How to Use AI for Content That Actually Converts with Siva Chellamuthu (#74)

Pete Vera, Exit Algorithms: Welcome to exit algorithms, the podcast where we decode what it really takes to unlock growth, streamline operations and prepare your business for a high value exit. I'm your host Pete Vera. And today I'm joined by Siva Chalamutu. He's a technical leader and software engineer who's built products inside TikTok and Metta and is now the CTO and co-founder of posterchild.

ai a platform. using AI to help nonprofits tell better stories and raise more money. Siva, super excited to have you here. Welcome to the show.

Siva: Super excited to be here as well Pete. Thank you for inviting me to this show ⁓ happy to share ⁓ Knowledge from what we have built so far to the founders out there Pete Vera, Exit Algorithms: Awesome, yeah, I appreciate it. I'm looking forward to diving in. Do you mind starting off?

Can you share a bit about your background, your career journey and how you ended up at Poster Child? Siva: Sure. So my path has always been about using technology to give people access. It started with the small nonprofit work back in India.

Then I worked on consumer products at scale and now through Poster Child. ⁓ I grew up in South India and the small nonprofit that ⁓ I had as a student volunteer, ⁓ the lesson was like ⁓ talent was everywhere but access to tools, guidance and funding was uneven. So we helped thousands of kids and students get access to computer education, job readiness and certification opportunities. Some of them were sponsored by some college institutions and Microsoft back then.

Later, I built a music video app for Windows Store, which had millions of downloads. A fun fact there is like ⁓ I got sued by Taylor Swift lawyer. cease and desist. So, ⁓ yeah, the, yeah, that, ⁓ that startup taught me difference between, you know, building something people use versus building something that can serve as a business.

was, anyway, it's a fun gig. ⁓ then, ⁓ I spent my time at some startups, Zillow, ⁓ TikTok, ⁓ and Meta, ⁓ TikTok, especially I worked on. Pete Vera, Exit Algorithms: Ha ha ha. ⁓ no.

Siva: content creation at massive scale. So we created something called photo mode. It is the Instagram type of content inside TikTok among the short form videos which reached hundreds of millions of monthly users. Yeah, it's the scale ⁓ that stick with me and I was pulled back to zero to one problems where the product and market are being reinvented.

⁓ With respect to Post4Child, ⁓ we started it as a content automation idea, but the stronger signal came from nonprofits, where storytelling was not just marketing, it's directly tied to fundraising. I met my co-founder, Leandro, in July 2024. ⁓ The first idea was a social media listener to know ⁓ who is talking about your brand ⁓ and help... Small businesses create better content.

⁓ But nonprofits showed a stronger need. ⁓ Same content problem, but fewer resources and much higher stakes. ⁓ Organizations like Oakland School of Arts and Colorwave, which was ⁓ co-founded by my ⁓ co-founder, Leandro, ⁓ made the problem real for us. They had...

hundreds and hundreds of stories from their alumni sitting in their actual sheet ⁓ collected in the form of ⁓ survey testimonials, but there is no real tool to convert into ⁓ community stories, into trust, fundraising, and support. ⁓ So once we got deeper into the data, we realized the fundraising side was even bigger than the content. So that is when Postal Child become... Pete Vera, Exit Algorithms: Hmm.

Siva: more than a content tool, and it become an AI-powered storytelling and fundraising platform. Pete Vera, Exit Algorithms: Yeah, awesome. Wow, I love your story. Very unique background.

I like the, you know, the theme seems to be, know, it's a lot of your projects are mission based and, you know, you have a big, big outcome. I was curious, can you expound upon, you know, zero to one problems and its context, you know, within Poster Child? Siva: ⁓ For sure. So for me, I had an ⁓ operational experience back in India ⁓ with a small nonprofit, but ⁓ the nonprofit in US is totally different.

⁓ So you are literally starting from scratch. But luckily, there is a lot of open source data around it. ⁓ All of the... ⁓ organizations, information are public and they all do file IRS data.

So that become the data pipeline to, ⁓ you know, like create the data layer to power this product. So we ⁓ created a data layer of 1 million nonprofits, over 150,000 funders are among that pool. ⁓ And we learned 20 years of data, of ⁓ giving grants and That is ⁓ an opportunity I saw zero to one, and that resonates with handling things at scale at TikTok, ⁓ the larger data processing, the pipelines, and that came together at Postreel. Pete Vera, Exit Algorithms: Yeah, awesome.

Wow. Yeah. Were there any other lessons learned during your time at TikTok or Meta, these bigger companies ⁓ that you've taken over and maybe even on the business side over to Poster Child? Siva: Yeah.

So TikTok and Meta taught me how to think about content, data and infrastructure at scale. And like I said, I brought the same mindset into the sector, where the tooling has historically been underbuilt. At TikTok, when you ship a feature, it reaches hundreds of millions of users. That changes how you think through.

data pipelines, performance, experimentation, product quality. ⁓ For PostRoshel, that means building the data foundation early instead of only wrapping an NLM around a workflow. ⁓ So we pretty much started with the data layer, didn't just use NLM to do something for the industry. ⁓ This is on the fundraising side.

On the content side, what I learned ⁓ from TikTok and Meta is how distribution channels shape content. The same story has to change depending on where it goes, who's reading it and what action you want them to take. ⁓ This resonates with whether you are publishing ⁓ as a carousel or you're publishing as a, you know, like a single image post or a short form video. ⁓ That is the lens I brought to Post4Child.

⁓ Non-profits already have powerful stories, but they do not have the system to turn one story into a social post, a newsletter, a funder update, or a board narrative, or a campaign message. So that tool didn't exist at all. So we were not just ⁓ generating content, we are building a creation layer. that helps a nonprofit adapt the right story for the right channel and for the right person and for the right outcome.

So that's a bigger thing that I took from ⁓ and applied to this industry. Pete Vera, Exit Algorithms: Yeah, love it. Yeah, I think that emphasis on story is powerful. That, you know, there's a lot of literature out there now, but, you know, like telling a compelling story is really how you get people interested, right?

At the end of the day. That's awesome. Yeah. How has ⁓ the journey been so far building Poster Child?

⁓ I'm curious about the name and how has it been ⁓ hiring more people and evolving in this crazy world of AI we're in now. Siva: ⁓ Yeah, so I could, yeah, I could speak to ⁓ different questions. ⁓ But yeah, we simply help ⁓ turn raw community stories into ⁓ public funding data, ⁓ know, into content funding, intelligence and concrete fundraising actions. As the...

As you know, CTO, my role was to turn the vision into technical platform. ⁓ You know, like the data layer, the systems and founder intelligence, story engine and infrastructure. ⁓ So ⁓ I'll, so I gave you, I walked you through the briefly about the problems in the nonprofit industry and the solutions that we came up with and the current. Pete Vera, Exit Algorithms: That'd be great.

Siva: Yeah, the current traction ⁓ we have currently is like, ⁓ we have signed up 60 plus nonprofit organizations across 12 sectors. My favorite are ⁓ education, health justice, and community empowerment. So we have so far processed like ⁓ 180,000 customer stories with like around 300,000 community members contributing to it. ⁓ So, yeah, and our fundraising product has surfaced billions of ⁓ grants opportunities.

How that works is we read ⁓ non-profit funders ⁓ websites ⁓ for new opportunities, new alignments. ⁓ So this has overall saved a lot of time for customers. ⁓ Our estimation is like... ⁓ The comps person at nonprofits spend around like four hours creating a story from reading the material into classifying it to generating an output to chat, GPD or Canva or other different means.

Same on the fundraising side. they it is a real manual problem. The fundraising, they have to read these IRS forms, understand who are the key people. Pete Vera, Exit Algorithms: Mm.

Siva: and develop a strategy to go after ⁓ that particular funder. This is for one funder. So imagine this at scale. So same for stories.

So we're solving both storytelling and fundraising at scale. Pete Vera, Exit Algorithms: Yeah. Yeah, got it. ⁓ Yeah, and have you been able to, you know, implement AI and automation in your processes?

I'm curious, you know, once you take on a client, what's your process to ⁓ get them going and start an effective media campaign? Siva: Yeah. ⁓ So ⁓ the way we build Postal Child is like ⁓ the AI starts with raw material that they come up with. It could be anything.

It could be a voice note from a volunteer at the field. It could be an alumni's podcast or ⁓ their testimonial about what they feel. So ⁓ then it helps actually answer four practical questions. ⁓ What story we tell?

⁓ Who should hear it and which funders does it matter to? And what should they do next with that information? So we take that. ⁓ a quick question here is like, do you me to go into the AI workflows or on a business level workflow?

Pete Vera, Exit Algorithms: Well, let's start with the business level first. Siva: Yeah, sure. So ⁓ we take that and we help answer that question ⁓ of how that story output should be. And we help them ⁓ classify it based on whether it relates to their specific program or the specific ⁓ topic they want to talk about.

For example, an education nonprofit want to talk about ⁓ recent graduation ceremony. So we classify that. And on the other side, we have the entire DNA of that nonprofit. So we know what that nonprofit is with their latest information from across the internet, their mission ⁓ and their past impact metrics, their ⁓ social presence and their awards.

So we use this ⁓ information layer combined with the raw ⁓ data coming in and combined with their intent, whether they want to generate it for a recent graduation ceremony ⁓ or they wanted to target on a specific funder who funds ⁓ for ⁓ STEM opportunities. So that get ⁓ stitched into an output. ⁓ So with respect to your campaign question, we see this as a multiple ⁓ channel output. So with just one creation flow, we generate an Instagram carousel, we generate captions, we generate a blog post, we generate a one-pager that they could use for funders who funds in the specific category.

So from there, they can use the content in multiple ways. ⁓ And on a scale, they can use a community of like 20 students to feature a specific fundraising page or a showcase page. Pete Vera, Exit Algorithms: Mmm. Got it.

Yeah, fascinating. I'm curious, there, ⁓ you see what is an effective social media marketing campaign ⁓ look like? How many times per week are they posting and what type of content or platforms work the best? Siva: Yeah, that's a great question.

⁓ based on our ⁓ user base, ⁓ the first platform, ⁓ or they go one to one is like Instagram and LinkedIn. So LinkedIn is where they want to reach for funders, ⁓ a funder related audience. Instagram is where they want to reach for scale and reach. ⁓ So Instagram ⁓ have more views and engagement compared to LinkedIn.

But LinkedIn kind of acts as ⁓ a record for your donor audience. So most of the donors, a big network individuals are in LinkedIn. Pete Vera, Exit Algorithms: I see. Got it.

Yeah, kind of dual approach there. Did they post, you know, is it effective to post multiple times per day or ⁓ what are the most effective campaigns? What do they look like? Siva: Yeah.

Yeah. So the number one problem nonprofits want to use the product ⁓ is to solve consistency. So they ⁓ are doing multiple jobs at a time. So the number that we saw they inclined towards is ⁓ anywhere from one to three times per week.

But it really depends on the season of it. ⁓ If they are doing a fundraising campaign, They tend to do more posts ⁓ with us and offseason where they are doing something on the ground. Pete Vera, Exit Algorithms: ⁓ Okay, yeah. Do you see any mistakes that they're making with their content in general?

Siva: Yes, so it's the mix of things. ⁓ So at Poster Child, we categorize content into eight different types. A ⁓ community spotlight, a donor spotlight. A few examples are those.

And ⁓ is it an impact story? ⁓ Is it a thought leadership? ⁓ Or is it an event recap? So our classifiers help classify that.

⁓ with the content ⁓ given as an input. So also we are working on creating a planner which keeps the content per week diverse rather than all of them being a thought leadership, all of them being, you know, even tree caps or just featuring a person. So it has to be always a mix of things. Donor, when they scroll the timeline, they want to see ⁓ the groundwork, the community spotlight, and also the impacts ⁓ that's been happening for this quarter or for this week, ⁓ and also some thought leadership.

How are you seeing the field that you are working on? Pete Vera, Exit Algorithms: Nice, yeah, that's really helpful. Yeah, having a variety of content types makes a lot of sense. Great, yeah, and another major theme of the show is AI, how people are implementing AI.

I know, of course, Poster Child is AI, basically native, right? Every aspect of it is AI-driven. Maybe a better question is how are you personally leveraging AI? whether it be productivity or in your day-to-day life.

Siva: Yeah, that's a great question. So yeah, yeah, I can answer this in twofold. So I'll start with ⁓ how we use AI for the product. And then I'll also talk about how we use AI for the teamwork and startup work ⁓ in the engineering work.

⁓ So with respect to the industry, ⁓ yeah, I think ⁓ small teams are being empowered. ⁓ with these ⁓ latest ⁓ tools. Two years ago, it's impossible to build a tool. ⁓ It's not impossible, but it ⁓ needs ⁓ larger ⁓ people to process, let's ⁓ say, 200,000 stories.

It takes a team of 50. So now with the LLM flows, it can be done at scale. ⁓ meaning ⁓ over the night, ⁓ LLMs with its context understanding can categorize and ⁓ bring out ⁓ these 20 stories are ready to go. These are the high impact ones, which has changed a lot of things for ⁓ small teams.

And I see this resonate for all the verticals. ⁓ It's been like people helping people move from intent. to context, to decision, to action, and then outcome. ⁓ But humans should still own judgment, relationship, and final decisions.

⁓ So ⁓ this is same across all verticals. The way I see it is, it is cutting down the research synthesis, prioritization, and giving you the first draft. ⁓ Pete Vera, Exit Algorithms: Mm. Siva: It is giving the capacity for small teams to operate like a much larger organization.

Pete Vera, Exit Algorithms: Mmm. Siva: Yeah, so that's ⁓ my perspective on the industry, nonprofit and also other verticals. But with respect to how we are using as a company, ⁓ AI should change how your own team builds, ⁓ researches, and test things and ships the product. ⁓ We use it for AI ⁓ across engineering, ⁓ product thinking.

⁓ market research, and drafting our concepts and iterating from that. ⁓ Yeah, so ⁓ we use ⁓ for ⁓ developing code ⁓ and with a team of three, ⁓ we kind of achieved a larger output ⁓ for this past one year. Pete Vera, Exit Algorithms: Mm. Sure.

Yeah. ⁓ incredible. I love that. Love that insight.

Yeah, I'm curious. Do you have any strategies to share around ⁓ generating the right prompt? I think that's where some people get stuck is they use it like a search engine or just they're not giving a quality enough input. then, of course, the output is less is pretty much garbage, too.

So. I'm curious your thoughts on that. Siva: Yeah, that's a great question. So it speaks to ⁓ the concept of how to use AI, right?

So AI sounds like, ⁓ the output sounds like garbage if you ⁓ don't give the right context. The model is only as good ⁓ as context you give it. If you prompt a generic model with the generic data, like if our customers go to ChatGPD and here's a story, that one of our community members shared, can you convert into social media captions? It will, but there is no relations to ⁓ how it knows about the organization, ⁓ what are their past successful social media posts are, how do they write their content.

Literally, it won't have any context to their brand voice, their historical performance. ⁓ So that is what is important when you come to doing something with the AI. ⁓ Our content creation goes through multiple layers. ⁓ This is the prompt.

We are dynamically stitching together. So a general prompt is ⁓ what ⁓ content to generate. Is it an Instagram ⁓ captions, carousels? ⁓ what ⁓ carousel type it is.

Is it a thought leadership? Is it an event recap? ⁓ And how the organization usually do the event recap. Here's an example.

⁓ And what is the content output requirement for the specific platform? Let's say LinkedIn has like word limit, right? So that, so all of these ⁓ things needs to be dynamically put together and tested. And even you should test with the different LLMs to see how they perform.

But on a deep technical level, ⁓ we also built evals ⁓ to measure the ⁓ delta of changes between each model, how they differ, and also overall used feedback both from customers and another LLM feedbacking how the output is to improve the overall quality. Yeah. So don't treat prompt just as a one-time thing. It is an evolving thing, and it is also a dynamic thing.

Pete Vera, Exit Algorithms: Nice, love it. Yeah, I great answer. I appreciate that ⁓ insight. I think that's ⁓ really useful for people trying to step up their game.

⁓ Yeah, well, it's been a great conversation, Siva. Last question I give you is what is one practical tip for business owners wanting to step up their game with their marketing game with AI? Siva: The marketing game. ⁓ Yeah, so, yeah, I have given some level of insights while we were talking.

It is the content that you generate should be tailored for the right audience and it should be the right type and you should use the right channel. It was hard to... tailored like that ⁓ before LLM. But now with LLM and the rules built around it, ⁓ you can just take one raw data, either video, audio or some corpus of text.

⁓ If you have the rules built in for each ⁓ channels, each type of audience, ⁓ each type of ⁓ content that you wanted to output. and that gets you at scale. ⁓ Imagine in pre-LLM days you only had capacity to create content for LinkedIn or just Instagram and you get lazy to do short form videos but now with the ⁓ system you built and you spend time like maybe weeks to build one time and now ⁓ that's automated and you have a reach to multiple channels and the equation is if you move from like one platform to many platform you are reaching more people different audiences and you're growing business Pete Vera, Exit Algorithms: Well said.

Yeah, no, amazing. And thank you, Steve. It's been incredible conversation. Where can listeners find you and learn more about your work?

Siva: ⁓ Yeah, I'm available in LinkedIn. We're active in LinkedIn and Twitter. ⁓ And also we have a blog ⁓ at postreel.ai.

⁓ Happy to share the links. Pete Vera, Exit Algorithms: Great, yeah, I'll put that in the show notes. Thanks so much for coming on today's show. See you, bye, I really appreciate it.

Siva: Yeah, thank you for the opportunity, Pete. And it's been a great experience to share the knowledge with the community, especially in the lens of content marketing. And that's one of the core things that we do at Poster Child. Thank you.

I appreciate it. Pete Vera, Exit Algorithms: Absolutely.