AI product strategy: key steps, examples, and best practices

Seemingly every product manager is tasked with creating an AI product strategy.

The C-suite expects you to build one; your sales team asks about it (because customers and prospects ask them); and your marketing team wants one so that they can soon join trending AI conversations.

The reality is that certain AI use cases for your product are significantly more valuable than others. Being able to identify and weigh these use cases appropriately as well as execute any effectively requires a well thought out AI product strategy.

We’ll break down how you can build yours, but first, let’s align on what an AI product strategy is and why it’s important.

What is an AI product strategy?

It’s a comprehensive plan for using AI in your product. This includes how you’d prepare data, select a model(s), integrate with 3rd-party applications, and deploy the AI functionality.

We’ll go into each of these parts in more detail in our step-by-step section.

Related: What is a product experience strategy?

Why you should integrate AI into your product

While it depends on the AI features you build, incorporating AI into your product provides a wealth of benefits.

• Enhanced customer experience: AI features can help your users find resources faster, complete tasks quicker, make better decisions, and more. Taken together, your customers can be more productive and save ample time

• Competitive differentiation: Your competitors likely haven’t built out—let alone executed on—an AI product strategy. Being able to do so will, as a result, give your organization a competitive advantage 

• Revenue growth: Assuming the AI product features you build are unrivaled and improve the customer experience, they should help you close more deals and retain more customers

• Media coverage: While this isn’t a primary reason to invest in AI product features, they can help you get coverage from prominent media outlets, spurring further awareness of and demand for your product

How to build a strong AI product strategy

AI product strategies naturally differ depending on the type of product you offer, your customers’ and prospects’ needs, and more. But here are some steps that can guide you toward an effective strategy, regardless of your situation.

Step 1: define the problem AI should solve

Understanding the issue that’s most critical and that AI is best positioned to address is foundational to any AI product strategy. 

Here are some examples:

• If you’re experiencing a slow time to value, you should prioritize AI capabilities that remove friction in adopting your product

That’s exactly what Causal, a financial planning platform, aimed to accomplish by launching their AI Wizard. Here’s more from Causal’s CEO and co-founder, Taimur Abdaal:

• If you’re receiving feedback that information is too difficult to find in your product, you should prioritize building enterprise AI search functionality
  
  
• If there’s a certain product workflow that takes your users a long time to complete, you should prioritize incorporating AI to automate it or provide personalized recommendations for completing it sooner
  

• If your product generates a sizable amount of data or code that needs to be interpreted quickly, you can leverage AI to help distill the key takeaways

Semgrep, a cybersecurity platform, for example, aimed to use AI to both automate repetitive product workflows and highlight potential security threats from a given set of code. Here’s more from Sachi Shah, the Head of Product at Semgrep:

• If users are unsure of the best ways to use your product, you can build an AI chatbot that can help them pinpoint their best use case(s) as well as tips for putting it into practice

Step 2: assess your data requirements

Your product will need certain types of customer data to power its AI features. For instance, a recruiting automation solution would likely need candidate data to power AI features; while a financial planning platform would probably need customers’ ERP data.

On top of that, you’ll need to normalize the integrated customer data, or transform it into a standard data model. This ensures that it’s embedded accurately before it’s added to a vector database, where it can be retrieved for relevant queries and fed to the LLM you use to generate outputs.

And lastly, you’ll need to align on the frequency at which the data needs to get synced—whether that’s daily, every few hours, or in real-time.

Based on all these requirements, you’ll need to determine whether you can build the integrations and transformation logic natively or if you should outsource this work to a 3rd-party integration solution.

{{this-blog-only-cta}}

Step 3: choose the right AI model 

While there are only a few LLMs you’d choose from—namely from providers like OpenAI, Meta, Google, Anthropic, and Mistral—each has relative strengths and weaknesses.

You’ll need to understand the specific capabilities that are most important for your specific use case(s) and then evaluate the models' relative efficacies across these capabilities. It may even be worth testing the models to validate their claims and to see which are most reliable.

Step 4: plan for user experience and adoption

Your users may not know how to use the AI feature and may be reluctant to try it—especially if they’ve used your product in a certain way for a long time.

To help ease the transition and get them comfortable with your AI feature, you can provide simple instructions for using it and/or offer nudges that help guide them.

You can also provide opportunities for users to provide input on any aspect of the feature, whether it's the outputs generated or the UI that’s provided—allowing you to refine it further. And you can even task customer success managers with scheduling customers calls for the explicit purpose of training them on the feature. 

Step 5: bake security into the AI feature

If your AI feature shows any security weaknesses, your platform’s credibility will quickly nosedive.

To mitigate and proactively address any security vulnerabilities, you can invest in a variety of initiatives:

• Use strong encryption protocols, like AES-256, for data at rest and in transit
  

• Adopt role based access controls (RBAC) so that only select users can use and/or modify the AI feature, and regularly review these roles to ensure they’re implemented correctly over time
  

• Use scopes for your customer-facing integrations to toggle off any data you don’t want to sync

• Leverage API observability tooling to uncover potential security incidents related to an integration quickly 

Step 6: go to market slowly

The roll out can consist of a few key markers: a pilot, a beta, and a GA push.

1. Pilot test: You’ll need to develop a minimum viable product (MVP) of your AI feature and launch it to a very small subset of users. Ideally, these users opt in to the MVP because they can immediately benefit from using it and are willing to be design partners.

2. Beta test: Once you’ve identified and tackled the p0, p1, and p2 improvements from the pilot test, you can roll out the MVP to a larger audience that’s chosen at random. This larger audience will naturally have more varied experiences with the feature, so you can expect even more feedback. 

At this phase, you can also perform A/B tests to suss out the AI feature’s impact and whether it’s addressing the problems you identified at the beginning. And you can determine if the AI feature works at scale or if there are infrastructure issues that are causing delays.

3. Push to GA: Finally, once the AI feature meets your beta customers’ expectations and works as intended, you can begin to roll it out to every customer.

Examples of AI product strategies

To make this more tangible, let’s walk through a few AI product strategies in action.

Guru builds cross-category integrations to power a comprehensive AI enterprise search

As part of their AI product strategy, Guru wanted to help every type of user leverage their enterprise AI search platform—from customer support reps to HR specialists. They also wanted every user to have easy access to their platform.

To put this into action, Guru built product integrations across ticketing systems, HRIS platforms, file storage applications, and more. 

Once customers connect to these systems, their employees can ask Guru all kinds of questions from the application they’re already working in (e.g., Microsoft Teams). Guru’s LLM can then use the relevant information provided in the connected system to generate responses in plain text (with the help of natural language processing, or NLP).

Peoplelogic uses normalized customer data to fuel reliable AI agents

Peoplelogic, an AI for HR platform, offers “Nova”, a suite of AI agents for people ops.

To help their AI agents consistently generate accurate and helpful outputs via email or Slack, Peoplelogic normalizes their customers’ employee and candidate data via Merge.

For example, one of their AI agents—“Noah”—can provide accurate summaries of candidates for interviewers by using normalized data from candidates' ATS profiles.

Here’s more on the benefits of using normalized customer data from Matt Schmidt, Peoplelogic’s CEO and co-founder.

Ema uses enterprise-grade secure integrations to prevent security issues

Ema, which offers a universal AI employee, needed to make sure that their AI agents only shared information with users that fell within their level of permissions in their employer's file storage solution.

To help enforce this, Ema uses Merge’s file storage integrations—which have access control levels (ACLs) baked in. 

Here’s more from Swati Trehan, the Head of Operations and Strategy:

Best practices for product managers in the AI era

As you and your PM team navigate building AI features, it’s worth keeping the following in mind:

Let customers lead the way

Your customers may already have ideas for leveraging AI in your product. And in some cases, they may be executing on these use cases without you or your team being aware of it.

That’s exactly what happened to the team at LaunchDarkly, a feature management and experimentation platform. Here’s more from Karishma Irani, LaunchDarkly’s former Head of Product:

To collect their ideas, you can run surveys that include open-ended questions (to prevent influencing them and to allow them to share unexpected ideas); you can schedule focus groups with a few customer champions where they can share ideas and refine them together; and you can schedule 1:1s with your most forward-thinking customers to go deep on their particular ideas.

Related: Best practices for creating product objectives

Become fluent in AI

Your team shouldn’t just focus on building particular AI features; you should also commit to having a deep understanding of how LLMs function and stay up to date on any changes. 

This in itself can help inspire new ideas, identify more effective implementations, and ultimately help you leverage AI in the most impactful ways for your customers.

Frank te Pas, the Head of Enterprise Product from Perplexity, shares more on why PMs need to have a deep understanding of AI:

Leverage AI at every phase of product development

Given its versatility, LLMs can and should fundamentally transform how you build products.

It should automate the repetitive tasks that plague your UX; it should take the technically-difficult tasks off your customers’ shoulders; it should inspire your team to launch and support new products. In short, you should use AI in every possible way that’d benefit customers.

Here’s more from Frank on how you can become an AI-first company:

Use a unified API solution to support the integrations your AI features need 

Your AI features and products likely need customer data, whether it lives in their file storage systems, CRMs, ticketing tools, ATSs, and so on.

Instead of tasking your engineers with building and maintaining customer-facing integrations themselves, you can use a unified API solution to add all the integrations you need through a single integration build.

On top of that, a unified API solution normalizes all of your customers’ data so that your LLM can generate more reliable outputs, and it offers the integration observability features your customer success team needs to handle integration issues independently.

Learn how unified API solutions can help you take best-in-class AI features to market faster and more easily and why Merge is the leading unified API solution by scheduling a demo with one of our integration experts.