AI tools for Product Teams (part 1)
As the tool ecosystems evolves rapidly, here are some great tools you could use to get the job done within a product team.
Introduction
I don’t remember such fast pace in technology as it is today with AI. LLMs are flourishing, and it’s difficult to discern which one is better than the other, forget about the differences in which one to use.
The same has propel companies to start introducing AI features to their products and others companies to burst into the scene that are claimed to be a full-fledge AI product.
Regardless the difference between brand-new products and existing ones, I think it is relevant to take advantage of them. What can we do today much faster than 5 years ago? How can we take that advantage to our favour to deliver better products and faster than before? How can we get some inspiration to introduce some of these capabilities in our products? These are some of the questions that inspired me to write about it.
Disclaimer: I’m on the team who believes that AI is not replacing, but making us better. The caveat is that if you can do more with less, companies will think and have another alternative that was not available before, this is even more relevant on a bear-market vs a growing one.
Mental Model
I’ve decided to use a simple mental model I learned from Jeff Patton, where he describes the continuous product improvement cycle.
For simplicity reasons and because I love enabling constrains, I’ve decided to limit to 4 the number of tools for each of the parts described above. The criteria was a combination of existing tools I used in the past I truly like and brand-new ones that caught my attention.
Sense Tool Stack - Quantitative
Amplitude
Amplitude has launched recently four great capabilities that are AI native. Since they gather a lot of data, they are very well positioned to offer insights, predictions, and recommendations to lowering the entry barrier for users who may not be familiar with queries, events, retention charts, and so on.
I loved when you use the Ask feature that it returns the potential charts that might answer the question:
If you don’t want to listen to the suggested charts, the answer it pretty neat, and you can open the chart and do all the previous actions with it. See how at the bottom of the chart they are showing the query in a format familiar with long-time Amplitude users.
Another one that caught my attention and I suffered in the past was confidence in the data. Even if I did the best to standardise the way events and properties were created, it slowed the entire insight generation process down. This seems to be a huge leap forward to always work on data you trust in a simple and straightforward manner.
I’ve worked on Alert functionalities that allowed customers to become more proactive and using webhooks to send that to the proper team to take a look. Having the same capabilities without having to configure anything, in the context of the chart I’m looking at, is becoming increasingly needed. This is where the efficiency comes into play.
MixPanel*
Although I believe MixPanel is a great tool, the number of AI capabilities is limited to creating content, charts in this case, using NPL and asking further questions. That content can be inserted when building a report. It’s called Spark.
Pendo
If you have ever used Pendo, you know that it’s quite different from the previous two tools. At Nexthink we have used it extensively for generating guides and content for every release and onboarding of new users, as well as tagging our product to inform our decision-making process. They released an AI feature that suggests steps with content in the tone you choose.
Another interesting feature they have announced is workflow insights. With Pendo you can check the most common steps your customers are following (based on the tagged you have provided) within your product. Therefore you can get information on whey they may be dropping or not executing a particular action. Although some of the suggestions could be obvious, it triggers the idea that you should revisit the flow, because something may be causing friction or problems.
Last but not least, similar to the Data Governance in Amplitude, Pendo’s core action is the manual tagging of pages, clicks, and actions. I recall spending some time deciding the name of the page and event. Only to realise later I have to change it and sometimes it did it in the wrong way. That’s why the have incorporated AI to assist the tagging and make sure it’s done properly, with the proper standards.
Heap*
I haven’t used Heap in depth beyond playing with some capabilities available within their free account, but I knew quite a few product folks that recommended it to me as an alternative to Pendo.
They have branded all their AI capabilities behind “AI co-pilot” and it includes three functionalities: chats, summaries, and follow-ups.
As you may predict, chat is the conversational interface that is present in other products. You ask for a relevant event, behaviour and the product will reply back. Same for the subsequent questions (2) along with the ability to forward you to the area of the product to interact a bit more (1). One interesting thing that it was quite interesting to me, was the CTA “Understanding my approach” (3), which prompts the co-pilot to explain its reasoning on why it shows events, time, and properties for that answer:
The other two functionalities are interconnected. Summary assists you in the creation of your chart by adding a name and description in the form of a questions - This is brilliant IMO, because it helps the readers of the chart what this chart is answering and it could be used to feed the chat functionality (1). The follow-up feature presents, within the chart created, a list of other questions that are related to the content of the chart, that would take you directly to chat (2), and you can keep interacting with it:
Notice how they are adding a visual cue to highlight the fields that have been filled by AI, and how the idea of getting questions to go and interact with chat reinforces the engagement by interacting with chat. I’d expect they keep adding points in the product that will render this capability at the center, especially to dig deeper into analytics topics that are the ones taking more time to uncover.
Enterpret*
I bumped into this tool in one of the newsletter I read often and what caught my attention was the claim of “Transform customer feedback into product growth”.
The reason I’ve decided to include this tool, it’s because I haven’t seen a full-fledged solution that addresses this problem effectively. The vast majority of tools are covering just a tiny percent of the customer feedback (sometimes just surveys, other just the ideas they submit, and so on). Also, the frustration of being in meetings with just some part of the story, and having to collect many different points of views, which takes time and it has my bias incorporated.
One of the main traits of this kind of tools is the large unstructured data you get from a wide variety of tools. Therefore, what I like about the AI offering is how machine learning taxonomy automatically captures every layer of feedback to give you actionable and relevant insights that reflect your customer's voice. More feedback, better LLM predictions, better insights, and greater user value.
The biggest visible feature is Wisdom: The AI Copilot for Customer Insights. It allows customers to Summarised queries:
Dive in and get deep: see how on top of the summary you get references on the number of times that reason was mentioned, and you can jump straight in the source that originated it, and share it with the rest of the team working on this issue.
Generate content: you can use the customer feedback to generate relevant content to address your customers, as well as for your product team. Imagine that you can use this feedback to create a PRD that contains the most relevant insights, issues, and data points to make a point on why you should invest time fixing X.
Conclusion
As you can see, regardless of the type of product you are using, all of them have adopted the text interface to fulfil their main use case. Using the Kano Model as an analogy, chat interfaces seem to be the basic of AI. It seems that if you don’t have it, you’re already behind.
Then, depending on the capabilities each product is offering, some AI feature are performers, and will vary on the main problems customers have to interact with each of the products. Honestly, I haven’t seen any feature that is an attractor or delighter, but I totally believe they are coming.
Although some of the tools are very similar (Amplitude & MixPanel and Pendo & Heap), they have adopted different approaches. That’s how horizontal AI can be and how much room for keep expanding humans’ ability is left.
Next steps
I the following article, we will continue diving into the AI tools for product teams for the qualitative part of Sense.
As a spoiler, we will be analysing the following tools: FigJam, Dovetail, Gong.io, TL;DV, and Fathom AI.