i-Exchange Knowledge Base Overhaul

What's i-Exchange?

i-Exchange (or information exchange) is a process knowledge base used by frontline colleagues at Santander during every customer interaction. Simply put, it's a big instruction manual for banking processes.

How did I get involved?

In a previous fraud role, I used i-Exchange every day to follow the correct processes. It was full of bugs and usability problems. It was so frustrating. I wanted to do something about it, so I pitched my ideas to the Product Owner. They offered me a job.

I started a discovery phase to understand more about the problem. I wanted to know more about i-Exchange, its users, and constraints.

Research plan

I teamed up with a UX researcher and together we built a research plan. I wanted to understand why people use i-Exchange, how users interact with it, what makes a good knowledge base, what content is held on i-Exchange, what causes colleagues difficulty, and what constraints we needed to work within.

The research needed to show what users' goals are when they open i-Exchange, how they access, navigate, interpret and apply information, where i-Exchange falls short compared with other knowledge bases, and how to prioritise effort going forward.

This work supported all areas of the bank because all customer interaction colleagues use i-Exchange. It also supported the business goal of making it easier for colleagues to make customers' better happen.

Research methods

In order to fulfil my research plan, I used a variety of research methods. I spent time with many different colleagues from contact centre to branch to really understand their needs.

• Surveys

  Surveyed colleagues to understand their experience with the current system.

• Competitor analysis

  Analysed other knowledge base platforms to identify best practices and opportunities.

• Interviews

  Conducted in-depth interviews to understand pain points and the context behind survey responses.

• Workshops

  Facilitated sessions with frontline colleagues to generate user-centred ideas.

• Shadowing

  Observed colleagues in their daily tasks to understand real world usage patterns.

• Journey mapping

  Mapped the site and user journeys, so I knew where to focus my efforts.

Site map

I created a site map to build a thorough understanding of the current website structure. This helped me understand the information we held, how it was organised, and if we had logical information architecture.

Content audit

By auditing i-Exchange content, I gained a better understanding of what users might search for and the additional services to consider as part of the redesign, particularly announcements and Community forums.

What the heuristic analysis found

The heuristic review surfaced problems across Home, Search and Articles. To include all of my findings here would be too much, but the biggest issues were clear.

1. The homepage lacked hierarchy

   A cluttered navigation bar, no audience filters, a decorative hero section, and a mixed communications list made important information harder to scan than it needed to be.

2. Search punished small mistakes

   The engine used exact-match logic with no spelling tolerance, natural language interpretation, autocorrect, suggestions, or fallback result when a query failed.

3. Results did not support confident decisions

   Article metadata was limited to title and ID, so colleagues had to open articles before judging relevance. Role and audience context were invisible at the moment of choice.

Hopeless prompts

Users prompted to use filters after every search query, if only there was a simpler way to do this.

Exact match search

Outdated exact match search bar that was buggy and often led to void results.

Missing information

Article information limited to title, topic and ID. Not enough context to judge relevance before clicking through.

Misplaced iconography

Iconography can help distinguish UI elements, but every result had the same book icon making them meaningless.

Within articles: there was no version history, no feedback mechanism, broken breadcrumb navigation, and content hidden inside collapsed accordions that the search engine couldn't index. Users had to manually expand every section before searching a page.

Broken breadcrumbs

Clicking the navigation options in the breadcrumbs was buggy and resulted in page not found errors.

Missing article features

Users expect functions like sharing, expanding menus and copy to clipboard. Competing knowledge bases always made these available.

Faulty accordions

Content held in accordions could only be searched if accordions were expanded. Users had to manually open multiple accordions to search for key terms.

No version history

Challenging for QA when they needed to review a customer interaction completed on a previous version.

Messy nav bar and no footer

Old nav bar had a messy kitchen drawer with all sorts of systems hidden in a coaching and development drop down.

Wasted hero section

Hero section banner was decorative and overlayed with the search bar and welcome message. Additional banners below make Home feel like an advertising board.

Missing filters

No options to filter audience from Home, so all initial searches were unfiltered.

Confusing communications

Combined list of incidents and communications made it hard to distinguish messages. Manual filters were available but testing showed they were not used.

To include all of my findings in this case study would be too much, so I'm going to focus on the biggest problem that my research revealed.

For a website called Information Exchange, you'd think we had this nailed. However, as a previous user, this was no surprise to me. It's the reason I joined the team. There were key flaws in the system that could be broken down into two pain points.

Pain point 1: Trouble searching

"The search engine is slow, doesn't interpret queries and has no allowance for user errors. This repeatedly wastes time for colleagues and customers."

Exact matching was prone to error and did not support users to find processes quickly. It lacked natural language interpretation and advanced search features such as spellcheck and genius suggestions. If the user made a small spelling error, that search was essentially voided, with no results or suggestions.

Pain point 2: Finding relevant information

"It needs to have an update on the way we find information and the way in which you navigate the system, it is outdated."

"Anything that would assist navigation would help but honestly, I don't know what."

It's frustrating for users to navigate through different business area content, the majority of which will never be useful to them. There are over 6,000 communication articles, how-to guides and technical documents on i-Exchange, categorised into relevant audiences. Setting up i-Exchange to make it easy for each user to access relevant information was a huge challenge.

Now I knew what I was dealing with, and had achieved the goals of my research plan, I was well equipped to start ideating potential solutions.

I'd heard murmurs of a better search engine that was too difficult to implement and was warned about long development times on ServiceNow, the platform that hosted i-Exchange.

How to improve the search engine?

Some ideation sessions require workshops and collaboration. This problem was more technical. Our systems needed a serious upgrade.

There are many ways to improve search engine optimisation, such as changing the way source content is written, including metadata to tag certain search terms, and tweaking the search engine in the back end to weight certain attributes of the source documents differently.

However, the Product Owner had already attempted to use the above options to improve SEO. I would be trying to optimise thousands of documents for a search engine that realistically wasn't up to scratch. We needed a more fundamental fix.

AI powered search results

After hours of reviewing system requirements and technical constraints, I discovered that ServiceNow offered an AI search engine that we could use. In fact, we were already paying for it.

AI search offered improved speed, machine learning, advanced reporting and analytics, natural language interpretation, autocorrect, and genius results.

How to find relevant information on i-Exchange?

I scheduled workshops with contact centre and branch colleagues to tackle the problem of finding information. I framed this as navigation to make the theme simpler to form a workshop. I wanted to open the floor up to my users to let them come up with their own solutions before starting to create any new designs.

It was really difficult in the workshops to keep people on task. I invalidated my first workshop, as I spent the first half of the session talking to them about why we were there in the first place!

This is why I created the handout for the next session, as it meant that people who attended at least understood why the workshop was taking place.

First, I asked how they make toast. This simple exercise gets people empathising with others' experiences and recognising that we don't all do things the same way. Then I asked a simple question: what issues do you have with finding information on i-Exchange?

Everyone put their thoughts down on post-it notes. These comments were really broad, so I used affinity mapping to draw out the key themes.

Personalisation, system issues and quality of content were the three main themes. I was in the process of upgrading the back end systems and had the ball rolling on improving content as part of another project, so I honed in on personalisation.

I asked each participant to respond to a single prompt:

If you could change i-Exchange however you wanted, what would you do to make it more personalised?

If users were able to save their audience, I could automatically apply their preference to filter searches, communications and other content.

With the personalisation concept validated by the workshops and AI search approved for deployment, I moved into design. I started with quick wireframes to explore layout options before committing to higher fidelity.

Wireframing: Getting ideas down

The first wireframe explored how to lay out communications and favourite processes. The second brought in a profile feature to emphasise personalisation and make the saved audience feel like a personal setting rather than a filter.

Hi-fi: Refining before testing

Once the wireframe direction was agreed, I moved into high fidelity. The first version went straight into moderated user testing. Feedback on the footer navigation and the separation of communications led to a refined second iteration.

Now I had some rough prototypes for the homepage, search experience, and articles, it was time to test them. I knew I could create something more polished, but I wanted to validate my designs before spending more time refining them.

User testing

I was limited with user testing tools due to colleagues requiring a downloadable plug-in that IT could not approve. I was not swayed by this and persevered with moderated testing.

Sitting down with colleagues and seeing them interact with the prototype gave me ideas for how to organise the tab and footer menus more effectively. Testing with the comms and incidents team meant that I could refine the communications and incidents list to what really mattered. They also gave me approval to develop a new customisable banner space.

I was able to show them how the saved audience feature would filter their searches and the communications list. The concept landed strongly because it removed a repeated task from their day.

Search testing

Search testing was incredibly important to make sure I delivered on the upgrade with as little friction as possible. Every customer interaction depended on i-Exchange, so it was really important to get this right.

I set up a team of business analysts to run 4,000 manual searches in the test environment. Real user queries were tested on the existing search engine and the AI search engine, with and without audience filters applied. Results were graded as a promoter, neutral and detractor.

Using these gradings, I created a net promoter score to compare both search engines and how much impact using a filter made. This was more than enough validation to push ahead with the upgrade and move into implementation.

Decision to proceed

Testing was complete, the new ideas were validated with users, and I was ready to polish the UI and implement the changes.

Testing was more constrained than ideal: a required browser plug-in couldn't be approved by IT, which ruled out unmoderated remote testing. I persevered with in-person moderated sessions, but it reduced the sample size and the confidence I could take into production. More on that in the retrospective.

I owned the full development life cycle, so I planned the implementation of the search upgrade and scheduled a product backlog to release saved audiences, the new home page, search page and articles.

I adopted a dual-track agile approach for implementation. With the search engine upgrade being worked on by developers in the long term, I could break down saved audiences and UI upgrades into smaller chunks in the short term. This meant frequent releases that put technology in the hands of users as early as possible.

Long term - AI search. The search upgrade was an enormous piece of work that took nearly a year to complete, including challenges with ServiceNow bugs found during development. In November 2024, I deployed the new AI search engine to i-Exchange. I saved months by launching a minimum viable product with an interim UI, then iterating over time as the back end upgrade completed.

Short term - UI changes and saved audiences. I broke saved audiences into sprints so we could deliver changes much faster. First we created a space to store the data, then allowed users to select their saved audience, then displayed it, and finally applied it from the home page. The home page followed a similar approach, with small but frequent updates leading to the full release every two weeks.

Colleagues were apprehensive about change. Many had used i-Exchange for years and were worried that moving too quickly would make their jobs harder. It was important that the new system felt familiar and did not turn everything upside down.

The wider project covered feedback, articles, navigation and search. These screens show the biggest pain points from this case study, plus some smaller issues that were fixed along the way.

Home

• Navbar and footer

  Navigation was simplified, useful links were expanded, and the links colleagues relied on most moved into a new footer.

• Banner and cards

  The static hero became a clickable banner, with smaller cards available for internal initiatives and events.

• Saved audience and AI search

  Users could set a saved audience from the homepage, improving relevance across searches, communications and content.

• Communications and incidents

  General communications and urgent incidents were separated, colour coded and filtered automatically by saved audience.

• Favourite processes

  Favourite articles were added to the homepage under My processes, so regular tasks were one click away.

• Frequently viewed

  The homepage also surfaced articles each colleague viewed most often, personalised by article view count.

Saved audience and AI search

Saved audience available from Home, promoting accurate searches of relevant content with a temporary prompt to encourage adoption on launch.

Navbar and footer

Navigation options simplified. Useful links expanded and included in a new footer.

Banner and cards

Clickable banner making use of the hero section. Smaller cards available to promote internal initiatives or events.

Favourite processes

Added a favourite articles list to Home under My Processes, making sure users had access to their preferred articles.

Communications and incidents

Communications and incidents separated and given colour and icon coding. Saved audiences apply to these lists automatically.

Frequently viewed

Frequently viewed articles included under My Processes. Each user's list is personalised on their article view count.

Search

• AI powered

  Machine learning, natural language processing and improved speed made the search engine more tolerant and useful.

• Save my audience

  Users could amend their saved audience on the search page or remove filters when they needed results from another area.

• Genius suggestions

  The engine could generate summaries from processes and suggest the best option before users clicked through.

• Spellcheck

  Autocorrect suggestions helped users recover quickly when they made mistakes typing under time pressure.

• Relevant data and buttons

  View count, open in new tab and share link actions were added directly to search results.

• Analytics reporting

  Reporting became available for top queries, knowledge gaps, search synonyms and failed searches.

Save my audience

Saved audience can be amended from the search page, allowing users to remove filters to see other business area processes.

AI powered

AI search offers improved speed, machine learning and natural language interpretation.

Genius suggestions

Pulling from processes directly, the engine provides AI generated summaries and suggests the best option.

Spellcheck

Autocorrect suggestions for when users do not have time to go back and correct mistakes.

Relevant data and buttons

View count, open in new tab and share link options now available directly from search results.

Analytics reporting

Full reporting available on top queries, knowledge gaps, search synonyms and failed searches.

Articles

• Digital adoption

  Purple banners highlighted processes that could be self-served digitally, supporting the business goal of promoting digital adoption.

• Version history

  Full version history was made available for each article, helping QA teams review the correct process version.

• Article links

  Copy to clipboard, link sharing, expand all accordions and add to favourites were updated in line with article conventions.

• Feedback

  Feedback options gave users a way to share thoughts, helping knowledge managers see which articles needed improving.

Digital adoption

Purple banners on processes that can be self-served digitally, supporting the business goal of promoting digital adoption.

Feedback

Feedback options allow users to provide their thoughts, helping knowledge managers know which articles need improving.

Article links

Copy to clipboard, link share, expand all accordions and add to favourites updated in line with conventions for article documents.

Version history

Full version history available for each article, allowing QA to do their work and users to experience less confusion during changes.

Once the AI search engine deployed, total search queries fell rapidly, a clear signal that colleagues were finding what they needed on the first attempt rather than repeating searches. Spellcheck, genius suggestions, and machine learning were all working. The analytics dashboard surfaced emerging search trends and knowledge gaps that the content team could act on directly.

Saved audiences reached 50% adoption in the first month after launch without any mandatory training, and tripled the proportion of filtered searches almost immediately. Santander estimates the saved audience feature alone saves at least 120 hours every month across the business, with the true benefit likely much higher when customer time, error reduction, and cognitive load are factored in.

Over the 12-month period, i-Exchange NPS increased by 11.7%, turning around a long-declining trend and bringing user sentiment back on track.