports both. Bottom-up methods look for the relationships between the different pieces of content, and uses metadata to describe the attributes found.

We undertook the re-architecture of the PeopleSoft web properties for a number of reasons. First, the three sites all had their own user experience, different architectures, and varying core goals. The sites also had overlapping content and users. Partners, who had to navigate all three sites to get all the information they needed, had the worst experience because they had three sites to navigate and understand.

Content was often duplicated across the three sites. This made updating the site time-consuming and difficult because files had to be updated in many places. It wasn’t uncommon to find different versions of a document on each of the sites, or even within the same site. Each site had its own style guide, which added to the varying experiences.

The sites also differed in their technical back-ends. Each site had its own search engine and content management system. Many types of databases were employed on the sites, and the structure of the data varied from database to database. Different information systems teams, as well as content development teams, supported the sites.

In February 2001, we started a project seeking to create a single site, with a unified technical infrastructure and three distinct user experiences. This new system would use Interwoven’s content management system, TeamSite, to store and generate the files for all three sites. The sites could share the same content assets where possible, reducing creation and maintenance overhead. Users would have the same type of experience on all of the sites, due to the shared graphic identity, branding, style guide, and information architecture. Once users learned one site, they would be able to transfer that learning to the others.

While we used many methods and tasks as part of this enormous project, this case study will focus on just one small piece of the bigger picture: the bottom-up information architecture methodologies. We did extensive user and stakeholder research, usability testing, and top-down IA, but a thorough discussion of them is beyond the scope of this article. The architecture portion was the first part of the project to be completed. PeopleSoft hired Lot21 and Adaptive Path to help with the architecture development.

Information architecture has a bottom?

All information architectures have a top-down and a bottom-up component. Top-down IA focuses on the big picture, the 10,000-foot view. It incorporates the business needs and user needs into the design, determining a strategy that supports both. Areas of content are tied together for improved searching and browsing. It determines the hierarchy of the site, as well as the primary paths to main content areas. Top-down IA can be as large as a portal or as small as a section home page.

In contrast, bottom-up IA focuses at the lower levels of granularity. It deals with the individual documents and files that make up the site, or in the case of a portal, the individual sub-sites. Bottom-up methods look for the relationships between the different pieces of content, and uses metadata to describe the attributes found. They allow multiple paths to the content to be built.

Both top-down and bottom-up methods are necessary to build a successful site, and they are not mutually exclusive. They work together to take the users from the home page to the individual piece of information they need.

Content inventory

Before we could do any designing, we had to first understand what we were dealing with. The first step we took was conducting a content inventory, which counted and documented every page on the site. It recorded specific information about each page that would later be used during the content analysis.

We created a separate Microsoft Excel spreadsheet for each site’s inventory. Each main section or global navigation point got its own workbook or “tab” in the spreadsheet. This made it much easier to work with the large files. The name of the page, URL, subject type, document type, topic, target user, and any notes about the page were manually recorded. There was room allotted in the spreadsheet for PeopleSoft to record the content owner, frequency of updates, and whether the page was a candidate for ROT removal. (ROT stands for Redundant, Outdated, and Trivial content.)

The final inventory consisted of more than 6,000 lines in the spreadsheets. Only HTML pages were recorded. Pages in Lotus Notes databases were excluded, though the different views were documented. Of the information recorded, link name, URL, and topic were the most useful and we referred to them again and again throughout the project. The other fields were still useful though. By filling those fields out, we were able to think more critically about each page, and get a better feel for and internalize what the sites had to offer. If we had just captured the page name and URL, or used an automatic method for gathering the information, this depth of knowledge would have been lost.

In addition, each page was assigned a unique link ID. At the beginning of the inventory, we envisioned using the link IDs as a way to refer to the pages since the page titles were often inconsistent and unreliable. In reality, the link IDs were too complex and numerous to use. No one could remember that meant the volunteer request form. The link IDs did prove to be helpful in other ways. By simply scanning a page in the spreadsheet, it was quick and easy to determine how broad or deep a section of the site was. They were also helpful during content migration in mapping the content on the old site to the architecture of the new site.

Unified content map

The content inventory spreadsheets were highly useful for detailed information about individual pages. But more than 6,000 lines of information are a bit hard for people to get their arms and brains around. The spreadsheets were not very good at giving a high-level view of the content on the site. For that we created the unified content map. Once the inventory spreadsheets were completed, we were able to pull out the different document types and content types we had found. We identified the larger content areas (e.g., general product information, customer case studies) and then listed out the individual examples that existed on the site (e.g., component descriptions, functionality lists).

The content areas of all three sites were mapped together in the same document, forming the unified map. We then identified content types that were duplicated between the sites. These overlapping items indicated areas that we wanted to investigate further to understand why they were duplicated. Was the document modified slightly to better serve a particular audience? We found out that in most cases, the documents were identical. Usually the content owner simply didn’t know that the document already existed elsewhere, or the technology used made it difficult to share assets. These overlaps were a driving force for structuring the content management system so a single asset could be used in multiple ways, for multiple audiences.

Classification scheme analysis

Beyond understanding the types of content that were on the PeopleSoft web properties, we also had to understand the organizational schemes that were in place on the sites. By looking at how the content was currently structured, we would have more insight on how it could be improved.

Classification scheme analysis was done on the products and industry classifications of all three sites. The names of the industries and products appeared in different places throughout the site, beyond the products section. For example, in the “Events” and “Customer Case Studies” sections documents are classified by product and industry. Each instance of the classification was recorded in a table, so the terms could be compared.

The first thing we looked for in the table was inconsistencies in wording from list to list. Inconsistencies illustrate the need for controlled vocabularies on the site because there are so many ways to describe the same thing. These inconsistencies were used as the basis for variant terms in the product and industry vocabularies. We also looked for “holes” in the classifications – places where terms were not used. Holes could indicate places where content needed to be developed, or needed to be removed because it was out of date. These sections were flagged so they could be examined during content migration.

Content analysis

Once the content inventory was complete and we had created the unified content map and classification scheme analysis tables, we had the daunting task of analyzing what we had documented. We used these tables and maps to help us find the patterns and relationships among the different types of content.

We looked for ways the content could be better tied together. On the previous site, content lived in discreet silos and there was very little interlinking. We discovered that there was actually a lot of information that could help prospective customers better understand our products and services or processes, such as implementing a PeopleSoft solution. For example, there are consulting services offered by PeopleSoft, as well as our Alliance Partners, which are specifically focused on the task of implementation. Training classes are available for both the technical implementation team, and the end users who will be using the new software. Once we saw these connections, it became clear that we needed a new section of the site devoted to implementation. User testing confirmed this, and we also learned of other types of information users needed, like a listing of supported platforms PeopleSoft software runs on.

Through content analysis we were also able to create the metadata schema to use on the new site. Some attributes such as products or services were obvious from the beginning. Others, like language and country, became obvious only when we saw how many documents we had that were non-English or appropriate for only North America. Twelve attributes in total were identified, and they are used to describe content on all three sites.

Creating the product lens

Information about the different products was spread out across the sites. This was especially true on the Customer Connection and Alliance Connection sites, where there are support documents in addition to sales and marketing information. Users had to go to multiple sections of the site to find all the information they needed. High-level marketing material could be found in the “Products” section, but support information was in its own area. Documentation was separate from support, and upgrade information was separate from both support and documentation. This model supported users who came to the site knowing what they wanted – support information for Global Payroll. The model didn’t work for users coming to site wanting to see all information related to Global Payroll. There was no central place that aggregated the links to the various resources together.

A goal for the new site was to support both types of users. We began by combing through the content inventory and the sites themselves to find all information related to products, no matter where they lived in the sites. Examples of content we found are support information, consulting services, training classes, and industry reports. We wrote each item down on a sticky note.

Working together with the Customer Connection team, we organized these sticky notes into different groupings. The sticky notes worked very well in this exercise. The “unfinished” nature of the notes encouraged people to be more critical and they felt freer to make changes. The whole team participated by moving the sticky notes around and discussing the reasons behind the movement and connections among notes. While coming up with the groupings, we didn’t think about final nomenclature. We instead focused on capturing a name that described the essence of the group. We ended up with titles like “What Others Are Saying About Product” and “Working Beyond the Product.” Things you would never want to see in a global navigation bar. We refined these labels later on once we built out the product pages.

These groupings formed the basic structure of the product module pages. Because there was so much information related to the products, we decided to divide the module pages into different tabs. The public would see three tabs— “Features,” “Technical Information,” and “Next Steps.” Customers and partners would see two additional tabs—“Support” and “Upgrade”—once they had logged into the site.

The information available on these tabs is supposed to be specific to the individual product. Ideally, a link to release notes on the “Global Payroll Support” tab would take the user to just the Global Payroll release notes. Unfortunately, due to technical limitations with our current database structure, we have to link to the release notes area in general. Users must then drill down to the information for Global Payroll. As we update the databases, we will be making these links more specific. Until then, we feel it is an improvement from before, when the user would have to backtrack out of the products area and drill into the documentation area to find these notes. We are at least getting them to the right neighborhood.

Site comparison tables

Not all of the bottom-up work occurred at the beginning of the redesign project. Once the new architecture was determined, we still had to populate that structure with the content. To aid in the migration and creation of content for the new site, we turned again to the content inventory.

The content inventory was performed in May 2001. Planning for the site migration didn’t take place until September. Even though specific pages on the sites had changed since the inventory, the bulk of the inventory and the structure it represented were still correct. We modified the inventory spreadsheets to include the new site structure, complete with new link IDs.

These tables began as a means to double-check that all the content had been accounted for in the new architecture. It also allowed us to see holes where we would have to create new content. As plans for migration continued, the use of the tables expanded. They provided a means for estimating the number of pages that had to be migrated. A column was added to indicate if the page was part of a database not scheduled for migration. Columns for the content approver and the migration team member names were also added to the spreadsheet. This made it clear to everyone who was responsible for which sections. This also helped in balancing out the workload among the whole team.

Once migration started, the usefulness of the comparison tables quickly faded. On-the-fly changes to the architecture occurred at the lower levels of the site as we worked with the migration team to slot the individual pieces of content. The tables quickly became out of date, and it took too much time to keep them updated.

State of things today

The new, Customer Connection, and Alliance Connection sites launched on December 21, 2001, on time and on budget. Since the launch, site inquiries, one of our major success indicators, are up significantly over last year.

But just because the site is live and successful doesn’t mean our work is done. We are continuing to refine and tweak the site. We are conducting various user testing and usability sessions to see how customers and prospects like the new site, and where they are having difficulty. We are retiring older databases and migrating the content into Interwoven TeamSite. There are areas of the site that we simply didn’t have the time to examine in detail during the redesign. We are now tweaking the architecture of these sub-sections such as “Training” and “Assess Your Needs” to better support the content we have and make it easier for users to find what they need.

Later this year we will be implementing PeopleSoft’s portal software so customers will be able to better log and manage their support cases and have more control over their site experience. The work is really just beginning.

Chiara Fox is the Senior Information Architect in PeopleSoft’s web department. Before joining PeopleSoft, Chiara was an Information Architect at the pioneering consultancy Argus Associates.


  1. After reading “Re-architecting from the bottom-up” I decided to go take a look at the site. I was a little surprised when I tried to validate a page according to its “DOCTYPE declaration” and found errors. Why bother declaring such things if you have no intention of adhering to it?

    It seem that “site designers” tend to look at this as being unimportant! I wonder why?

    BTW, I used the W3C validator at:

  2. Thank you for sharing this experience, in special, the bottom-up IA. I would like to know more about IA experiences in designing corporate portals.

  3. Great article, the address of the page changed since yesterday, I think you’re probably doing some backend changes. Just thought I’d let you know that the images aren’t visible on this page..


  4. I really enjoyed the case study and I’ve been able to share the example to colleagues about how effective IA can be to an enterprise’s business. Any chance you can share the 12 attributes for describing the pages?


Comments are closed.