COLLEAGUE CONCIERGE
Helping colleagues find their next step.
AI SERVICE DESIGN · RESEARCH · BEHAVIOUR
I researched how colleagues navigate unfamiliar workplace tasks and tested how an AI concierge could make guidance easier to act on.
My contribution focused on discovery, comparative response testing and the design principles connecting the assistant to the services behind it.
STATUS: IN BUILD — DISCOVERY AND DESIGN DIRECTION
From finding help to acting on it: the proposed Concierge experience connects discovery, guidance and the next step.
THE ORGANISATION ALREADY HAD A CONCIERGE LAYER.
It was people who knew where to look and whom to ask.
In our internal study, 39 of 52 respondents asked another person for help with an unfamiliar workplace task. Most used several sources along the way.
That made the opportunity more specific: help colleagues navigate the gaps between services, with enough context to understand what to do next.
HELP HAS TO BE FINDABLE.
Before an assistant can help, someone has to discover it, understand what it offers and decide whether it is relevant.
I mapped the route from an initial search through to finding Concierge, evaluating its usefulness and receiving an answer.
Each transition asks something of the colleague: another decision, another unfamiliar interface or another judgement about whether to continue.
The design direction was to make help discoverable within the work already happening, with a clear connection between the question and the task it supports.
The invocation map examines the effort involved in finding, assessing and using Concierge before a task can continue.
AN ANSWER NEEDS CONTEXT.
Finding an answer did not necessarily mean someone felt ready to act.
Internal responses highlighted questions about whether guidance was current, whether it applied to the person’s circumstances and what could happen if they got it wrong.
Those concerns shaped the response patterns I tested. Alongside practical instructions, the richer version included information about applicability, ownership, review dates and routes to help.
The aim was to give colleagues a basis for judging their next step.
Left image: The richer response prototype combined practical guidance with context, source information and a route to support.
TESTING WHAT PEOPLE COULD ACT ON.
I tested the three patterns with 101 external respondents across two workplace scenarios.
The structured alternatives were strongly preferred to the information-heavy version. The richer C response was selected by 63% in the Cardiff scenario and 61% in the temporary-manager scenario. It also generated the highest reported confidence.
The study tested the response as a whole. It did not isolate which additions produced that confidence, or establish whether people would complete a real task correctly.
The resulting design direction was clear: put the action first, make reassurance available and keep a route to help within reach.
THE ANSWER DEPENDS ON THE SERVICE BEHIND IT.
A question such as where to sit in an unfamiliar office can depend on local information, an understandable map and someone who can resolve the gaps.
The Hove map and FAQ material made those dependencies tangible. They helped focus the work on information quality and the smaller connections between services that a broad AI ambition could easily overlook.
RESEARCH INTO DESIGN DIRECTION
52
Usable internal discovery responses.
101
Completed external testing responses.
3
Response patterns compared.
IN BUILD
Discovery informing the developing service.
The research informed use-case consideration, further information-quality work and the proposed response structure. These are research findings and design decisions; live-service impact has not yet been established.
WHAT I'D CARRY FORWARD
I’d start with the moments where people lose their way between services. Those gaps can reveal needs that are easy to miss when discovery begins with the organisation’s most prominent processes.
I’d also separate confidence from correctness. A response can feel reassuring while still being outdated, incomplete or unsuitable for someone’s circumstances. Preference testing helps shape the experience; operational testing needs to establish whether people can act successfully and recognise when to seek help.
The next stage is to test that whole path: finding Concierge, receiving guidance, taking action and recovering when the answer is insufficient.