Conveyor AI

 

A single engineer working on a proof-of-concept over 8-weeks costs $30,000. Conveyor AI is a platform featuring a drag-and-drop UI with a pre-built library of AI components that allows anyone to create functional AI models in minutes with absolutely no code.

NOTE: Since this project involves features of a product that was unreleased at the time, certain artifacts cannot be shared.

The Problem

 

My Task: The 4-week Incubator Challenge is a part of a designer bootcamp created by IBM. During this challenge, teams of 5 designers work together to design a solution for whatever startup the group is assigned too.

Problem Statement: How can we help new users understand the value proposition of Conveyor AI, their ROI (Return On Investment) for their use case, and how to make the most of the product within minutes of starting?

Research Insights:

  • For this project, we were responsible for doing our own user research, all of which is described below.

Additional Context:

  • As Conveyor AI is a startup, they didn’t have a lot of manpower to support us with. Since the product had not been released yet, we didn’t have any current users to work with.

Scoping the Problem

 

My team and I felt that the initial Problem Statement was rather broad. We employed Enterprise Design Thinking to further specify and scope the Problem Statement to be something that could be achieved in the brief 4 weeks that we had together. After a full week of speaking with stakeholders to understand the product (after all, how can we teach our users how to use a product we barely understand ourselves), we arrived at our two Hills:

  1. It should take no more than 15 minutes for a new user with basic development knowledge to deploy a simple AI app with Conveyor AI.

  2. A low-code user should be able to use documentation to learn a core Conveyor AI feature or successfully recover from an error with in
    4 clicks and/or 5 minutes.

The Solution

 

This was a project that I was particularly excited about as it was right in my ballpark: teaching people is my specialty. Getting a chance to design the learning experience for a brand new product was a great way for me to employ my pedagogy in industry design. In response to the Problem Statement, we produced two deliverables:

  1. Template Finder

  2. Documentation

The Research

 

To better understand the problem space, we had to conduct some initial research before generating our solutions. Here’s what we did:

  1. Interview stakeholders

  2. Conduct usability testing

  3. Complete a heuristic evaluation

  4. Run competitive analysis on learning/onboarding experiences

Research Step 1: Interview stakeholders

We interviewed 5 potential users with a pre-written script for consistency

 

The questions we wanted to answer were:

  1. What are our user’s goals and pain points when using low code software?

  2. What types of resources do users find most helpful when learning new software tools?

  3. Which use cases benefit from low-code AI software?

We interviewed the type of personas we wanted to target for our solution

Research Step 2: Usability testing

With each of our interviewees we also ran usability testing with an existing prototype.

 

The questions we wanted to answer were:

  1. What’s confusing and requires additional tooltip support?

  2. What parts of the experience do users find most confusing and frustrating?

  3. How do users navigate our documentation hierarchy?

Research Step 3: Heuristic evaluation

Every member of our team ran an individual heuristic evaluation for the existing prototype using Jakob Nielsen’s 10 Usability Heuristics

 

The questions we wanted to answer were:

  1. Are there any usability issues with the current prototype?

    • Going from Projects to Workflows doesn't make sense to me when it was explained to us that Projects are use to group multiple workflows. Shouldn't workflows be created first, and then grouped?

    • When I think of a "Project" in other software such as Adobe, I think of the elements of a Projects as things that speak to each other to create one cohesive whole. Do the workflows of the same Project interact at all?

    • what is an input? What is an output? Not sure if these are industry-specific terms or Conveyor AI terminology

    • I clicked on Audio File Upload component and it asked for an “audio data type”. How do I know what qualifies as an audio datatype?

    • I don't know what the component can "output" until I click on it and put it on the canvas. Would be more streamline to know what the inputs and outputs are beforehand.

    • The titles of the components are not clear on what exactly they do. For example, there is a component called Subscription.

    • The component library isn’t organized in any particular order. Would be helpful to the user if we created some kind of hierarchy to organize the components so that they can better understand which components suit their use case.

    • there is one button labeled “Configure” but it is not clear what is exactly being configured. “configure” is also a rather vague term.

    • Some of the text in the design uses title case and some uses sentence case. Would be good to pick one and stick with it.

    • When hooking up the components with each other, there’s no clear indication whether I’m hooking them up in the “right” way. Could add some kind of disabled state to clarify which components can be hooked up with which

    • Is there a standard “step 1, step 2, step 3…” etc to building these models? If so, we can provide the user some guidance.

    • Disable certain inputs and outputs based on the component the user is currently clicking on to prevent them from hooking up the model incorrectly.

    • come up with a way to enable people who aren’t able to utilize drag-and-drop to make the product more accessible

    • Design should try to reduce clutter as much as possibly since canvas-type applications tend to be prone to that.

Research Step 4: Competitive analysis specifically on the Learning experience

This was a step that undertook myself as someone who specializes in learning experiences.

 

The questions we wanted to answer were:

  1. How do other low-code software applications help their users learn their product?

  2. Given our short time frame, what can we as a small group achieve that will be the most effective and helpful to Conveyor AI in the long run?

Synthesizing our Research

 

After our research was completed, we still had two weeks to form together our solution. We devised a two part solution:

  1. Documentation - Although this deliverable wasn’t something that excited us, it was something that the CEO strongly pushed for.

  2. Template finder - This was a deliverable that we as a team felt would be more engaging to users and that would better leverage our skills as a team of designers, and also gave Conveyor AI opportunities to grow in the community space in the form of sharing templates. However the CEO of Conveyor AI didn’t feel as strongly about this experience as he did about creating documentation.

My contribution to the solution

 

Using the results of my competitive analysis in combination with our extensive user research I spear-headed the creation of our documentation with these steps:

  1. Identified the optimal information hierarchy

    • By walking through the final prototype of the product, I identified the optimal information hierarchy.

  2. Conducted research

    • As aforementioned, investigated how similar products are documenting their complex libraries and usages.

  3. Introduced Gitbook

    • Through research, I discovered this well-established documentation platform and advocated for its use in our project.

  4. Initiated the Gitbook

    • I created the Gitbook account, sorted out billing logistics, created a template for every page so that my team + Conveyor AI can continue to fill out the documentation consistently even after our 4 weeks working with them have concluded

  5. Populated the pages

    • Using the template I created, my team and I began writing the Gitbook pages with our best understanding of Conveyor AI. This was a challenge as we weren’t SMEs nor content designers!

  6. Wrote the handoff guide

    • I created a Box Note of general rules to follow so that the Conveyor AI team can easily follow the flow of our work and continue off of it.

What I learned through this experience

 

These are my takeaways from this experience that I’ll carry with me for the rest of my career as a designer

  1. It’s ok to not fully understand the product

    • Taking the time to learn the product well instead of rushing into the design aspect of the project can lead to a lot of undoing and time wasted in the future. My team and I were a little panicked when halfway through our four weeks we were only just done with research. However as we designed the product we realized that our research made our design process more efficient.

  2. As designers, it’s ok to have a different stance from other stakeholders and still assert your point of view

    • In spite of the fact that the CEO wanted to drop Template Finder and have us just focus on documentation, we persevered with Template Finder while also delivering documentation in the form of an easy-to-use third party platform widely trusted by similar products. We truly believed that Template Finder was something that future users would love, and as designers, we felt that it was right to assert the voice of the user.

  3. A REALLY good team can make a world of a difference

    • We didn’t operate as a strong team simply because we were all nice to each other. We stepped up for each other when one person was sick/ felt that they needed assistance. We weren’t afraid to offer a counterpoint and engage in difficult discussions. Although the problem space was daunting, our teamwork made me enjoy working with them everyday.

    • We still have monthly calls to catch up! Even though two of us aren’t even at IBM anymore.