Guide to designing a GenAI product: From vision to content strategy

Working with Gen requires designers to shift their mental models from deterministic to probabilistic output. Not only are you working with a new material, the technology is so new so there aren't any best practices (yet).

This guide is an overview of the technology and lessons I've learned in my own AI consulting projects working at PH1 Research and from the amazing experts we've had as guests on the Design of AI podcast (Spotify - Apple).

Over the coming weeks I'll publish additional articles to dive into key areas that this guide can't answer in detail. Please subscribe to my LinkedIn newsletter to get updated when those topic-specific articles are released.

Sections in this guide

1. Background & reality-check

2. Rationale for AI

3. AI product vision

4. AI product strategy

5. AI product principles

6. Design's role in crafting GenAI products

7. Content strategy

Background & reality-check

We're two years into the latest AI hype cycle and I encourage you to be pragmatic. Not every business needs to deploy an AI product and not every use case would benefit from GenAI capabilities.

Past technology hype cycles have taught us that a rush to capitalize on emerging internet technologies leads to the implosion of once-promising businesses and products. The launches of Siri, Alexa and other chatbots/assistants fueled a 2010s hype cycle that led to a comedic list of failures featured in Verge's list of biggest flops of the decade:

Companies saw the hype and misunderstanding that surrounded artificial intelligence and thought to themselves: “A-ha, we can sell that.” They produced AI toothbrushes, AI smart beds, AI alarm clocks and dishwashers, promising that “advanced machine learning algorithms” would adapt to the problems in our life, while cranking out the same old products relying on IF/OR functions.

This all changed when ChatGPT proved the power of Large Language Models (LLM). LLMs are trained on large data sets and capable of understanding and generating natural language and other types of content to perform a wide range of tasks. LLMs are a specialized application of machine learning and natural language processing that has deep learning capability.

By leveraging neural networks to teach the data models and transformers for training models in less time, OpenAI, Anthropic, Google, Meta, Mistral, and others have unlocked use cases that are business-ready.

The technology is readily-accessible, tokens are increasingly cheaper, and the context-awareness is shockingly impressive. AI chatbots are improving at an even faster rate than computer chips, meaning that we should expect more impressive new capabilities soon.

Moreover, as our guest Dr. Amy Bucher, Chief Behavioral Officer at Lirio mentioned, academic researchers are studying many new capabilities that are yet to be commercially available.

Obviously it is hard to know if this time the AI hype will be outweighed by true value generation. But, we know that AI needs enormous computing power to succeed and that Nvidia and competitors are well-capitalized to deliver on that.

the indicators are that the combination of new computing power available from Nvidia and others, LLMs rapid improvement, plus yet-to-be explored academic research avenues point to a bright future.

Action items for designers

👉🏽 Understand the technology so you can take on more technical UX problems

👉🏽 Pressure test the technology so you're aware of what it can or can't do well

Rationale for AI

So far in 2024, investors have pumped $26.8 billion into 498 generative AI deals. That's on pace to be larger than the GDP of most of the countries in the world. The hype is detached from reality and it's expected that most AI startups will fail.

But buried in the hype is increasing adoption and excitement. According to a survey by McKinsey, AI adoption has more than tripled since 2017 and adoption of GenAI doubled between 2023 and 2024.

McKinsey: The state of AI in early 2024

And this growth in adoption is directly related to a growing list of potential use cases. IBM's Armand Ruiz created a comprehensive resource of potential use cases across almost every industry. However, according to Gartner, estimating and demonstrating business value is the number one AI adoption barrier.

Based on the conversations we've been having across industries, the reality is that every business is experimenting with GenAI, perhaps not publicly. Fear is growing too. Wait and startups may leverage GenAI to chip away at your competitive moat. And worse, some businesses feel that OpenAI may completely commoditize many use cases to such a point that they could no longer be monetized.

From my perspective, the rationale for leveraging GenAI is undeniable for organizations where a significant amount of their business is dependant upon any of the following:

• Improving the quality of document processing

• Increasing the scale of customer service and self-service tools

• Maintaining and automating code completion and bug detection

• Delivering tools to make data more comprehensible and actionable

• Turning data and archives into actionable workflows

• Enhancing access and utility of creative tools

• Observability and monitoring of internal or external data sets

• Managing and deploying personalized content at scale

As such, you need to ask yourself if GenAI is what you really need before you proceed.

Action items for designers

👉🏽 Self-critically evaluate if your business would benefit from GenAI and if so, build a case

👉🏽 Build service blueprints to understand, assess, and model if and how GenAI can benefit

AI product vision

While 90% of VC-backed companies plan to launch generative AI in their products, many of those will fail because of technical and/or strategic issues. Teams are burning a lot of effort and R&D budget to explore potential ways of leveraging AI. They must find a way to prove value and monetize the technology.

Rather than taking the "slap AI on it" approach, you need an AI project vision to ensure that each AI initiative ladders up into your business architecture.

Here are two guides to help you craft a product vision: Mural & ProductPlan. But given the uniqueness of GenAI, your AI product vision must achieve the following:

• Enhance the business mission

  : This product must deepen the customer's relationship with the business, not add a tangential benefit

• Enhance the core product/platform vision

  : This tool sits within a suite of tools or products —ensure that it enhances, not distracts

• Focus on aspirational

  : As much as product is often defined by "problem to be solved," customers pay for solutions and states they want to achieve

• Enable exploration

  : Building with GenAI is exploratory, so the vision cannot lock the team into a solution by being too specific

An example of this is Microsoft 365 Copilot:

Harnessing the power of AI, Microsoft 365 Copilot turns your words into the most powerful productivity tool on the planet

The AI product sits as a layer on top of other products within the platform. The vision is that Copilot unifies and simplifies using the entire 365 platform. It amplifies what a user is capable of by interpreting a user's actions by adding context from the Microsoft Graph and LLM. This enables the Copilot to understand the context of what you're trying to achieve and guide you to solution(s) for a range of tasks.

Overall, Microsoft's vision enhanced the business and platform. It also opens continued opportunities to explore the technology's capabilities and new paths to deliver value.

Action items for designers

👉🏽 Your product shouldn't be an add-on —it should be an enabler within the ecosystem

👉🏽 Map how the product facilitates core JTBD across that ecosystem

AI product strategy

We can't all be Microsoft, Google, Meta sitting on mountains of data and funding; They can try every strategy until they find one that works.

The problem is that most of us have been trained to believe that a plan is a strategy.

Roger Martin is one of the most prominent strategists. Here he shares why this confusion exists.

https://youtu.be/iuYlGRnC7J8

Creating a strategy is intimidating. Alex M H Smith is the king of unpacking the process of making high impact strategies. He wrote No Bullsh*t Strategy: A Founder’s Guide to Gaining Competitive Advantage with a Strategy That Actually Works to help and the lessons apply to AI because he challenges us to avoid common mistakes by focusing on what makes what you're selling unique and studying competitors.

And here's the hardest thing for leadership to accept: Being built with AI doesn't make it unique or competitively advantageous. Harvard Business Review analysis shows that B2B customers don't care about AI —they buy because of the value that a product/service delivers them.

Source: Harvard Business Review "The B2B Elements of Value"

And my own customer research showed this as well: The term AI is too unspecific for customers to append specific value to. To the dozens of potential buyers that I interviewed, "AI" is equivalent to a sticker that says "New."

The secret to your AI product strategy is embedded in understanding value drivers from the perspective of your customers. Use these questions to triangulate those value drivers:

1. Walk me through how you use the product, why you use it, and the decisions our product helps you inform?

2. What were the reasons why you've considered switching to a competitor's product?

3. If you lost access to our product for an entire week, what would be the most challenging consequences of that?

4. What single workflow do you wish our product could automate for you?

5. Imagine a new product coming to market that completely eliminates the need to use our product/service —what makes that new product so unique and important?

Customers won't write your strategy for you but they will describe capabilities that they're willing to pay much more for. It's your job to determine how to best leverage this novel technology to unlock 10x value.

Action items for designers

👉🏽 Conduct in-depth customer interviews to triangulate the most importance value drivers

👉🏽 Develop a strategy that leverages AI to deliver 10x or more new value to customers

AI product principles

Today's AI products are plagued by issues that must be reconciled in your product strategy.

Ethical issues include the cost to innovation, bias and discrimination, lack of trust, security problems, lack of informed consent, and many more.

Leveraging GenAI comes at a cost to your customers and your business. A poorly-executed project can harm your balance sheet, brand, and build resentment within teams.

AI-mature organizations navigate this is by developing product principles (aka pillars) that define the responsible use of AI.

Google AI's responsible AI principles are some of the most comprehensive.

Objectives for AI applications 1. Be socially beneficial. 2. Avoid creating or reinforcing unfair bias. 3. Be built and tested for safety. 4. Be accountable to people. 5. Incorporate privacy design principles. 6. Uphold high standards of scientific excellence. 7. Be made available for uses that accord with these principles. AI applications we will not pursue In addition to the above objectives, we will not design or deploy AI in the following application areas: 1. Technologies that cause or are likely to cause overall harm. Where there is a material risk of harm, we will proceed only where we believe that the benefits substantially outweigh the risks, and will incorporate appropriate safety constraints. 2. Weapons or other technologies whose principal purpose or implementation is to cause or directly facilitate injury to people. 3. Technologies that gather or use information for surveillance violating internationally accepted norms. 4. Technologies whose purpose contravenes widely accepted principles of international law and human rights. As our experience in this space deepens, this list may evolve.

Others, like Anthropic created a constitution to introduce a worldview which was differentiated from OpenAI, where their founders defected from.

As the honeymoon period of GenAI nears its natural conclusion, designers should expect customers to demand higher standards. We're already seeing this happen in the creative industries:

• Adobe was forced to scale back plans to train AI using artists' work

• Founder of Procreate, an illustration app, declared he @$*#! hated generative AI, and his company’s products wouldn’t adopt it

• Generative music startup in a locked in a legal battle with the music industry:

  Suno's view is that incumbents want to stifle competition

  , whereas the

  industry sees generative music as stealing an artist's life's work

These issues will eventually spill into other industries and you need to be prepared.

Product principles are your opportunity to explicitly align your corporate mission, product vision, and product strategy in a way that deepens your relationship with customers.

Action items for designers

👉🏽 Determine the non-negotiables of leadership, the product team, and your customers

👉🏽 Build principles that are specific, yet flexible enough to keep your team focused on delivering positive impact to your wide range of stakeholders

Design's role in crafting GenAI products

Yes, that was a lot of content to read before finally get to the first section about design.

It goes to show the immense weight and power that designers have in creating AI-powered products. Rather than fear AI, this is a time to embrace new potential streams for our own careers: Conversational designer, behavioural designer, service designer, model designer, and GenAI is remaking the core UI/UX role as more adaptive capabilities are added.

Designers are essential in this moment of great change because they're the best equipped the navigate uncertainty and to fine-tune outputs to deliver value.

A fantastic resource is Google's People + AI Guidebook, offering best practices and examples for designing with AI.

Emily Campbell was one of our first guests on the Design of AI. She has been at the forefront of defining the shape of AI. Visit her website to learn about key UX patterns:

• Wayfinders: Give users clues about how to interact with the model, particularly when getting started

• Inputs: Submit the user's prompt to the AI within its surrounding context

• Tuners: Let users refine or remix their prompt to get improved results

• Governors: Maintain user agency as the AI works in order to understand and direct the AI's logic

• Trust indicators: Give users confidence that the AI's results are ethical, accurate, and trustworthy

• Dark matter: Potentially nefarious, but certainly ambiguous patterns that impact user trust with questionable user value

• Identifiers: Differentiate the AI from other features and highlight its use case

GenAI creates many new UX challenges/opportunities that designers must explore solutions to. In this video Emily Campbell explains how Github Copilot has an assistive feature to predict user intent.

As we move into an era that will be increasingly dominated by AI-powered solutions, designers must learn how to craft with their new superpowers:

• Situational awareness

  : Leverage situational data that enables the system to render the user's intent (e.g. Predict what you're trying to create based on behaviours & environmental context)

• Task assistance

  : Leverage the information the model was trained on to offer assistance on specific tasks (e.g. Drafting content, creating images, modifying code, analyzing data)

• Task automation

  : AI agents that are trained to complete specific workflows once instructions are provided (e.g. Build and update dashboards, manage and document meetings, build an application)

• Personalized journeys

  : Monitor the user's progress and modify the journey to improve their likelihood of advancing forward (e.g. Coaching plan to help someone quit smoking by monitoring days smoke-free and their positive/negative steps)

Action items for designers

👉🏽 Shift from a task completion mindset (i.e. focused on conversion) to one where you're focused on the user's success (i.e. How might we enable them to have the best experience)

👉🏽 Use journey maps to explore the current state of an experience and explore potential improvements, then reverse engineer where and how to leverage GenAI

Content strategy

ChatGPT transformed the paradigm of user interfaces by delivering a conversational experience that adapted and learned from us. Now that conversational interfaces are the default for interacting with GenAI, books like Erika Hall's Conversational Design become critical to every designer's toolkit.

A designer's goal should be to reduce the unnecessary friction of prompt engineering. Interfaces and user journeys can be designed to infer the situational context necessary to take an OK prompt to a BEST prompt.

Designers also need to train users how to converse with a technology that they don't necessarily understand, trust, or need. When I've been researching and testing new AI products, the same concerns are raised by users:

• What does this "AI" capability actually do?

• How do I know if it is accurate?

• Why should I stop using the interface I'm comfortable with?

Interfaces must aim to mitigate these concerns to ensure that the user maintains their focus on a successful outcome.

One of GenAI's capabilities that is most exciting is the ability to create adaptive user experiences. A recent study explored how ChatGPT performed when it was asked to create personas based on survey data and to create web pages based on those personas. The results were promising, showing comparable outcomes but with big time savings.

The technology isn't ready yet to replace this highly-nuanced workflow but it there is growing evidence that generative AI can be used successfully to personalize user journeys in ways which are too effortful to do manually. Writer is one such platform, enabling brands to fine-tune models to generate content.

Action items for designers

👉🏽 Consider your expanded role in directly and indirectly supporting users through the process of rendering their intent into successful outcomes

👉🏽 Consider the possibilities of personalized user journeys that will soon be possible thanks to generative AI

Conclusion

GenAI introduces a range of complexities and risk that need to be thoroughly understood. But the technology also expands the designer's toolkit in ways which we could never have imagined. Regardless, design is about to go through a major period of change.

I hope this guide helps each of you find the path and specialization that best suits you.

Over the coming weeks I will publish deep-dive articles into the topics discussed in this guide. Please subscribe to my LinkedIn newsletter to get updated when those topic-specific articles are released.

And if you need expert support with your GenAI project, please reach out my consultancy PH1 Research. We specialize have worked with Spotify, Microsoft, Mozilla, government agencies, higher education, and more on emerging technology projects.