W.i.P / Forge

Like all good stories,

there is absolutely no world building and we’re going to jump right in.

The Solution:

Okay, but that is?

That’s Forge*, which leverages Gen AI's capability for dynamic content generation to create personalized online experiences.

*Note: Due to restrictions showing this work, certain elements like the brand identity and other discernible features have been changed, replaced with visual content created just for this study.

Like all good stories,

there is absolutely no world building and we’re going to jump right in.

The Solution:

Um... That means what?

Oh, Gotcha, let me switch to regular people speak. Forge is a Generative AI Agent (fancy chatbot) builder and Digital Campaign (fancy online experience) management tool. With near endless possibilities, this ready-to-go and scalable toolset works hard to help you achieve your end goals. And, like all good prequels, we’ll start with why and how Forge came to be, and where it may be going.

Like all good stories,

there is absolutely no world building and we’re going to jump right in.

The Solution:

That’s Forge*, which leverages Gen AI's capability for dynamic content generation to create personalized online experiences.

*Note: Due to restrictions showing this work, certain elements like the brand identity and other discernible features have been changed, replaced with visual content created just for this study.

If you're viewing this study on a small screen, web will present a more interactive experience.

Okay, but that is?

That’s Forge*, which leverages Gen AI's capability for dynamic content generation to create personalized online experiences to pair with an agentic ecosystem.

*Note: Due to restrictions showing this work, certain elements like the brand identity and other discernible features have been changed, replaced with visual content created just for this study.

If you're viewing this study on a small screen, web will present a more interactive experience.

Um... That means what?

Forge is your customer’s advocate.

Forge operates like a white label you get to customize to your business and your campaigns, like, let's say, a web page for your newest collection of streetwear. Within the Forge platform, you're defining the guardrails, building the content and shaping the feel of that specific campaign. Whether customers are concerned about allergens in textiles, want to see themselves in the shirt, or are ready to check out, Forge is there as an AI advocate energizing your business and collecting valuable insights.

Forge provides actionable, properly coded insights that can help you adjust what kind of pre-fabricated or generative content should be included in a campaign. Forge will integrate within your existing ecosystem as an AI advocate, acting as an overlay if that's all you need. The same security and reliability is to be expected whether you're building entire campaigns or just deploying advocates across any channel. This remains true regardless of whether you fully utilize all the content creation and management functions.

Being consistently affirmed in the positive by research and advisory companies, we ensure that nothing - whether new features or tweaks to existing structures - is done haphazardly. For the purposes of this study, my sanity, and in respect to the average human lifespan, we're going to streamline this and focus on the relationship between Forge and the Small Business (SMB) Segment.

1

Forge is your customer’s advocate.

2

Ideation Station.

Since Forge is an agent and content management interface, there are only so many ways to design the UI, but far more ways to define the interactions within that interface. So, with Market & User Research in hand...

$4B+ addressable market for SMB

82% would pay premium for AI-powered features

91% want real-time content generation abilities

Existing solutions fall into fragmented categories, with no proper unified platform or thread to easily connect the pieces. Knowing there is a healthy competitive market, an agnostic or unified approach will obviously be a key point for success. SMB users want one program to rule them all and do everything. Figuring out the right ratio of in-house, bespoke control and third-party support is a Herculean task that will consume a lot of time.

Finding out the personas, or distilling down the right segment research, will be one way (of many) we start to refine and define Forge, and the platform that supports it. SMB is a huge category of human beings, so who are we selling this multi-channel, integration-friendly AI campaign creation and agent support product to?

$4B+ addressable market for SMB

82% would pay premium for AI-powered features

91% want real-time content generation abilities

Has a moderate monthly budget

Need for reduced costs and increased growth

Wants automated content to reduce mental load

Collecting thoughts, that’s hard, so we use products like Notion to help.

Information architecture, user flows, journeys and all that other fun stuff is also hard, so we use Miro to make it a bit easier.

Ideation Station.

Has a moderate monthly budget

Need for reduced costs and increased growth

Wants automated content to reduce mental load

Collecting thoughts, that’s hard, so we use products like Notion to help.

This may be a new product, but there is a lot of pre-existing grounding. Users want a simple, no-code solution to their complex marketing funnel and
e-commerce concerns. These same users are divided between wanting something to plug-in to their existing kit, and that of an all-in-one site and agent builder with Conversational and Generative AI features. So, figuring out the peculiarities of how the user and their intended audience will interact on both the front and back end is far more complicated than shown in the chart above. So, how are we going to design for both the known and the uncertainties?

This may be a new product, but there is a lot of pre-existing grounding. Users want a simple, no-code solution to their complex marketing funnel and e-commerce concerns. These same users are divided between wanting something to plug-in to their existing kit, and that of an all-in-one site and agent builder with Conversational and Generative AI features. So, figuring out the peculiarities of how the user and their intended audience will interact on both the front and back end is far more complicated than shown in the chart above. So, how are we going to design for both the known and the uncertainties?

Information architecture, user flows, journeys and all that other fun stuff is also hard, so we use Miro to make it a bit easier.

Forging ahead.

3

Forging ahead.

Generating and/or populating content on a site, or within a chat window, are two different asks. The first needs to be developed; the latter is an upgrade to structures in place. Even so, it's important to start from the beginning to see if we can better shape that evolution. Nobody wants a sequel with the same story as the first. Well, most don't

We’re building our wireframes, prototypes and design systems in Figma, go figure, totally wild.

There is another reason to focus on the AI Advocate. Forge can be truncated by delaying the Dynamic Content Assembly Platform (DCAP), the site generator, but the opposite is not true of the Generative AI Agent Builder. Given how well Jakob's Law is often validated, the general structure of each rapid prototype (low to high fidelity) shouldn't be a shock. The flourishes are in the functions, interactions, and visual style, and not the structural build. X needs to go to Y in the expected way, so the challenge is balancing our need for creativity with the biases of user expectations. They need to be visually appealing, and easy to debug. And, well, not every egg gets to the omelette.

Well, sometimes we do use ProtoPie, we’re not Figma exclusive (but that is Figma).

The key part of making something functional is making sure it's functional. So, the approach to design will naturally deviate from start to finish. The AI Advocate lives on the web, and the internet is not a cohesive place, to say the least. At every actionable stage of design, we are testing the validity of our ideas (as I'm sure we all 100 percent do, every time, right?).

Ensuring there is functionality, usability, and visual consistency across different browsers and devices is one of the most important and repeated stages of internal testing. Even the first solution at the top, well... it's changed a bit. How things manifest into existence in our mind and on a website doesn't always align. Even the example below will change.

We’re all seeking validation, and plug-ins like Storybook helped in that.

We’re building our wireframes, prototypes and design systems in Figma, go figure, totally wild.

Ensuring there is functionality, usability, and visual consistency across different browsers and devices is one of the most important and repeated stages of internal testing. Even the first solution at the top, well... it's changed a bit. How things manifest into existence in our mind and on a website doesn't always align. Even the example below will change.

Well, sometimes we do use ProtoPie, we’re not Figma exclusive (but that is Figma).

We’re all seeking validation, and plug-ins like Storybook helped in that.

4

Pushback & Tact.

The study focus has been on the consumer-facing experience, but this is heavily influenced by where advocates are deployed, how they are deployed, and what AI models are being used, to name a few known unknowns. These are a lot of things to consider, so if interactions don't make sense or the UI '’looks like a failed exercise in modernism,” to quote someone who won't be named (Stephen), that's bad Gen AI design. Proper visuals and controls for user input, advocate output, and error management are key.

As Expected

Easier than Expected

Harder than Expected

Intuitiveness

25%

59%

16%

A lot of factors can lead to task abandonment. Having an intuitive interface is one hedge against that, along with a consistent goal. The first few seconds of experience are crucial, but every other second after that is important as well. It takes so little time for a user to give up on an AI agent experience. Tasks need to be simple, error recovery clear, and navigation fluid.

Customization from the builder is something that will also impact the experience of users. Like a video game, a developer can't really control how modders will play with their creation kit. They can, however, make sure the builder is easy, the components work well, there is clear guidance, and that some checks and balances exist. And like any good game, there needs to be a balance between giving away too much or too little, so there is a market for robust post-release support, adoption, and relevant add-ons. No one wants a game that only felt whole after buying the DLC.

We've been talking about the AI advocate's experience and appearance, but we're not just selling bespoke chatbot skins. Those agents are not only being visually constructed, but their interactions are being constructed as well. The performance and accuracy of those interactions need to be trustworthy and fluid, no matter how you scale or integrate, on whatever channel, with whichever model. That is a vital marketable point that needs to work with both the agent and DCAP.

While gathering insight into the product in dedicated interview panels, we're also fine-tuning our personas and looking for insights from those people that might lead to some amazing advertising inspiration.

For our user research, one of the platforms we use is Maze.

Pushback & Tact.

While gathering insight into the product in dedicated interview panels, we're also fine-tuning our personas and looking for insights from those people that might lead to some amazing advertising inspiration.

For our user research, one of the platforms we use is Maze.

5

Market Madness.

You probably saw this coming, but the DCAP doesn't hit the same high standard as the Gen AI Advocate Builder. Well, it works, but it doesn't work well all the time. In A/B testing of a partly generated marketing funnel page versus that of a designer-created marketing funnel page, due to generation errors and other issues like frustration clicks, there was not enough positive consistency for DCAP compared to the latter. (mock simulation below)

The reliability of the full package needs to exist to avoid truncating Forge. While Forge is presented as one cohesive product, these have always been two silos with a lot of cross-collaboration and a desire to cross the finish line in one package. However, nothing is being scrapped. DCAP will rise again, not just in this release.

Immediate marketing direction shifted to support a less robust platform and product update. I know, anticlimactic right? So, while the sequel is being written, let's wrap up this current production. The truncation didn't dramatically alter how the AI advocate builder and customization progressed, but did allow extra time for user testing and brand voice finessing, which we took advantage of to help reduce our biases.

When ideating copy, we sometimes like to use products like Claude to assist.

Immediate marketing direction shifted to support a less robust platform and product update. I know, anticlimactic right? So, while the sequel is being written, let's wrap up this current production. The truncation didn't dramatically alter how the AI advocate builder and customization progressed, but did allow extra time for user testing and brand voice finessing, which we took advantage of to help reduce our biases.

Market Madness.

When ideating copy, we sometimes like to use products like Claude to assist.

6

Now we’re in the End Game

So, did all these decisions pay off? Personally, I always feel there is room for improvement, and perhaps that is a great trait to have, despite being a bit of a Negative Ned. Nonetheless, there were quite a few actionable insights gleaned from the process and results, and I'll share some SMB related ones below without much commentary.

20-32% reduction in cart abandonment with the use
of AI Advocates.

AI Advocates provide real
time analytics & insights, allowing for up to 4x ROI in applied campaigns.

Gen AI personalization can double CLV when features like multilingual support and tailored content are deployed.

83% reduction in response time, with some common issues being resolved in seconds.

This case study was constructed to show thought process, so it made sense to structure it like a regular A-to-Z narrative. While there is healthy debate over case study structure, I appreciate whatever works best for the project. Other studies published in W.i.P may not follow this structure and will likely be more condensed in length.

Now we’re in the End Game