Nexmine

About the project

Company

McKinsey & Company
August 2023 - September 2023

Contribution

Design Specialist

Team

Engagement Manager
Data Scientists
Front & Backend Dev Resources
McKinsey Leadership

The challenge
A major generative AI use case that McKinsey had been exploring with various clients is a smart maintenance advisor that is informed by client data. The tool would be a copilot for all maintenance workers and technicians on a mining site whose job it is to successfully maintain pieces of equipment.

How do we make leverage available data to inform a conversational AI chat interface that addresses real pain points of a maintenance worker? How can this tool make someone's day at work easier in a way that traditional analytics can’t?

McKinsey had some of the technical aspects ready from tools they had built for previous clients. Our job was to figure out how this tool might be used, what users may ask it, and making sure we have the right data to support those answers along with a streamlined UX that would wow a board of directors.

Our initial requirements included the following:
‍

Help diagnose a problem
Help outline a maintenance procedure
Identify and order necessary tools and parts
Identify potential safety hazards
Facilitate communication with shift supervisors
Help update a maintenance log and reports
potential safety hazards of a maintenance issue

We drafted an initial storyline to illustrate the original vision we discussed with the client in early meetings.

My roles

I led all design efforts on the project including research, facilitating ideation, journey mapping, design and prototyping.

Project impact

Our proof of concept was presented to the board of directors and it was unanimously approved to be built. The AI tool went into development in Q4 of 2023.

Research Overview

Personas

Our client identified three different types of users. We spent the majority of our time focusing on Greg, the maintenance worker, who would be the primary user of this product.

We spent the majority of our time focusing on Greg, the maintenance worker, who would be the primary user of the product.

Research Methodology

We aligned on a qualitative approach to uncover the voice of the user that would inform a current state journey along with identified pain points and potential future state concepts.

After some internal ideation and the creation of a future state journey map, our second round of research would consist of validation and refinement of the journey map.

Round 1

9 Qualitative interviews with maintenance workers

Round 2

6 Qualitative interviews with maintenance workers from the same group, focused on journey map validation

Questions we addressed

What are you doing during your workday?
What are your major pain points?
What technology are you currently using to do your job?
What operational and logistical tasks might be helped by a tool like this? How?
If you had a magic wand, what would you create to make your job easier?

Analysis

Some themes emerged during our research and we identified four opportunity areas worth exploring:
‍

Routine maintenance
Is a specific function of a piece of equipment working appropriately?
General issue diagnosis
The equipment is broken, what is the appropriate course of action?
Visual diagnosis
The equipment is broken and something looks out of the ordinary, what am I looking at?
Repair logistics
I need parts and tools to fix this equipment, how do I get those and how do I document everything?

Pain Point Themes

Tests are run to determine if equipment is indeed broken, this can be time consuming and not always effective.
Even though manuals and work history is digitized, looking for the right information can be arduous and time consuming.
It’s common to not know which maintenance bays are open and available, it takes calling around to figure it out. Sometimes unclear where to send the broken equipment and repair gets delayed.
The process of ordering parts is very manual, we either go personally or send someone to go in person
It’s easy to forget to update the fleet management system which causes problems downstream
No central knowledge base, if someone leaves the company process is lost

‍

Journey Mapping & Ideation

I put together a fidelity current state journey map in Miro based off of research insights that we would eventually use as ideation session stimulus.

I facilitated a ideation session with our internal team that consisted of one engagement manager, three data scientists and two developers.

Based off insights from research and our internal ideation, I created a future state journey map that tells the story of a typical scenario at a mine during which our AI assistant would be used. I connected each of the activities to one of the themes that came out of our research.

The top portion of the journey map has five swim lanes that represent the location of each touchpoint along the timeline.
‍

The day starts at the maintenance bay and bounces between the operations center, the app that we are building, the mine pit and the warehouse that stores the needed tools and parts.
‍

In the bottom two swim lanes, we identified potential future state concepts as well as each pain point we we would address.

We validated this journey map with an internal McKinsey & Company expert who has a wealth of knowledge and experience in mining to help us create a more realistic scenario. He gave us suggestions about the order of touchpoints and how we should accurately describe them. As an example, he told us that tow trucks are referred to as floats.
‍

Our second round of user research was solely focused on the validation of this artifact.

Design

Our research showed that the virtual assistant tool would be mostly be used on tablet and mobile by maintenance workers and supervisors, desktop by operations personnel.

Core Features

Task-centered dashboard
Spanish language support
Documents section
Admin settings
AI chat interface leveraging the ChatGPT API
Suggested prompts
Chat history
Ability to export chat manuscripts
Image recognition
In-line safety guidelines
In line work orders

We built a functional prototype in Figma, the goal of the prototype was mostly to illustrate some of the features that the client had originally indicated that they wanted in this tool and also to show some features that we had added based off of insights from user research.

This tool was eventually coded so that we could use dummy data and that coded version served as a proof of concept to eventually sell the board of directors on getting this thing built. We presented it to the board, they approved it and the tool went into development in December of 2023.