WorkTechAdvisor
TechAdvisor | Caterpillar
I transformed a fragmented, expert-driven recommendation process into a clear, explainable experience that helps dealers confidently recommend technology at both single-asset and fleet scale.
Figma
FigJam
Confluence
Azure DevOps
Word
Excel
ChatGPTOutcome & impact
Got just a minute?
It’s a long one: two phases, nine months. Grab a coffee.
A customer calls his CAT dealer. He runs a site with 100 machines, productivity has dipped, and he wants to know which technology will fix it. The sales rep doesn't have the answer, so he digs through technical docs, loops in three CAT technical teams, and promises to get back. By the time a quote arrives 2–3 weeks later, the opportunity is already lost.
When I got the brief, it seemed like a straightforward recommendation tool. But research revealed a different challenge: dealers didn't need another tool, they needed a way to answer customer questions within 30 minutes, before the customer hung up or moved on.
Dealers struggled to confidently recommend technology due to fragmented information, complex configurations, and reliance on experts, slowing decisions and increasing errors.
I led the end-to-end UX: shaping the recommendation strategy, translating constraints into design decisions, and designing a scalable experience in close collaboration with Product, Engineering, stakeholders and SMEs.
Final design highlights
Context
This was a 0→1 product initiative delivered in two strategic phases, covering the entire product lifecycle from discovery and research to design, validation, and delivery.
TechAdvisor is a smart, centralized platform designed to help dealers and customers discover, understand, and get capability recommendations for Caterpillar assets.
Why · the problem
Caterpillar offers a growing portfolio of digital solutions and machine technologies. While these technologies deliver strong value to customers, dealer teams struggle to confidently recommend, configure, and support them.
The challenge
Reducing dependency on specialist and long internal coordination cycles
Easy for dealers to digest the recommendations data
“To transform an expert-driven process into a dealer-ready recommendation experience that is fast, explainable, and scalable, reducing coordination dependency and enabling confident recommendations, so dealers can move from weeks-to-minutes decision making.”
Enabling data-driven foresight to support smarter and faster decision-making.
Recommend the right technology with confidence
End to end process
Discovery ⇢ Concepts ⇢ Concept testing
Initial discovery
I worked closely with our UX researcher and we conducted discovery sessions with users across Western Australia and the US. The goal was to understand the real shape of the problem before designing anything.
Instead of asking "what features do you want?" we focused on how recommendations were made today, what steps caused delays, and where users hesitated or escalated to experts.
CAT Dealers
Gender split
84% Male15% Female
7 dealer-side experts
Remote, via video call
1 hr per participant
Western Australia, US
Account Managers, Quoting Specialists, Product Specialists
3 dealerships, 2 continents, one shared struggle.
Insights · what we studied from the users
What did every interview point back to?
GunaDealers don't think in products, they think in problems, and they need answers quickly before the customer hangs up.
The reframe
If TechAdvisor shows a machine's current state upfront and explains why each recommendation fits, dealers will be able to answer a customer inside a single call, without escalating to a technical specialist.
From insights to concepts
Our goal in concept design was not to create a perfect system, but to design a flow that removes hesitation for users at the moment of decision making. Some of our core goals are:
Concept testing · what we learned
Why test again after several interviews?
GunaInterviews gave us the understanding of the problem, but users had nothing concrete to react to. Concepts changed that: now we could check whether our direction was right, watch how dealers actually hunt for technical information during a cross-sell or upsell, and see where their current tools fall short.
We tested the concepts with the same set of users who we did interviews with.
From concepts ⇢ real product
The plan vs the reality
Single asset first as the foundation, multi-asset on top. Helios looked feasible then, so multi-asset was going to work exactly like single asset: a dealer searches, the system pulls the fleet of one customer, one product family, one machine model or prefix. One interaction model, two scales.
What is Helios?
GunaHelios is a technology which helps the system to pull all the assets data directly from different sources with ease, in a matter of time, quickly.
Helios came back as a 12–18 month implementation. The multi-asset entry point depended on it.
We changed the door, not the house. Fleets come in by import instead of search. Everything after the entry point was already built on the single-asset foundation: capability selection, recommendation structure, card patterns. None of it had to change. And when Helios lands, search can be added as a primary door with no redesign.
Defining the MVP
Concept testing gave us the direction; phase 2 was about sequencing it. We built single asset first, not as a smaller version of the product, but as the foundation: the card patterns, the state indicators, the recommendation structure everything else would inherit. Multi-asset would layer on top, using the same interaction model at fleet scale.
Single Asset Recommendations
Focus 01Designed as the foundation of trust, clarity and for scalability.
Multi-Asset Recommendations
Focus 02Designed for real-world dealer workflows, where recommendations are rarely made for one machine at a time.
The critical decision
Multi-asset would work like single asset: search for assets, get recommendations. Helios integration made that possible.
Helios came back as a 12–18 month implementation. Our entry point for the entire fleet experience sat behind it.
Ship the door we could build now. Dealers import a group via Excel instead of searching for it.
Pushed backInitial direction
ShippedFinal decision
Why not just wait for Helios?
GunaWaiting meant shipping nothing for over a year while dealers had needs right now. Import was the less elegant door, but it opened immediately and, because everything behind it was already built on single-asset patterns, search can be added later without a redesign.
Only the entry point changed. The foundation stayed the same.
Key iteration
We explored a minimal approach where the import process was compressed into fewer screens, reducing the perceived effort of getting assets into the system before making recommendations.
RejectedFewer steps
The whole import on one screen, minimal clicks, fastest possible entry.
Fewer clicks, but it broke the upload pattern users knew from every other CAT application. Unfamiliar means hesitation, errors and a new learning curve.
ShippedStepper-based flow
Guided steps that match the upload pattern used across the CAT ecosystem.
More steps, but users could predict the next move. Predictability beat saving two clicks.
Constraints
Max 200 assets in a group
A validation runs before import, splitting accepted and rejected assets with a reason against each row.
One product family per run
A product-family filter appears before generation, making each run a deliberate choice instead of a surprise.
Solution
I need an advanced way to search the assets, and I can't recommend anything until I know exactly what this machine already has. Today, finding that out takes longer than the customer call itself.
Customer-specific asset specs must be shown upfront.
We designed a fast, intelligent entry flow that helps users quickly find the right asset and understand its current state before exploring recommendations. Deeper technical details are available on demand via a side drawer, ensuring users have the right context before making upgrade decisions.
Impact →Dealers reach the right asset and its full technology state in seconds.
Recommendations are never for one machine. My customers run a fleet, so going asset by asset just isn't realistic.
Fleet-level recommendations are the real-world use case.
Fleet recommendations start with an Excel import. Every row is validated, and accepted and rejected assets are separated with clear reasons, ensuring only valid data moves forward.
Impact →Getting a 200-asset fleet into the system became a minutes-long task, with errors caught upfront instead of discovered after.
Every import lands as a named asset group showing its created date, processing status, and last-processed time at a glance. Users open any group to view its assets, watch processing complete, and generate recommendations the moment a group turns ready: no re-imports, no asking around.
Impact →One list answers "where is my fleet, and is it ready?" with full traceability from upload to recommendation.
I'm overwhelmed by the capabilities CAT offers; we're swamped by a sea of solutions. I have no clear picture of what to pick for the right customer problem in that quick span of time.
Dealers asked for use-case guidance, not product lists.
Capabilities appear as cards grouped by use case, by customer problems, not product names. The system auto-detects site connectivity and pre-selects what the asset supports, and every card carries a visual state indicator separating what's already active, what's available, and what can be enabled.
Impact →Users skim, scan, and shortlist from the full CAT catalogue without expert help, and the available-vs-added confusion from testing disappeared.
Evaluation needed to support two speeds. A quick view opens any capability's benefits and recommendations in a modal: no waiting, no generating, no page loads. When the choice is close, users pull options into a side-by-side comparison to see exactly how capabilities differ before committing to one.
Impact →Faster decisions with less cognitive load.
The multi-asset screen reuses the single-asset design to keep the experience familiar. Only three things change: connectivity is selected manually, the context reflects the selected asset group, and capabilities are single-select because this runs across many assets.
Impact →Users carried their mental model straight from one machine to 200: fleet capability without a second learning curve.
Getting recommendations isn't the finish line. I still have to understand it quickly, and I have to explain it to the customer in words that make sense to them.
Concept testing showed recommendations need stronger context, not just actions.
For a single asset, users can review recommendations alongside asset details. Capabilities are organised in the left navigation for quick access, while recommendation use cases are grouped into accordions to progressively reveal details. An overview summarises each capability and its key benefits, users can edit recommendations before implementation, and actionable recommendations are highlighted with icons for quicker interpretation.
Impact →Recommendations stopped being expert output to decode; they became customer-ready material a dealer can walk through live.
Fleet recommendation output leads with a high-level summary: capable, incompatible, and active assets at a glance, with every upgrade category quantified across the group. A detailed view then breaks down exactly what each asset needs: subscription, hardware and software. The whole thing downloads as a single package for the customer conversation.
Impact →A 200-asset recommendation run reads in one screen: scan the totals, drill into any asset, hand the customer a number.
Design validation
Step 01
Moderated usability testing with 7 dealer participants across Australia. The experience scored 85, above the referenced SUS average of 68. The effort rating is the average across four core flows: asset discovery, import, capability selection, and recommendations.
Step 02
HFI evaluation model, reviewed by usability experts across navigation, content, presentation, and interaction. The craft held up to scrutiny.
Impact
Weeks → Minutes
Time to a customer-ready recommendation, against the 2–3 week baseline mapped in discovery.
Max 200 assets per run
Fleet-scale recommendations, previously impossible.
Expert led → Dealer led
The dealer completes the recommendation with far less escalation.
GunaThe knowledge already existed. Access didn't. Every answer sat behind multiple teams and hand-offs. By putting expert reasoning where dealers already worked, a 2–3 week process became a conversation that could end on the same call.
What's next
When Helios lands, multi-asset gets the search entry point we originally designed for. Nothing behind it changes: the import flow and the search flow drop onto the same foundation. The pivot cost us an entry point, not a rebuild.
Today a fleet run is generated on demand and downloaded before it clears. Persistence was the right model; the effort wasn't available in this release. It's the first thing I'd push for, because "generate again to look again" is a tax on exactly the fleet-scale workflow we opened up.
Both ceilings are system limits, not design ones. The validation and filter patterns we built to make them survivable are the same patterns that come out when the ceilings lift.
Each of these was a live trade-off during the project, not an afterthought. The version that shipped was designed so that none of them requires a redesign to resolve.
Choosing single asset first was about shipping something solid. Then Helios came back as a 12–18 month implementation and the multi-asset entry point collapsed. Only the door had to change; everything behind it already sat on single-asset patterns. I'd made a scalability decision months before I knew I'd need it, and I now make that decision deliberately rather than by instinct.
The brief asked for a recommendation tool, and we'd sketched a capability catalogue. Seven interviews later the real need was confidence inside a 30-minute decision window: a different product with a different measure of success. We'd have built a good catalogue that nobody could use on a live call.
AI helped me speed up the design process by quickly validating my design logic, exploring initial concepts, and identifying gaps early. It also helped me create first drafts of UX copy, making stakeholder reviews more efficient. AI helped me move faster, while the final design and content decisions remained mine.
Nearly every meaningful decision on this project was made against something I couldn't remove: a 12–18 month dependency, a 200-asset ceiling, an engineering effort budget. What I took from it is that constraints don't dilute design work, they're what make design decisions legible. The version of TechAdvisor that shipped is the version that could ship, and I can defend every trade-off that got it there.
Guna Sekaran
If you have any more questions or want to know more details, don't hesitate to get in touch. Meanwhile, take a look at my other work, my experiments, or learn more about me.
Open for opportunities
I partner with teams to turn complex problems into clear, human-centred solutions.
