Interview with Macro, Pi Product Manager: With over 3 million global users, is it still necessary to continue learning Gamma?
A tool that makes creating PowerPoint presentations easier than ever before is quietly changing the way many people work. It combines AI with traditional presentation documents, requiring almost no learning of complex operations; simply input your ideas to get a complete presentation...
A tool that makes creating PPT presentations easier than ever before is quietly changing the way many people work.
It put AI Combined with traditional presentation documents, almostNo need to learn complicated operations; simply input your ideas to get a complete demo.
For those who have never made a slideshow before, it makes content creation tangible and accessible for the first time.
exist WAIC After the exhibition,Pi has surpassed 3 million users worldwide..
Every day, training instructors, science popularizers, and even small workshop owners who have never opened PowerPoint create their own presentations here.
Some people use it for online open classes, some use it for sales roadshows, and some just want to give a decent introduction to their product for the first time.
The product was created by a small team of a dozen people. Without massive advertising campaigns or complex growth strategies, they relied on the product itself and a few trade show events to bring it to millions of users.
Macro They said that what they initially hoped to create was something that could bear...MultimodalA "content container" for content. Text and images can be directly placed inside, and audio and video will be added in the future.let AI By combining generated content with manually edited content, a truly usable result can be produced..
Product Name: Pi (Presentation Intelligence)
Slogan:AINative content creation and sharing Agent Platform, more thanAIPPT
Main functions:AINative content creation and sharing Agent This platform represents the current state-of-the-art (SOTA) standard in content creation, integrating knowledge management, a design engine, and...AIFull-stack native editorintelligentbodyThe Copilot system provides global users with professional-quality, accurate, and comprehensive services.High efficiencyA harmonious experience in content creation and sharing.
Official website: Pi
The following is our interview with Macro, the product manager of Pi:
Global user base surpasses 3 million
๐ฉโ๐ฆฐK Sister Research Society
Hi~Macro, what have you been up to lately?
๐ง๐ปโ๐ป Macro
Mainly working on WAIThe C exhibition was a great success overall, from the pre-event activities to the closing ceremony. Our user base has reached another milestone, surpassing 3 million users globally. Then there are some brand promotion matters.
๐ฉโ๐ฆฐK Sister Research Society
Is the project being promoted simultaneously both domestically and internationally?
๐ง๐ปโ๐ป Macro
Our goal is to be world-class.AIWe hope to first validate the product and technology in the domestic market, refine the details, and achieve a higher level of product completion before making a full-scale push.
From language models toMultimodalContent container, redefined AI Creative Forms
๐ฉโ๐ฆฐK Sister Research Society
Why did you want to make this product in the first place?
๐ง๐ปโ๐ป Macro
We feel that since we've already entered...AIThe times call for change; the traditional, inefficient way of creating PowerPoint presentations must be abandoned. In the past, most people would first search for templates online, then painstakingly modify them word by word, often wasting a lot of energy on layout because the template slots didn't match the content. It was extremely tedious.
If a product can not only improve efficiency but also provide creativity, then it is a truly excellent product.
๐ฉโ๐ฆฐK Sister Research Society
What were your initial ideas for this product?
๐ง๐ปโ๐ป Macro
When we first developed this product, the model was just a language model, and most products on the market were just chat boxes.
We judge that the future is definitely...MultimodalTherefore, we hope to use a kind ofThe "content container" can hold text, images, and even music and video, etc.,let AI The generated content is combined with manually edited content to produce usable results.
Based on extensive competitor research and our experience with editors, we chose to develop our own similar one. Gamma The editor, asMultimodalThe carrier of content.
Our product currently supports text and image modalities, and music and video modalities will be gradually added as the model capabilities mature in the market.
Previously, content was created by a creator and then distributed, which was a one-way process. We hope that content can also generate interaction.
Therefore, we've incorporated interactive methods such as questionnaires, and in the future, we will also support users to interact directly with content during the content creation and distribution process.AI It will assist in answering users' questions.
This was the direction we initially planned, but some features are still being iterated.
๐ฉโ๐ฆฐK Sister Research Society
This can be understood as you now focusing on demonstration scenarios and integration.MultimodalCapabilities. The platform currently supports image generation, but video generation is not yet available. Is this because the results haven't met expectations?
๐ง๐ปโ๐ป Macro
Yes, we haven't added video functionality directly yet. This is because if users spend a lot of time configuring videos page by page, it would negatively impact the overall experience.
The same applies to image processing; our users generally don't need to write it themselves. PromptThis will generate the first version of the image. It's only used when extremely high quality is required. Prompt Adjust or regenerate.
We also hope that users won't feel pressured when creating videos on the platform. Prompt Instead of being troubled by technical skills, you can directly generate content simply by using natural language.
Users only need to focus on the content itself, as well as basic aesthetics and layout..
๐ฉโ๐ฆฐK Sister Research Society
Like manyintelligentbodyBased on the themeautomaticWill you consider including pictures later?
๐ง๐ปโ๐ป Macro
Yes, we have that now too.automaticWith picturesAI It will generate images for you based on the context of your content, or use the "decorative images" that come with our theme.
We are also considering using multiple Agent It helps users collect information and analyze which data and images are more suitable for the current content scenario.
We are following up on this technology, but we also have concerns, namely data copyright issues. Therefore, I am currently conducting compliance research and preparing a report. If the risks are manageable, we will catch up as soon as possible.
At the same time, we are also building our own image library, which has collected a large number of images without copyright risks, and will be made available to users in the future.
๐ฉโ๐ฆฐK Sister Research Society
Why did you consider adding features related to design and management, such as custom fonts and interactive questionnaires?
๐ง๐ปโ๐ป Macro
Let's start with the design and custom themes. We're essentially creating content layouts, and layouts are inseparable from aesthetic design.
Therefore, we hope to open this channel to users with high aesthetic standards or design skills, so that they can create more beautiful works.
Relying solely on defining themes ourselves would certainly have limitations. We now have a WeChat group of about a hundred designers who are both our initial users and co-creation users. They create some exquisite themes for us, and we also provide certain creative rewards.
In the future, if users have their own visual needs, they can also create them themselves through this method, instead of having to rely on us to customize themes.
Changing the layout, typesetting, and text only requires a single mouse click.
๐ฉโ๐ฆฐK Sister Research Society
Can you put anything into your editor?
๐ง๐ปโ๐ป Macro
Yes. We support uploading PPTs for secondary formatting, and high-quality works shared by users in the community can be used directly. New coding formats and HTML web pages are also compatible. We can also adapt to future changes in interaction methods.
I think that currently each household uses Agent If you really want to delve deeper into PowerPoint creation, you'll ultimately have to rely on your own editor. And now, when it comes to developing editors, most will likely follow our direction.
๐ฉโ๐ฆฐK Sister Research Society
There are many presentation document and PPT generators on the market now. The traditional method is to use templates to fill in content, and a single sentence can generate one. How do you balance the relationship between creative effect, design effect, and content production efficiency?
๐ง๐ปโ๐ป Macro
Based on our own observations, there are four main technical architectures for creating PPT presentations on the market:
- The earliest templateโFor example, final draft,CanvaThe most traditional approach is to move the template online and let users apply it themselves.
- Template +Large ModelcopywritingโFor example, pixel blooming,AI In PPT, mark the slots in the template and use...Large ModelGenerate the text and insert it.
- Our self-developed editor modeโAfter the model generates content, it is formatted according to the editor's definition.
- AI Coding generates HTML-formatted PPTs.Many manufacturers are now following suit.
We are more optimistic about the third model for two reasons:
- High content completenessIn traditional templates, if the template only has 4 points and the model generates 6 points, the last two points may not be able to be placed; with our template, we can arrange as many points as are generated.
- More typographyintelligentSmoothOur model can understand semantics; for example, it can understand content with a timeline.automaticUse a timeline layout, and present parallel content in a parallel structure.
Compared to using Agent Alternatively, with our HTML-generated solution, our advantage lies in the editing experience and speed: generating an 8-page slide can be completed in approximately 30 seconds;
Agent Most of the solutions take several minutes, consume a lot of tokens, and are inconvenient to edit; many of them can only change the text, and changing the layout or color requires interacting with natural language again, which takes another 10 or 20 minutes, resulting in a very poor user experience.
Here, changing the layout, typesetting, and text only requires a single click of the mouse.
๐ฉโ๐ฆฐK Sister Research Society
Now everyone is trying to use it. AI Coding generates web pages or presentation documents, but it can basically only export PDFs, making it difficult to convert them into editable PowerPoint presentations. Do you think this model will impact your market in the future?
๐ง๐ปโ๐ป Macro
It's not fair to call it a direct impact. All companies are still iterating their products, and we're also internally researching this model.
We also hope to introduce more AgentFirst, we need to solidify the pre-existing content for users, such as helping them with search and analysis. This will result in more complete and visualized data. Agent Its obvious advantages.
However, considering the editing experience, secondary editing, and exporting, this model is likely to converge with our current framework in the long run. It's just that creating a truly user-friendly editor takes time, so I believe they will eventually become similar.
From the domestic version Gamma Start by sticking to a small team approach
๐ฉโ๐ฆฐK Sister Research Society
Why do you want to target the domestic version of Gamma?
๐ง๐ปโ๐ป Macro
I used Gamma and Tome back in 2021 and observed that many people in China were using Gamma, especially in the tech sector, where people are willing to embrace emerging products. From an operational perspective, we wanted to attract this user group first.
๐ฉโ๐ฆฐK Sister Research Society
We've observed that Tome previously shifted its focus to marketing, and now it's even shut down the business. What are your thoughts on this phenomenon?
๐ง๐ปโ๐ป Macro
Tome differs from Gamma and us in that it focuses on free editing. Judging from the results, they didn't stick to their initial direction, which may be related to the team's background.
Compared to the same period last year, Gamma is currently operating very well. The small team is doing well, with a good ARR and stable income.
Therefore, our analysis suggests that Tome's transformation is more likely due to team factors or, more accurately, capital factors.
However, we remain optimistic about this sector, and many companies have recently been doing... Agent The team is also entering this niche market.
๐ฉโ๐ฆฐK Sister Research Society
Does the team size and organizational structure also align with Gamma?
๐ง๐ปโ๐ป Macro
Our team currently consists of about 16-17 people. The team leader (CEO) strongly advocates for a small team model because, in today's world, it's unnecessary to have people crammed into every aspect of the business like in traditional companies.
In the future, we also hope that the team will not exceed 30 people.This completes the entire product closed loop.This is very similar to the Gamma pattern.
๐ฉโ๐ฆฐK Sister Research Society
What is your current product-to-R&D ratio? Have you explored a model like the one Andrew Ng mentioned, where "one product manager is paired with 0.5 R&D personnel"?
๐ง๐ปโ๐ป Macro
We're not so excited about pursuing a high proportion. Now, it's basically one person playing multiple roles. For example, I used to be a product manager, but now I'm responsible for part of the product, most of the operations, and some business development. I also mentor one or two junior colleagues who are a little less experienced.
We currently have a slightly larger R&D team because developing a good editor is quite a complicated task. AI Coding can completely help.
We have about ten people in R&D, and the rest are in production and operations, so the ratio of R&D to production is roughly one to one. However, I don't think this ratio applies to our internal operations, because several R&D staff also have product design capabilities and awareness.
People who had never made a PowerPoint presentation before became core users of the mini-program.
๐ฉโ๐ฆฐK Sister Research Society
For which user groups did you initially design this product?
๐ง๐ปโ๐ป Macro
In the initial stage of launch, we did not conduct targeted promotion. Instead, we hoped that more people would try it out first and see what kind of content creation they could use the product to accomplish.
From the perspective of data feedback,It works best for creators of training and science popularization content.Because they create content very frequently and have a large audience, they often come to modify or recreate content three or four times a week. This type of user has the highest engagement and the most paid content, and their content best matches what we want to present.
Their work goes beyond traditional PowerPoint presentations, incorporating web pages, videos, and other elements that perfectly complement our product.MultimodalThe original purpose of containers.
๐ฉโ๐ฆฐK Sister Research Society
Other companies generate content on web pages. But you also created a mini-program. Why?
๐ง๐ปโ๐ป Macro
Firstly, considering domestic usage habits, many people don't use computers frequently, and we hope that mini-programs will make it more convenient for more people to use and experience them.Large ModelThe resulting changes in content creation;
Secondly, it's about creating a closed-loop content ecosystem. Mini-programs are an excellent distribution channel in China, allowing content created on the web to be distributed more directly to friends or other users via mini-programs.
๐ฉโ๐ฆฐK Sister Research Society
What are the differences in user proportions or usage behaviors during the operation of mini-programs?
๐ง๐ปโ๐ป Macro
The biggest difference is age. Mini-program users are generally older, many are in their forties or fifties or even older.
They were curious, "Can this help me make a PPT?", and many people in China haven't actually done much PPT work.
To give a vivid example: When we were in school, our older teachers would typically create a PowerPoint presentation with just a few lines of text and a single image, almost no formatting, but that was considered quite good among their peers. Many people in that age group, however, couldn't create a PowerPoint presentation at all.
I have an uncle in his sixties who often goes out to sell his own products. It's a small workshop, and no one helps him with PowerPoint presentations. After hearing that I was making this product, he became very eager to learn.SimpleHe was able to create a product description using his phone after just a little instruction, and he was very happy with it.
During our promotion efforts, we also discovered that there are indeed many people in this category.
๐ฉโ๐ฆฐK Sister Research Society
Yes, they are actually a group of people who are being ignored.
๐ง๐ปโ๐ป Macro
๐ฉโ๐ฆฐK Sister Research Society
Are there differences in payment behavior among users of different age groups? For example, which group has a higher payment rate: users in their forties and fifties or younger users?
๐ง๐ปโ๐ป Macro
The paid conversion rate is still higher among younger users. Slightly older users, if they are more eager to learn, will also be converted to the web version after using the product.
Because on a webpage, once you've created two pieces of content, you can basically learn how to use our product. The overall barrier to entry is low; it doesn't involve any coding, and you can complete it by clicking the mouse and typing on the keyboard.
The fault tolerance rate for B-end clients is less than 1%, so we are currently prioritizing the C-end client.
๐ฉโ๐ฆฐK Sister Research Society
In terms of business model, do you prefer the C-end or the B-end?
๐ง๐ปโ๐ป Macro
Prioritize the development of the consumer market (C-end).Large ModelThe core value lies in improving efficiency, but in B2B scenarios, the accuracy requirement is extremely high, and the error tolerance rate may be less than 1%. In contrast, the error tolerance margin is larger in the C2C sector, so we are currently focusing on the C2C sector.
Once the basic model level can effectively resolve illusions and ensure error-free operation, we will invest more heavily in the B2B sector. Currently, we are also paying attention to B2B opportunities, but our core focus remains on the B2C sector.
๐ฉโ๐ฆฐK Sister Research Society
Would you consider opening up your editor and presentation document generation capabilities to external users? For example, could you provide them to other companies via API or MCP?
๐ง๐ปโ๐ปMacro
We are already working on this layer, and have already connected with some well-known companies, such as Kanter (an international consulting firm).
They have internal models and their own database to generate content for their users. Now, they want to present the results to clients in a visual format, so they've called upon our capabilities to generate slides.
We also created a video demonstration of this case study, which is now live.
In addition, some educational institutions have also accepted our API to create slides for training scenarios.
This time WAIAt the C exhibition, many companies were quite interested in us, including some conglomerates, and they all came to inquire about our relevant capabilities.
W debutAIC. Online activities bring in tens of thousands of new users every day.
๐ฉโ๐ฆฐK Sister Research Society
In WAIDuring the C event, you launched a lucky draw activity. Could you share it?
๐ง๐ปโ๐ป Macro
Lotteries are a classic and effective method in traditional operations, as people are quite interested in unknown prizes and the thrill of probability.
We are at WAIThe company's first public appearance on social media marks its first time showcasing itself to the public, and the company hopes to use this opportunity to introduce itself and its products to a wider audience.
At the event, our booth on floor H3 was constantly in a queue; it was probably the only one on the entire floor with a long queue.Yu ShuThe longest queue outside Rokid.
However, the on-site traffic was limited, so we also launched a simultaneous online viral marketing campaign.The campaign was a great success, with tens of thousands of new users joining every day..
๐ฉโ๐ฆฐK Sister Research Society
How did this lottery event achieve viral growth and spontaneous user promotion?
๐ง๐ปโ๐ป Macro
We spent a lot of money preparing the prizes, including our own merchandise, all of which were of high quality. The T-shirts were washable and even featured embroidery.
This is also related to our founder's genes. He has a high pursuit of aesthetics and quality, so even when giving away a large number of peripherals to users, he put a lot of thought into it.
In addition, we also incorporated Labubu's current popularity into our giveaway campaign. After two weeks of running the campaign, we discovered that Ele.me had also started using Labubu for their marketing and viral marketing campaigns, and this approach proved to be quite effective.
๐ฉโ๐ฆฐK Sister Research Society
What are your main growth methods and channels?
๐ง๐ปโ๐ป Macro
We currently have several operational initiatives being rolled out gradually. The team hasn't hired any particularly outstanding operations experts; we've basically been learning as we go, but the overall operational framework is now complete.
The first segment involved a very small amount of investment, but it was really small, almost negligible compared to several competitors.
The second part is collaboration with social media KOLs.AIDuring C, we also discovered some high-quality bloggers, and many bloggers spontaneously came to our booth to check in. We will also do some networking activities later.
The third part is operational activities. We plan to launch a small event every month or two, but the themes will be different.
This time WAIActivity C primarily focuses on promotion and user acquisition. Later, we may hold smaller creative competitions to bring together high-quality creators and foster a more collaborative environment.High efficiencyThe co-creation mechanism.
The content is more fancy, and the long-term goal is an interactive platform.
๐ฉโ๐ฆฐK Sister Research Society
Actually, you want to build a creator ecosystem and provide personalized services for advanced users, right?
๐ง๐ปโ๐ป Macro
Yes, that's one way to understand it. On another level, we're currently developing tools, but in the long run, we don't want to be limited to just tools. We aspire to become a content platform, a carrier of the next generation of content paradigms.
Content initially started with text and images, like on Toutiao; later it evolved into short videos. What form will content platforms take in the future? Everyone is exploring that, and we are too.
One thing that excites us internally is that future content will be interactive, but how exactly it will be interactive is still being explored step by step.
๐ฉโ๐ฆฐK Sister Research Society
Have you recently explored any new directions for product capabilities?
๐ง๐ปโ๐ป Macro
Yes, we've been working on something lately: making the content a bit more fancy. The current editor layout is structured, but many domestic users are used to the traditional whiteboard mode of PowerPoint, where they can assemble and draw their own content.SimpleIt can be done, whether it's complicated or not.
User research indicates that users want more fancy results, so we are developing this capability and expect a major upgrade to the generation and layout capabilities by the end of August.
๐ฉโ๐ฆฐK Sister Research Society
A huge thank you to Macro for sharing so many stories about Pi with us today, showing us new possibilities for content creation. We also wish Pi continued rapid user growth and look forward to Macro's next visit to share more surprises.
๐ง๐ปโ๐ป Macro