A home page 500 sellers
would actually use first
Salesforce ships a generic home page. For commercial insurance sellers at Travelers, that page showed a list of tasks with no prioritization, a news feed nobody read, and metrics disconnected from their daily goals. Eighteen months of research, usability testing, and a live pilot built a replacement that sellers open first instead of navigating around.
Parts of this case study have been altered for portfolio purposes. Names, interfaces, and data are recreated, with no proprietary Travelers systems or real customer information shown.
| Agency | Written premium | My flow | Agency flow |
|---|---|---|---|
| Offenhauser & Co | $1,584,286 | ||
| Insurors of Texas | $1,612,850 | ||
| Bolt Insurance | $1,464,925 | ||
| Comerica Inc | $1,287,341 |
| Rep | Open pipeline | Bound YTD | Status |
|---|---|---|---|
| J. Torres | $1,148,000 | $612,000 | On pace |
| M. Chen | $890,400 | $503,900 | On pace |
| A. Patel | $402,100 | $188,200 | At risk |
| G. Sorg | $735,600 | $441,000 | On pace |
| K. Eakes | $519,300 | $276,400 | Watch |
| Scheduled 2+ wks | Under 2 wks | YTD NB$ | |
|---|---|---|---|
| In person | 10 | 12 | $200,000 |
| Virtual | 8 | 14 | $150,000 |
| Account | My written PY | Agency written | Flow |
|---|---|---|---|
| Suffolk Insurance | $1,500,000 | $1,500,000 | ↑ |
| Marsh USA, Hartford | $1,200,000 | $1,200,000 | — |
| All Protector Agency | $500,000 | $500,000 | ↑ |
| Marsh USA, Morristown | $30,000 | $30,000 | ↓ |
| Account | Effective date | Stage | Strength | Est. premium |
|---|---|---|---|---|
| Tabcom LLC | 7/1/2026 | Prospect | Medium | $130,000 |
| Lancaster Bible College | 7/1/2026 | Prospect | Weak | $199,923 |
| Gertrude Hawk Chocolates | 7/1/2026 | Prospect | Weak | $650,000 |
| Hendrick Manufacturing | 7/1/2026 | Submission | Weak | $270,000 |
| Acmwad Partnership LP | 7/1/2026 | Submission | Medium | $94,200 |
| Lehigh Carbon Cmnty College | 7/1/2026 | In Play | Strong | $318,400 |
| Desert Breezes Resort | 4/10/2026 | Propose | Medium | $88,600 |
| Infinity Healthcare Mgt | 5/4/2026 | Bound | Strong | $412,750 |
The problem with "default"
Salesforce ships a generic home page. For commercial insurance sellers at Travelers, that page showed a list of tasks with no prioritization, a news feed nobody read, and metrics disconnected from the goals those sellers are measured against. Most reps navigated away within seconds of logging in.
The business had two versions of the same question: why is CRM adoption low, and why do sellers say the tool does not help them sell? The answer was the same in both cases. A home page built for the median Salesforce user is not built for a commercial insurance Account Executive managing a $3M renewal book.
I open it, see nothing useful, and go straight to my accounts list. The home page is wasted space. Account Executive, discovery interview
This project is the eighteen months between that sentence and a page those same sellers now open on purpose. It is also, and I want to be straight about this, a project where the most interesting findings came after launch rather than before it.
Short for Account Executive: the seller who owns the agency relationships and the book, works new business and renewals through the pipeline, and is measured on written premium, retention, and agency visits.
Short for Territory Manager: the leader who runs a team of Account Executives. They need the same data aggregated rather than personal, which turned out to be a different page and not a filter on the same one.
Who was in it
One designer, a standing group of sellers who kept showing up, and a lot of people with a stake in the page. Research ran as weekly advisory sessions in Mural with Account Executives and Territory Managers pulled from across the commercial business units, a separate pool of usability participants who had never seen the screens, and a pilot instrumented well enough that we could read what people actually did instead of what they told us they did. Findings went straight into the Salesforce development team's sprint, which is the only reason a page this configured shipped at all.
What sellers actually needed
Three research phases, six seller and leader roles, and data from every major commercial business unit. Every design decision below traces back to a specific finding, and where it does not, I have said so.
Discovery interviews
How sellers structure their day, what they check first, and what Salesforce currently helps them do or does not. Run across roles, because we already suspected the answers would diverge.
Contextual inquiry
Watching sellers in their actual workflow. This is where the gap between stated needs and observed behavior showed up, and it is the phase that changed the most minds internally.
Advisory sessions and testing
Weekly working sessions in Mural on one component at a time, then two rounds of moderated usability testing with sellers who had never seen a screen.
What sellers said they needed at the start of their day
Where the default page failed
Today, not last week
Sellers wanted to start the day knowing what needed their attention now. The default page was a record of what had already happened, which is a different thing and a less useful one.
The common core was too small
AEs wanted account health. Territory Managers wanted team pipeline. BDDs wanted submission activity. One page could not serve all three without configurability, and the overlap was not worth designing around.
Nobody knew how old the data was
Data in Salesforce was not always current, and sellers had learned to distrust the tool. Any redesign had to surface data with timestamps and source attribution before it surfaced anything else.
How the advisory sessions ran
One component per session, in Mural, with the real widget pasted at the top of the board and four columns underneath it: what value does this offer you today, what action would you take, what information is missing, and what would you change. Sellers wrote their own stickies. I did not transcribe for them, because the phrasing turned out to matter.
The board below is the New Business Premium session, and it is the one I point to when someone asks whether this kind of session is worth an hour of a seller's week. The value column is thin. The missing information column is enormous. That imbalance is the finding, and no survey would have produced it.
One component, one hour
Narrow enough that people can be specific. A session about "the home page" produces opinions. A session about the NB Premium gauge produces the sentence about not knowing where you stand against your goal.
Rank before you discuss
The Pulse Check ran first, silently, so the loudest person in the room did not anchor everyone else's score. Then we talked about the spread.
Never make them classify their own feedback
I sorted the stickies into value, action, missing, and enhancement afterward. Sellers are good at saying what is wrong and bad at saying what kind of wrong it is, and it is not their job.
From findings to frames
Discovery research produced 47 distinct findings, consolidated into 12 design requirements. Each requirement has a direct line to the usability tasks it resolves. These are the five that drove the most significant structural decisions, and the ones I would defend in a room.
Sellers open Salesforce and leave immediately
No actionable content above the fold. The page was a lobby, and everybody walked through it without stopping.
A priority strip and a pipeline widget in the first column, sorted by due date and opportunity size rather than alphabetically. This eliminated the blank-page navigation reflex that showed up in every contextual inquiry session.
Territory Managers and BDDs have different data needs than AEs
A single layout served no one well. The leader view in particular was being used by people whose job it did not describe.
Role-based widget configuration. Each persona gets a distinct widget set, with no shared compromise layout. Configurable on first login and adjustable at any time. This is the decision that cost the most to build and the one I would make again.
Renewal and engagement dates were buried
Sellers tracked them in personal spreadsheets because Salesforce put them three clicks deep, which meant the system of record was not the system anyone recorded in.
An engagements widget on the home page showing the next seven days, with a direct link to each agency record. The next-7-day view turned out to be the least-used feature in the pilot logs, which is covered below and is not a comfortable finding.
Sellers distrusted the data
They had no way to know when a number was last updated, so they assumed the worst and went to the report instead.
Data freshness timestamps on every widget. Shown in subdued type: present for the seller who needs it, invisible noise for the seller who does not. Sourced from the Salesforce sync logs rather than typed in by hand.
Eleven widgets visible by default
Sellers could not identify what was important, so they treated all of it as unimportant. Contextual inquiry showed them scanning past most widgets without reading them.
Six widgets on the initial view, chosen by highest-frequency task per role. Additional widgets stay accessible through "Add to home." Fewer widgets that matter beats more widgets that don't.
The Figma exploration ran to more than forty numbered layout options before it settled. Most of the variation was not about which widgets existed but where the KPI strip sat, whether the NB Premium gauge belonged at the top or the bottom, and how much vertical space a pipeline funnel is worth. Those three questions took months.
What the first test showed
Five sellers across three roles. Eight tasks. Three tasks failed, all for the same reason: the layout did not communicate which widgets were interactive and which were display-only.
The SUS score of 55, which is a grade D, matched what we saw in session. Sellers completed tasks but expressed confusion at each step. The prototype was structured correctly and its affordances were absent, which is a genuinely useful kind of failure because it is cheap to fix.
Three changes followed directly from the session notes. Clickable widget headers got a visible underline-on-hover state. Renewal and engagement dates moved from the second column to the first. The pipeline bar chart was replaced with individual metric tiles that each link to a filtered report view, which is the change in the comparison below.
The default page as sellers found it. Eleven widgets, ranked by nothing, with the largest piece of real estate given to a feed that zero sellers reported using.
Testing with real users over eight weeks
The pilot ran for eight weeks with a cohort of AEs and Territory Managers across two commercial business units. Every component on the page was instrumented, and I wrote the analytics key that maps each tracked event string back to the component it belongs to, so the usage log could be read as a home page rather than as a list of strings nobody recognized.
Before the pilot, the median seller spent under twelve seconds on the home page before navigating away. By week four the dwell time had increased and the navigation path had changed: sellers were clicking into accounts and opportunities directly from the home page instead of using the Accounts tab as an entry point.
Then the log told us three things we did not want to hear, and those are the three worth writing down.
My Pipeline Refresh alone accounts for 3,366 of the 8,998 tracked interactions. Sellers were not browsing the page. They were pressing refresh on the one number they are measured against.
Three findings the log produced that no session did
Refresh is the whole story
My Pipeline Refresh was used 3,366 times, more than the next two features combined. That is not engagement with a home page. That is a seller checking one number repeatedly because the underlying data is not live, which means F4 was solved cosmetically rather than actually.
Usage is concentrated, not broad
Commercial Accounts logged 4,236 interactions. Select logged 7. A page adopted by two business units and ignored by three is not an adopted page, and averaging across them would have hidden that entirely.
Requested does not mean used
The next-7-day engagement view, asked for repeatedly in advisory sessions, was used 53 times in five weeks. The pipeline stage filter, 14. The group that helps you build is not a reliable guide to what gets used.
What the survey said alongside it
Sixteen sellers answered a post-pilot survey. Thirty-seven percent rated the page easy or very easy to use, which is not a good number, and the honest read is that the page is better than what it replaced and still not good. Sixty-nine percent reported no new challenges introduced, and the five who reported challenges wrote the most useful sentences in the entire study.
The second of those three is the one that stings, because it is F2 coming back around. We built role-based views, and a seller still ended up in the wrong one. The routing logic assigned the manager view by title rather than by what the person actually does day to day, and a Managing Account Executive does both jobs.
Eight decisions, each traced to a finding
None of these are styling changes. Each one moved structure, and each one has a research finding or a log line behind it.
Chatter feed removed
Zero sellers reported using the Chatter feed regularly, and it held the largest piece of real estate on the page.
The space now carries the highest-frequency use case for each role. For an AE that is the pipeline. For a Territory Manager it is team performance by rep.
Role-based layouts introduced
67% of sellers wanted their view to match their role. One layout forces compromises that satisfy no one fully.
Each role sees the widgets that match its daily responsibilities. Still imperfect, because assignment runs off job title and titles do not always describe the job.
Pipeline chart replaced with metric tiles
The bar chart looked informative and led nowhere. Sellers clicked it repeatedly in round one expecting it to open.
Each tile carries a direct link to a filtered report view, turning a display into a navigation point. The two View Report events together logged 1,313 clicks in five weeks.
Widget count reduced from eleven to six
Contextual inquiry showed sellers scanning past most widgets without reading them.
Six widgets on the initial view, chosen by task frequency per role. Everything else moved behind "Add to home" rather than being deleted.
Data timestamps added to every widget
Sellers had learned to distrust Salesforce data because they could not tell how old it was.
Last-synced time on every widget, in subdued type. This addressed the trust gap without changing the underlying data infrastructure, and the refresh numbers suggest it did not fully close it.
NB Premium gauge given a goal
"I don't know where I am relative to my goal" came up in the advisory board more than any other sentence. The gauge showed a number against no scale.
The gauge now reads against the seller's annual target, with the YTD figure and the pace both visible. A number without a goal behind it was the most-criticized component in the whole Pulse Check.
Leader views added for TMs and FVPs
Territory Managers opened six individual rep records to get a team overview.
The TM view aggregates team pipeline and surfaces at-risk accounts on page load, which replaced a six-click routine with a zero-click one.
Customers with Recommended Products demoted
It ranked lowest in the Pulse Check and logged 81 report clicks across five weeks. Sellers called it "nice to have, towards the bottom."
Moved off the initial view into "Add to home." Nobody asked for it back, which is the cleanest possible confirmation of a ranking exercise.
What this project proved
The platform is not the constraint
Salesforce's Lightning components are more flexible than most teams assume. The constraint was not the platform. It was the absence of research that could justify the configuration decisions. Once we had data, the development team could act on it. Enterprise CRM redesigns fail not because the tool cannot be changed, but because nobody has asked the people using it what they actually need.
The pilot log disagreed with the advisory group, and the log was right
Two rounds of usability testing got us to a page that performed well in a controlled session. The pilot showed what testing cannot: usage behavior over time, which widgets sellers actually returned to, and which features they understood in isolation but never opened in practice. The next-7-day engagement view was requested in session after session and used 53 times in five weeks. I would now instrument earlier and trust stated preference less.
Averages would have hidden the real finding
Aggregate adoption across all business units looked healthy. Split by unit, two were using the page daily and three had effectively not adopted it at all. The question worth asking of any adoption number is not how big it is but how evenly it is distributed, and I built the analytics key specifically so that question could be asked.
Role count matters
Designing for six roles across five business units means six sets of discovery interviews, six sets of usability tasks, and six post-pilot surveys. The temptation is to find the common core and design for that. The finding was that the common core was too small to be useful. A Territory Manager and an Account Executive have almost nothing in common on a Monday morning. The separate role views cost more to build and were worth it.
Still open. The BDD and Managing Director widget sets were scoped and designed but never piloted at the same scale as AE and Territory Manager, and both are on the research roadmap. Role assignment by job title is the clearest defect the pilot produced and it is not yet fixed. The three business units with near-zero adoption have not been interviewed about why, which is the most important open question on this project and the one I would start with. Second-round usability figures exist in the original study documents and are not reproduced here, because I have not been able to source them and I would rather leave a gap than publish a number I cannot stand behind.
Keep exploring
A year-long rebuild that combined legacy underwriting and sales platforms into one system, designed for the underwriters who live in it.
A one-day workshop that turned months of underwriter research into a working, coded AI assistant, then four rounds of iteration into a trust-first tool.
A team workshop redesigning how tasks get triaged and assigned across a distributed team, tested against a real backlog.