Last week I talked about the fundamentals of how to build your first AI experience (read about it here).
This article is about deciding what to build with AI and how it can fit into your product strategy.
(For context, I spent the last week drafting an AI strategy for beehiiv, and this was the exact framework I used)
This will cover:
Understanding how AI is currently being used alongside your product
Finding opportunities for AI to add magic into the product experience
Finding greenfield opportunities for AI in your industry
Prioritising and ranking
What not to do
Let’s get started.
Developing a product strategy typically centres around three concepts:
How can we make our current product fundamentally more valuable to our existing users (→ improving ARPU, Expansion, Churn, conversion)
How can we access new customers (→ TAM expansion, acquisition)
How can we ensure we maintain a competitive edge long term (→ long term growth & pricing power)
When we look at strategy through the lens of AI, the thinking should be exactly the same. i.e. how can AI help us achieve those three goals.
But, the question you should be asking is not “How can we add AI to our product?” but instead “how could AI make our product fundamentally better?” AND (this is important) “what will make our implementation better than others?”.
Here’s the step-by-step process to follow to answer that question:
Step 1: Mapping the hidden AI flows in your product
The first thing you should do is map out exactly how your users are using AI alongside your product.
This is where users are effectively creating a new hybrid user journey where they move from your product to an AI platform and back again (potentially multiple times).
Why is this important? Pre-AI, you could reliably assume how users interact with your product and what their journey was.
With AI, users are creating entirely new workflows, where AI tools (that you have no visibility over) are acting as an extension of your product. It’s key that you know what the new journeys are and why people are resorting to AI.
To map these, you can use any product management framework, although the ‘Jobs to be done’ (JTBD) is probably the easiest to think about (as the user is effectively bringing another tool into your flow to fulfil a specific task for them).
Here are some examples of hidden AI flows within beehiiv’s product (using my writing process as a reference):
Ideation & planning:
during the week I will drop ideas into Chatgpt/Gemini and using voice mode, I’ll outline all of the thoughts I have relating to that article (voice mode is better on chatgpt).
I will then get AI to take my thoughts and create an outline for the article. (Depending on the topic I’ll kick off some AI deep-research too).
When I’m ready to write, the article outline containing my ideas from earlier in the week are brought into the beehiiv editor.
Editor flow:
Once I’ve written this article I will pass it to Gemini (or more recently Claude Code) and get it to act as an editor, critiquing the structure, tone, succinctness etc.
I’ll then come back into the beehiiv editor, re-write certain sections, and re-share with the editor for feedback
I’ll repeat this multiple times, diving into specific paragraphs either I or the AI feel are weak.
Image styling:
This newsletter has a specific style for its header images. (I’ve found Gemini Nano Banana Pro to be the most effective for creating these)
Once I’ve reached the final steps of writing, I will leave the platform
Social sharing ideation:
once I’ve written the article I need to come up with roughly 3 posts for linkedin and X that are based on the contents of the article.
I will take the finished article into Gemini and prompt it to create 3 posts for each platform (one to hype it the day before, one for the launch day, and one for 2 days later).
I will then schedule these out on the relevant social platform
Website design (layout)
If I’m looking to create a complex layout of elements on my site I will pass in a screenshot of what I’m looking for into Claude, and ask it to explain how I need to structure columns and nest containers to achieve it. I will then apply it’s recommendations in the web builder.
After you’ve finished this process, you should have a list that looks something like this:
Task (JTBD) | Hybrid flow |
|---|---|
Planning, Ideation, research | Chatgpt (store title ideas) → Chatgpt (voice note ideas) → Chatgpt deep research → beehiiv editor |
Edit and refine newsletter | beehiiv editor → Gemini → beehiiv editor → Gemini → beehiiv editor …. |
Create a header Image for newsletter | beehiiv editor → Gemini (Nano Banana Pro) → beehiiv editor |
Generate social hooks | beehiiv editor → Gemini → Linkedin / X |
Figure out complex website layouts | Website builder → Claude → Website builder |
Step 2: Rank your hybrid flows
Take your list of tasks and flows, and begin to rank them on the following dimensions:
Threat
Is there a genuine threat that the AI they are using could displace our product and lead to churn? and if one of our competitors included this within their product, could this displace us/ threaten our value proposition?Replicability
How easy would it be for us to replicate this flow in-product?Journey improvements
How much better would the user experience be if this was in-product?Differentiation
If this was in-product, to what extent will our data allow us to give the user a better quality output than their current flow.Long-term advantages
If this flow was within the product and we had visibility over it, how much of an advantage would this give us long-term?
Task (JTBD) | Threat | Replicable | Journey | Different | Advantage |
|---|---|---|---|---|---|
Planning, research | 1 | 3 | 3 | 4 | 5 |
Edit newsletter | 1 | 5 | 5 | 5 | 4 |
Header Image | 1 | 5 | 5 | 3 | 3 |
Social hooks | 1 | 5 | 4 | 3 | 4 |
Website layouts | 3 | 3 | 5 | 4 | 4 |
The higher the score, the higher the opportunity and potential impact.
You now have a list of potential initiatives with clear sense of how important each of them could be to the product.
Step 3: Find opportunities for magic.
As a PM I often talk about creating ‘magic’, and AI has huge potential to add magic into your product.
Product Magic
A product experience that exceeds expectations, surprises and delights. It’s what fuels word-of-mouth, drives customer loyalty and has contributed significantly to beehiiv’s success.
To find these opportunities for magic, focus on your product flows and experiences, and look for:
areas of high friction
areas of high complexity
areas of frustration
Once you have these, you should be asking the question of “How could I make this experience 10x better that previously wasn’t possible/feasible without AI?”
Here are five examples from my work managing beehiiv’s website builder(and as someone with a website built on beehiiv):
Areas with frustration/complexity | How could this be 10X better? |
|---|---|
New user struggling to figure out how to adjust the design attributes to customise her design | Being able to tell a chat exactly what you want and have it make all the changes in the background. |
Duplicating multiple web pages and forgetting to update the meta titles/descriptions | Having something that flags if the meta title/description doesn’t match the contents of the page it’s for. |
Unknowingly setting your site up to not be optimised for SEO | Having something that reviews your site setup and pages, flags issues and can fix them for you. |
Linking to a page that is 404’ing but not realising when you publish your site. | Having something that can explore your site and check if everything is working as expected |
Having to manually update styling across whole site after deciding to change something small (eg corner radius for buttons) | Having something that can find all similar elements and UI across your website and proactively update the styling for you. |
(you get the idea…)
In the above examples you can see a few opportunities start to emerge where agents could significantly improve the experience:
Web Builder Agent (we launched this in November) - a chat experience that can help users interface with the advanced functionality of the web builder without having to understand web design. It also unlocks key functionality like being able to upload a screenshot you want to replicate, or ask it to cascade styling across a page/site.
SEO Agent - a background agent with a deep understanding of SEO best-practice that can review the setup of your site (everything from meta descriptions, redirects, H1/H2s etc) and intervene with recommendations, and make changes as requested.
Website review agent- a background agent that can review and interact with your site before it’s published, finding any issues (eg broken links), verify pixel tracking and more broadly give you the peace-of-mind that is usually achieved by having a dedicated web person on your team.
Why did we tackle the Web builder agent first?
One the best mediums for users to interact with agents is via chat and the most natural implementation for the Web builder agent was to introduce it as a chat within the web editor.
This will ultimately become the place where a user can interact with any other web-related agents we build down the line.
These first three steps will give you a clear idea of how AI features could augment your current product.
The next step is about blue sky thinking and ideating on net new products.
Step 4: Ask what is possible now with AI that was never possible before in your space/vertical?
This is the Greenfield thinking step: think beyond your existing product and focus on everything your users currently do as part of their day-to-day. Where do they spend their time, where do they invest their resources, what cause them the most frustrations etc.
TLDR what are the biggest problems you could solve for your users?
Here are some ideas I had for how we could empower the 1-person newsletter publisher:
Could we facilitate truly personalised brand outreach (at scale) for publishers to lock in partnerships?
Could we use AI to perform negotiations with advertisers on behalf of the publisher?
Could we use AI to take the insight/contents from the newsletter and generate an authentic video podcast using their voice & likeness + create all the marketing clips etc?
Could we use AI to constantly research and observe trends, proactively come up with topics for you to write about, research them and create an outline?
5. Align, rank & prioritise
You now have 3 lists of initiatives:
How users are incorporating AI alongside your product
Opportunities where you could improve your product using AI
New big picture opportunities for you to explore
Next, you need to merge this with everything else.
Align it with your core product strategy - ‘AI’ is not a strategy, it’s simply a new tool to help solve problems for your users. Whatever you decide to implement should amplify whatever you are doing as part of your core strategy. (Don’t get distracted by the ‘shiny AI feature’ that you don’t need)
Score and rank your AI initiatives - each idea should be judged on its impact, cost and likelihood of success. (I find RICE is a good framework for this, but any framework can be used)
Adjust your effort scores and re-rank - Because AI features are probabilistic (check out last week’s newsletter) ie the output is not guaranteed, it requires a lot more ongoing resources, both from a PM and engineering perspective, to monitor and improve it. So I recommend increasing your ‘effort’ scores for the AI initiatives to reflect the amount of work that will be required.
Prioritise against everything else you’re building - AI is just part of your strategy and takes up resources like everything else. Combine your new AI list with your existing product backlog for features and experiments, and rank them.
If you’ve done this correctly you should now have a much clearer view of all your AI opportunities, the impact they will have on your product and user experience and have alignment across the product team for how these should be prioritised.
Let me know if you found this useful!
See you next week.
Jake
