Overview
The problem
Failure to send proper notice can lead to millions of dollars in damages.
On construction projects, sometimes things can go wrong — events like weather delays and supply chain issues pop up when least expected. Project managers must then inform the appropriate parties about these unexpected situations by sending proper notice.
Unfortunately, many project managers fail to communicate notice well: they either don’t do it at all, or they fail to meet the requirements for communicating the notice properly, such as not following the schedule agreed upon in the construction contract.
These failures can result in legal compliance issues, which in turn lead to millions of dollars of losses for the project manager’s construction company.
the solution
My team decided to solve this problem by building a Procore integration that would take minimal information from a busy project manager, pull in relevant project data from Procore, and then draft a notice letter that project managers could then send to relevant parties.
Outcomes
After design completed, the feature took an additional 3 months to go through engineering and launch. Once available, this feature contributed to an average of $30 million in cost savings for our customers from avoided legal fees. Here are a couple of anonymized quotes from happy clients:
"The Notice Builder feature saves me so much time. I can just fill in the fields quickly — I don't even have to write full sentences when I'm answering the questions, and I get back a really solid draft that I can share out and edit."
"This feature is a game-changer for our teams. Many of our project managers are overwhelmed or don't know where to start with notices. This helps them get over the hump, and they can send over the AI letters to our review teams for approval before sharing with clients."
How we got there
Defining the problem space
I started with some competitive research to see what other products in our market space were doing. It turned out that this feature was fairly unique, as I couldn’t find examples of chat-based document writing within construction tools at the time. Expanding my review to general AI tools, I used Mobbin to review various products, and began to hone in on the idea of the “guided prompt”.
A bird's eye view of the competitive product research I did on a few different tools.
What I call “guided prompting” is the process of gently guiding the user step-by-step to create a context-rich prompt, instead of giving them a blank text box and asking them to write out everything they know. From previous research into the UX of prompting, I knew that creating prompts with the right level of detail is challenging for the average AI user.
On top of that, construction project managers were our key target audience — people who are extremely busy, and see anything that gets in the way of progress on their building as a hassle. This made it important for us to extract information as painlessly as possible from construction PMs and then fill in the blanks ourselves to create a solid notice letter.
Discovery and early ideation
With the research analysis in mind, I built out an object map to see how data types interacted with other data in the system. Following that, I sketched out a few low-fidelity concepts, then had a discussion with my team’s product manager and engineering lead to determine which approach would best suit our user. We decided to continue exploring the Accordions and Focus Mode options, cutting the Everything Scrolls and Multi-page options since the first would feel overwhelming and the second would add too many clicks.
The object map I created in FigJam.
Four concepts for layouts that I explored in low-fidelity. I used an iPad + Apple Pencil with the Figma app to draw these by hand.
Conducting user research
Before progressing further, we validated the concepts with our customers, so that we could make sure we were investing in the right area for development. I conducted research calls to learn about the pain points around sending notices, and to get feedback on two sketch concepts for the new feature. Afterwards, I created an affinity map in Dovetail and extracted key insights that emerged from my thematic analysis.
A broad capture of the thematic analysis I created in Dovetail.
Here are a few of our learnings from my user research:
Project managers don't send proper notices because…
they feel notices get in the way of day-to-day project progress
they're worried that sending a formal notice will harm their relationship with the client
they don't know the specific requirements for sending a formal notice letter
Customers prefer the form-based approach over the chat-based approach because it guides them on what they need to do
Customers want to reduce project team overhead by lowering wait times and speeding up the notice creation process so that builders can get back to building ASAP.
Reducing as much manual data entry as possible is critical to wider adoption.
High-Fidelity and Hand-off
Using the research findings, I was able to quickly take this feature into high-fidelity and iterate with my team. I created a full-flow prototype in Figma to demonstrate the interactions along each step of the journey, then wrote documentation to support engineering hand-off.
A zoomed out shot of the full high-fidelity flow & accompanying documentation for the final hand-off version of Notice Builder.
Since project managers already spent most of their time in a 3rd party tool called Procore, the final design was an integration that we injected into Procore's side panel for external apps. This kept the Notice Builder easily accessible to project managers without forcing them to open a new tab and navigate to a new tool, which might break their workflow.
A screenshot of what the Notice Builder integration looks like in Procore.
Now, project managers can simply fill out a few short questions, then sit back and relax as the AI wrote a notice letter for them. The AI-generated notice letter would build off of the form answers to draft a letter specific to the current situation, and pull in API data from Procore to fill in known information, like the jobsite's address and relevant contact information of the recipient.
Final Thoughts
Reflections
The design on this project took about 2 months — of that, research took roughly a month to complete. Document Crunch has a relatively small user base, and most customers are from the legal or back-office side of construction firms, since that is the persona that the product primarily targets. As such, getting interviews with "true" end users who were project managers or in on-site roles was challenging. A few of the people I talked to weren't project managers at all, but were in roles that managed them instead. Something that would help with future research projects is developing a persona-segmented database of end users that are open to providing feedback to the design team.