Usability testing worth millions

RBC Capital Markets ran an internal sales platform called Salesbook. Its Opportunities screen was the one place salespeople went to decide which clients to call, about which bonds, with what context. Over time it had grown into an eight-panel wall of data — Axes, Bond Matches, Client Matches, Trade History, New Issues, Portfolio Match, Feedback, Live Prices — that took hours of training to read and slowed the workflow it was meant to support.

I led UX research and design on the redesign: turning an unfamiliar capital-markets workflow into a testable model, validating it with salespeople on the trading floor, and shipping a high-fidelity prototype and visual direction for the product.

The engagement closed in April 2016 with the v1.6 visual delivery — the spec handed to Iris Software for the production Salesbook build. The screen that took hours of training became one a salesperson could enter from any starting point — a calling client, a trader's tip, a search — and read at a glance.

The engagement started with a short deck from RBC's Sales Technology group. They didn't open with a UI ask. They opened with a workflow problem — and named it in their own words.

This is what salespeople were actually looking at — eight stacked panels of dense financial data, side-by-side on a single desktop screen, packed with axes, bond matches, client matches, trade history, new issues, portfolio matches, feedback, and live prices. The shape of the problem was clear: not a missing feature, but a missing model of work.

The instinct in a project like this is to start with the screen: simplify the panels, rework the hierarchy, redesign the components. I held that off. The same brief had a second slide that gave away the deeper problem — and it wasn't the UI.

RBC mapped the existing flow as three rigid steps: a list of axes appears, salespeople pick one, and everything else updates around the selection. That worked when an axe was always the starting point. But salespeople actually entered the screen from anywhere — a client calling in, a trader walking over with a security in mind, a portfolio note from yesterday. The screen had everything they needed; it only had one way to ask for it.

So before any layout work, I pushed on the workflow itself. Free search as the front door, with security-first and client-first as named entry paths. The screen would adapt to the way the question was being asked, not the other way around.

The work moved across three organizations. Each had its own incentives, its own vocabulary, and its own definition of "good." Most of the design judgment in this case study lived in the seams between them.

The hardest part of the project was not screen layout. It was understanding what mattered in the sales workflow. Salespeople were not watching live markets the way traders were. They were looking for fit: which client, which security, what timing, what history, and what conversation would create a useful opportunity.

Early concepts were built with limited access to end users, so the team had to make assumptions and validate them quickly. Once we met users, I shifted the work from a generic dashboard concept toward task analysis: how salespeople start their day, move between client and security context, and capture feedback after a call.

The trading floor context shaped the product constraints. People sat close enough for verbal communication to matter. They worked from the office, not on mobile. Their layouts needed to be customizable because every desk and coverage area had different tab structures and habits.

The plan for each floor visit was deliberately small: about 15 minutes of structured Q&A per salesperson, then 15–20 minutes of silent observation. We optimized for seeing the workflow more than hearing about it — a sales floor is a place where what people do often differs from what they say they do.

Once the three workflows were named, I spent the first week of design on paper, not in Sketch. The cost of an idea on paper is roughly zero; the cost of an idea in pixels is a day. At this stage I wanted to be wrong fast and often.

The frames below are seven of the early sketches that stuck. They aren't pretty — they're working drawings, the kind that get scanned at the end of a session and forgotten until the case study. Click any one to take a closer look.

I built three directions in Axure rather than one. The reason was timing: salespeople had five-minute windows between client calls, and the only honest way to learn anything in five minutes was to put real options in front of them and watch which one they reached for. Each direction held the same workflow constant but moved one variable — tile density, visual tone, or the role of the sidebar — so user reactions could be attributed to a specific choice rather than aesthetics in general.

Below is the Axure prototype itself, captured as a 30-second walkthrough. The density is deliberate: a 24-client book across the top, an axe board of bond tiles, a full holdings table. Salespeople can't react honestly to placeholder data, so the prototype didn't have any.

We ran fast, five-minute sessions with salespeople on the trading floor. Two participants kept coming back across the engagement and shaped the prototype most directly. Below is what each of them said in their own words — and what shipped because they said it.

In parallel with floor testing, I took the three prototype directions to a wider review inside RBC. Five reviewers, each reading the same designs through a different lens — and almost no agreement between them.

The surface read was that everyone disagreed. The real read was that the disagreement had only one axis: information density vs. modern clarity. G. and C. were both arguing for density because the screen lived next to a Bloomberg terminal on a trader's desktop. E., A., and R. were arguing for clarity because the previous screen had failed them on exactly that.

The trap was to treat this as a vote. The work was to recognize that the two camps weren't actually in conflict — they were describing different parts of the same problem. The screen had to be denser than a modern web app and clearer than the one it was replacing.

The final visual direction did exactly what the synthesis said it would. Tile as the unit of an axe, scannable at a glance. Density gained back through layout, not by reverting to an eight-panel wall. Semantics taken seriously — every chart with its time-period, every dimension carrying one meaning, "Updates" renamed "Alerts." Modern enough that a salesperson would want to open it; dense enough that it belonged on a trading desk.

Consulting work usually ends at handoff, and you rarely see the long tail. I left this engagement at the v1.6 delivery in 2016. It resurfaced two years later, in RBC's own newsroom — not something I went looking for.

In March 2018, Kim Prado, the RBC executive who led the Salesbook program, was named Technology Innovator of the Year at the Women in Data and Technology Awards in London. The write-up credited Salesbook with streamlining "colleague access to client information which had hitherto existed in disparate locations," and her team with using "machine learning techniques to connect RBC inventory…with client data points to present attractive trading proposals to our sales teams."

That last line is the security-first workflow this case study is built around: securities in focus, ranked against client fit, surfaced as opportunities. To be precise about scope — I was the embedded UX lead on the Opportunities redesign inside a much larger program, not its owner, and the recognition is hers and her team's. But it's rare to get external, on-the-record confirmation that the model held up, shipped, and mattered enough to be named publicly. This is the closest thing I have to it.