How to build your own Product Manager GPT

Good morning. Today, I am writing myself out of a job—or maybe into a new definition of my job: becoming a leader of a well-trained GPT that does Product Management. Truthfully, with the recent announcements of GPT-4o and Google’s Project Astra, there won’t be much left to do for us knowledge workers in the future. As an example, in this post, we’ll look at the value a Product Manager adds and how to replicate that with a GPT. Likely applicable to other job functions as well.

I can’t stop innovation. But I can at least be at the forefront of it. Another 6-minute read. I’ll give you two minutes back next week.

6-minute read

The technology needed for it

Most people use it all the time. You know it…, your employers know it… and your co-workers definitely know it when they read your well-worded, deeply researched, insightful emails.

And why wouldn’t you? It’s a time-saving new technology for all of us. However, it is risky for employers because they don’t want sensitive data to enter third-party services without a thorough review.

The solution is coming. GitHub, Microsoft, Adobe, and Atlassian—all companies that provide software tools to knowledge workers—are working on or already providing enterprise GPTs. Those GPTs can be trained with your own company data within a trusted environment. Meaning that the data you use to train will stay in a closed system. And that is a hard requirement for enterprises.

That’s why enterprise software companies are bringing enterprise-grade Copilots to market. Data security is the unique value added.

If you want to train one for non-sensitive tasks or are not worried about data sharing, you can follow a simple tutorial on how to do it on OpenAI.

By the way, I know purists say technically, this is not “training” the AI. We give it files and data for retrieval. I will do my best to be “accurate.”

With these services on the rise, it is only a matter of time until we can all use our company's internal GPT.

Now that we have that outlined, let’s look at the work of a Product Manager.

What are Product Managers needed for?

There is a polarizing argument about whether PMs are actually needed in an organization. Aren’t PMs “just adding extra meetings and presentations to our calendars?”

My view: If designers and engineers know exactly what the market needs, sequence it in the right timeline to be built and brought to market, and everyone agrees and is aligned on everything, you might not need a Product Manager.

In reality, this is not the case, and there is a place for Product Managers to drive product initiatives.

I separate the value added into 3 buckets:

• Data - Data discovery, processing, and use to help with decision-making.

• Relationships - Working closely with stakeholders to keep them informed, agree on deliverables, and share the vision and goals.

• Management - Drive delivery of a product, clear blockers, and help people understand context, direction, and timelines.

Data is the place for a GPT to take over.

If that GPT gets fed all the data a Product Manager uses for their work; it can take over, concluding, helping with decision-making, and being an internal interface for all stakeholders to ask, “Why have we decided to do this?” “What is this for?” “Are our users asking for this?”

Let’s make this more actionable: How do PM responsibilities break down?

Now, we look at which would require a PM to gather and analyze data. This is what the GPT can augment.

Next, we break down the steps and data buckets and feed them to our GPT.

Onboarding PMGPT

Imagine you’d hire someone new to your team. What information would you share with them to ensure they understand your efforts' history, vision, and intentions?

Your GPT needs to know everything about your product, its features, and your vision for it. Teach it market dynamics and what the users in that market need.

It’s like a write up that would be a product and spec sheet. And another writeup about the market you are operating in. It also needs to know who the target audience is.

Questions that help to craft these documents:

• Why are we building this?

• Who are we building this for?

• What does our product do?

• Why is it unique?

• Feature sheet - any and all documents that describe your product externally - website docs, PDFs.

• With the recent announcements of GPT-4o and Google’s Project Astra, we will soon be able to feed it all our existing designs, prototypes, and architecture diagrams—any and all useful visuals to understand the product.

Superpower: Let your PMGPT pretend to be one of your user personas

Product Managers define personas to build a very tangible version of a user. We scribble down everything we know about our users in 4 different categories. That’s the data your GPT needs to learn.

About - Specify the demographics of your persona. 

Bio - What is the background of your persona? What is your persona doing? What are their motivations and beliefs? What do they enjoy?

Goals, Wants, & Needs - Goals and Motivations - must be linked to your product. They probably also want scrambled eggs for breakfast on a Saturday morning. But if you are building a finance app, that doesn’t really relate. Focus on where your product can help them achieve their goals.

Pain Points - Be careful; this is a very important section. It guides designs and implementation. This is the most important Persona section. It must be bulletproof and reflect a subset of your target audience. Detailing this will also inspire you and your GPT when brainstorming.

All Personas evolve over time. Do this to help your GPT evolve with it:

Get a regular data dump of whatever tool you use for customer feedback. Take in data wherever users leave feedback.

If you are in B2B, your Account and Sales Managers will likely all have long lists on a per-customer basis. Make it a task to regularly update your GPT on evolving user needs and pain points.

Add an impact number, for example, key customer data (by deal size) and their specific needs; your GPT can help you prioritize based on business impact. This can even plug into your CRM tool to always be up-to-date.

Now, you have a real-life bot trained to be your target customer. You can have deep, meaningful, inspiring conversations with them.

The Market you are in can have many implications.

Let your GPT know what market you operate in. This has implications for privacy, data regulations, and compliance. Note: PMs and PMGPTs will not replace the counsel of a legal advisor.

But on a high level, PMs must be aware of certain market regulations. So once all your PMs are replaced by PMGPT, ensure the PMGPT knows your go-to-market plans, locals, and the channels you plan to distribute your product through (direct sales, marketplace, etc.).

Next level of autonomy for your PMGPT

If you feed your GPT all the data above, you will have a great PMGPT that can run you through research, need-finding, and brainstorming sessions.

To add another level of autonomy, why not add your org chart and teams and size that work across engineering, design, customer support, and so on? Can be useful if you are looking for a specific stakeholder.

Now that it knows the teams and size let’s add one last dataset:

• Your planned Initiatives (or Epics) is what we call the projects we are planning to work on.

• The estimated effort it would take to build them out

• The criticality you give an initiative - ideally ordered by priority rank

• And more data about reach (how many users/customers need it)

The PMGPT can tell you which initiatives are “doable” within a given timeframe. And which might need additional capacity on any of the teams. It can also be prioritized via reach and impact.

The below doesn’t work with GPT out of the box. This is an idea that needs deeper integration; someone should build it:

When we get into the product execution phase, AI can set up all meetings that usually occur throughout a product's lifecycle: standups, sprint planning, backlog grooming, and leadership reviews.

We have seen the early stages in last week’s deep dive on ServiceNow. Their onboarding AI sets up 1:1 with relevant people on their first day.

Finally, PMGPT takes over the work.

Now that our PMGPT has all this background knowledge, it can create Product Requirements Docs for us. They won’t be complete, but we can use them as a good starting point.

To help your GPT learn your preferred format, upload a sample PRD.

You can use my Notion template here.

Then work with the prompts below to get your content created:

Useful prompts to use are:

For specificity, I would always describe my own vision for the next feature and product, including the problem statement, WHY we built it, and a high-level summary of WHAT we are building. We PMs take a lot of pride in that.

Then, feed it to your GPT. After that, the following prompt should give you a very valuable, time-saving output:

Every PM has a different structure for their user stories and requirements, but you should get the idea and be able to modify this.

If your GPT knows your Org structure, let it also draft up the stakeholders section.

That’s it; now you have PMGPT ready and running for you.

After all this hard work, you should invite your PMGPT to the launch party. Cheers.

Please send it along to someone who might be in desperate need of a GPT.

Have a great rest of the week,