Option 01
Power Apps
- Centralized the resource pool
- Search and categorization stayed shallow
- Licensing costs plus ongoing IT dependency
Limitation: Fixed storage, not retrieval. Needed IT to maintain long-term.
UX Research · Strategy · Service Design
Researching the information gap behind slow concierge responses.
A UX research and strategy project investigating why concierge staff struggled to find answers quickly—and translating those findings into a focused AI-powered solution.
Implemented at Ingleside, a multi-community senior living organization.
Read Ingleside’s project story ↗
89%
faster response time
30+
minutes saved per shift
100%
adoption across three communities
01 · Context
The concierge team had answers.
Ingleside's concierge teams are the first point of contact across three communities. Their information lived everywhere: binders, SharePoint, printed schedules, and coworkers' memory.
A simple question could mean checking five places for one answer.
The problem
Concierge staff spent 30 minutes a shift, on average, searching for information that already existed.
That time came out of time meant for residents.
Why this was critical
Slow retrieval became an operational and trust problem.
When staff could not retrieve reliable information quickly, the impact extended beyond one delayed answer. It affected consistency, resident confidence, and the people pulled away to help resolve the question.
Inconsistent answers
Residents could receive different information depending on the community, shift, or staff member.
Resident trust
Long waits and uncertain responses weakened confidence in the concierge desk as the primary point of contact.
Care-team disruption
Questions escalated to coworkers and departments, interrupting staff coordinating higher-priority work.
Knowledge dependency
Resolution speed depended on who was working rather than on a reliable system available to everyone.
02 · The question
Why was it taking so long to resolve a simple question?
Before designing a solution, I wanted to understand what was actually causing the delay. Two competing hypotheses emerged—and each would lead to a completely different intervention.
H1 · System cause
Information is difficult to find.
Information exists, but it is scattered across unstructured resources. A discoverability problem.
H2 · Human cause
Staff lack the knowledge or capacity to answer.
Staff may be undertrained, overloaded, or unsure how to resolve questions. A capability problem.
So I didn't start by designing a chatbot. I started by finding out which problem actually existed.
03 · Research strategy
Discovery had to answer "why" before design could answer "how."
I used research at multiple levels to understand the problem from both an operational and frontline perspective.
01
Desk discovery · Week 1–2
Stakeholder interviews
4Community Engagement Managers
I pressure-tested the initial problem framing, understood operational constraints, and gathered assumptions to carry into frontline research.
02
Field discovery · Week 2–3
Concierge research
13team members
7observed
6interviewed
3facilities
I segmented participants by tenure, shift, and facility to observe what happened when a real question needed an answer.
Research questions
What I needed to understand.
When you don't know an answer, what do you do? How do you search? Where do you get stuck? How often do repeat questions consume your shift? Is it faster to search yourself or ask a coworker?
User interviews
What people said
Interviews helped reveal beliefs, workarounds, confidence levels, and perceived barriers.
On-site observations
What people did
Observation showed the real workflow, including where self-service broke down and when staff turned to teammates.
04 · Current-state journey
Where does the resolution process actually break down?
The map connects the observed behavior to the opportunity: staff tried to self-serve first, but scattered and shallow search pushed them toward teammate escalation.
Resident asksA question arrives at the desk.
InterpretConcierge understands the request.
SearchChecks binders or digital resources.
StuckInformation is difficult to retrieve.
EscalateSelf-service breaks down.
Ask teammateReaches out on Teams.
ResolveAnswer finally reaches the resident.
Staff weren't avoiding the search. They were already trying to self-serve.
The bottleneck wasn't willingness or effort. It was retrieval. When information couldn't be found, staff were forced to rely on teammates.
05 · Findings
Three findings pointed to one root cause.
The research shifted the problem from a question of staff capability to a question of information access.
01
Information is buried, not missing.
The resources already existed. The problem was that staff couldn't reliably retrieve them when they needed them.
02
Staff were already working across too many systems.
Multiple tabs and tools were open during a single interaction. The solution couldn't become another system to manage.
03
The gap wasn't skill or effort. It was access.
Training, confidence, and escalation patterns didn't point to an undertrained workforce. Staff knew their jobs. The information wasn't accessible at the right moment.
06 · Research turning point
This wasn't a training gap.
It was an access gap. The evidence confirmed one hypothesis and did not support the other.
H1 · Confirmed
Discoverability was the problem.
Information was genuinely difficult to find. Search time averaged around 30 minutes, and self-service often ended in teammate lookups.
H2 · Not supported
Capability wasn't the primary problem.
The research didn't show that staff training or effort was the main driver of slow resolution.
Not a training gap. An access gap.
Why discoverability failed
Why did the access gap exist?
The information existed, but the retrieval systems did not support the moment of need. Concierge work required direct, contextual answers—not a longer path through folders, files, and search results.
Physical resources
Manual information could not be searched.
Critical answers were distributed across binders, folders, printed schedules, and handwritten notes. There was no shared tagging system or reliable way to locate the right page quickly.
Digital resources
SharePoint returned matches, not answers.
Keyword search surfaced every document containing a term, even when the result was not relevant. Staff still had to open files, read through them, and interpret the answer themselves.
The opportunity
Move from keyword-based document retrieval to contextual answer retrieval—so staff could resolve questions quickly without searching the entire knowledge base.
07 · Research → requirements
What the solution had to do.
The research translated directly into design requirements. Each requirement addressed a specific observed behavior or constraint.
01Near-instant retrieval.
Move from ~30 minutes to seconds.
Move from ~30 minutes to seconds.
02Contextual search.
Ask naturally rather than remember keywords.
Ask naturally rather than remember keywords.
03Zero additional systems.
Work where staff already work.
Work where staff already work.
04One source of truth.
Replace scattered resources.
Replace scattered resources.
05Easy ownership.
Managers keep content current.
Managers keep content current.
08 · Problem definition
How might we give concierge staff instant, accurate answers—without adding another tool to an already crowded desktop?
09 · Exploration
Digitize first. Then choose the retrieval solution.
AI could not compensate for fragmented or inaccessible source material. The first design decision was therefore an information-governance decision.
01
Digitize critical manual resources
I advised the team to convert the highest-value binder content and secured buy-in from the Community Engagement Managers for the digitization process. This created a maintainable knowledge foundation before evaluating any interface or AI layer.
Option 02
SharePoint + Copilot considered
- Flexible, quick to build, and easy to maintain
- Keyword search returned every document containing a word
- Staff still had to open and dig through documents
- Copilot could help, but added $4,000/year
Limitation: Search returned matches, not answers. Speed depended on the person, not the system.
Option 03 · Chosen
GenAI chatbot in Teams
- Contextual query resolution, not keyword search
- Quick, accurate, and trackable for analysis
- Already embedded in Teams, used daily
- No added training cost or licensing cost
Direction: Matched the existing workflow and offered the fastest path to real adoption.
10 · The solution
Chat Point
An AI-powered assistant that helps concierge staff find answers where they already work.
Microsoft Teams
A concierge receives a contextual answer in Teams, can verify it through the linked source, and use thumbs up or down to create an ongoing feedback signal for monitoring answer quality over time.
One source of truth
Chat Point pulls from structured SharePoint content.
Content stays with the experts
Community Engagement Managers maintain the information.
Focused scope
It answers information queries—it isn't a manager, emergency system, directory, calendar, or menu.
Clear boundaries
Defining what the tool doesn't do prevented the solution from becoming an AI system that tries to do everything.
11 · Design decision
We tested two configurations.
We tested both configurations to find the interaction that best matched the real-world need we observed: getting a clear answer quickly, without adding another crowded conversation or system.
Test 01
Group / team chat
Hypothesis: A shared space could support visibility and collective learning.
What happened: Group conversations added noise and made a quick answer harder to isolate.
Test 02
1:1 chat
Hypothesis: A direct conversation would better support the immediate moment of need.
What happened: The exchange stayed focused: one person, one question, one clear answer.
1:1 won.
The 1:1 configuration matched the real-world need we observed: concierge staff needed a fast, focused answer without navigating another crowded conversation or system.
12 · Impact
The research translated into measurable operational impact.
89%
improvement in response time to resident and guest queries
30+
minutes saved per concierge shift
100%
adoption across all three Ingleside communities
$4,000/year saved
by avoiding the Copilot integration cost.
by avoiding the Copilot integration cost.
Accuracy monitored continuously
through staff feedback and satisfaction data rather than a one-time launch metric.
through staff feedback and satisfaction data rather than a one-time launch metric.
5:00
average pre-launch response time
→
0:33
average post-launch response time
Calculation: (300 seconds − 33 seconds) ÷ 300 seconds = 89% faster.
Faster, more consistent answers gave concierge staff more time for personal service—not repeat lookups.
13 · Reflection
In an ideal world, I would have...
These were the highest-value next steps that timeline, access, and implementation constraints did not allow during the five-week engagement.