Email Love is supported by

April 24, 2024 - Articles

Email Peeps 41: Dipanjan Dey

“Dipanjan

Tell us about Kombai and how you became CEO of a company on the cutting edge of AI and email development.

Kombai is a purpose-built AI that can convert any email design to production-ready code in just a click. Email marketers, designers, and developers can use Kombai to reduce their time & cost to develop custom emails by 95%+ and improve their campaign performance by sending fully coded, infinitely creative emails.

To be honest, I personally, and Kombai as a company, landed on email development by a bit of fortunate serendipity. We started building Kombai as a general-purpose AI that can understand and code any UI design. We launched this in August 2023 on ProductHunt. Thankfully, the launch did well for us and ended up as one of the top 10 products of the year for 2023. 

When we saw how users were using us, we noticed many users were trying to code email designs with the tool. This surprised us because, at that point, our code wasn’t optimized for email clients. We were even more surprised when we spoke with some of these users. They told us they’d pay us if we could optimize our code to support the major email clients. So, we built some modifications on top of our base models to start generating code that’ll work on all email clients. That got us our first few customers. After that, we have been lucky to have some incredible customers and users who have been generous with their feedback and helped us figure out how to serve the email-related use cases better.


How did you train your LLMs to code emails?

We actually didn’t! 🙂

Because none of the existing LLMs models can accurately understand UI designs, we spent 15+ months building and training purpose-built ensemble models from scratch. In fact, 99%+ of our email code output comes from our own models. You can read more about the technology behind our AI models here.

We use LLMs to enrich our email code output for one specific, small purpose. Fortunately, this problem area is where best-in-class LLMs already have strong, generic capabilities. So we didn’t have to “train” them further.


Will AI eventually replace email developers, or do you see AI acting as a co-pilot to skilled developers?

Based on what we see in the market, I think the email development world will be bimodal. Email developers will continue to be critical for complex use cases, many using AI as their copilot. But marketers and designers who don’t have access to email developers will use AI to develop fully functional, high-performing HTML emails in seconds without compromising their design visions.

Today, email developers are often used for complex use cases that need specific human judgment. These judgments could be technical – e.g., specific customization of responsive behaviors, or logical – e.g., when the scope or design is incomplete. Both these kinds of judgments are uniquely human in nature. We believe all stakeholders are better off having real, high-skilled developers in charge of exercising these judgments. However, AI can very likely help them along the way by giving them a good V1 of the code that they can build on.

However, most email marketers and designers don’t have access to skilled email developers. They create emails in one of two ways, both of which leave a LOT on the table in terms of both email experience and campaign performance/ deliverability:

  • Using image-only/ mostly-image emails: Experts of all stripes have been telling us that using image-only emails is very bad for your spam risk, loading experience, and accessibility. In fact, it’s a staple in every email deliverability help doc you’ll find from any major ESP (e.g., here is the one from Klaviyo).
  • Using a WISYWIG tool: While these tools can be great for creating some emails that don’t have to stand out in an inbox (e.g., transactional emails), they severely limit the marketers’/ designers’ ability to deliver compelling, unique, on-brand email experiences. 

The majority of our customers and users today come from this group. We help them deliver fully coded emails at a fraction of the time and resources it’d otherwise take them. In such cases, instead of replacing email developers, we democratize access to email development to people who cannot use one today.


What’s in your toolbox?

Hardware: 

  • A Lenovo Thinkpad (I get a lot of grief from our dev team for still being on Windows. But, hey – someone has to have access to a real Outlook-on-windows!)
  • Sony wh-1000xm4 headphones

Software:

  • Calendly
  • Loom
  • Grammarly, ChatGPT
  • Google Workplace
  • Zoom
  • Slack
  • Linear

How will AI be used in other areas of email marketing?

In the immediate term, AI can be a copilot on many of the email marketing tasks, particularly for generating ideas and first drafts. 

We also see a lot of applications of AI in copy writing, particularly for cold emails. In addition, compelling use cases of AI are emerging in asset creation, segmentation, personalization, and customer behavior prediction.

On a related note, one interesting thing to watch out for would be – if AI makes it too easy to personalize/ optimize things at scale, how will subscribers/ readers react to the berge of hyper-personalized messages? Personally I have been much more immune to personalized cold emails today than a few years ago. We also hear from many email marketers and sales folks that the conversion numbers have either staggered or fallen despite all the AI-driven personalization.

Very likely, access to AI will be more and more democratized over time. So, I think marketers have to go back to our unique human abilities – to be creative and connect with our users/ readers/ subscribers to make their email campaigns stand out again.


What’s your favorite email campaign of all time? Why?

Recently, I came across the welcome email for Huggies® Baby Steps, which really stood out for the email geek in me! The email has client-specific user interactivity – it shows some rich user inputs in supporting clients and an alternative section with CTA buttons in non-supporting clients. It also helps that the email has great copy and design. 

Also, I’ve been a big fan of the emails from Canva. Their emails are designed beautifully and have all the email building/ coding best practices for high deliverability and performance. They also often include topical campaigns and initiatives.


What are some simple ways that brands can improve their email marketing with AI?

Thank you for this opportunity for some shameless plug. First of all, everyone should always code their emails with Kombai! 😛

Self-promotional kidding aside, I can think of a few lower-hanging fruits that most marketers can probably explore right away:

  • Image generation and editing:
    • But my favorite tool for email use cases in this category is Adobe Firefly. It lets you do a bunch of really cool image editing/ generation without needing graphic design skills. For example, it can do very realistic recoloring of any part of an image/ background, which can be very helpful in creating on-brand assets that fit in with the rest of your emails. 
    • You can also use Dalle or Midjourney to generate infinitely creative campaign images. However, some solid prompt-engineering skills might be needed to generate realistic/ on-brand images.
  • Copywriting and inspiration: 
    • Tools like Jasper, Copy ai, or even ChatGPT can be great to generate a V1, or candidate ideas, for any copy you have to write. They can also be helpful when you have to rehash existing content to create new copies or make segmented versions of a campaign, rewriting copies for different personas.
    • However, you probably would want a human to edit/ finalize the autogenerated copies, to avoid sounding too canned or “AI-ish”.
  • Predictive insights: 
    • ESPs have started to come up with their versions of AI, that can provide actionable insights on your campaign data, and help you schedule smart decisions. For example, Klaviyo AI can help you predict the next order date, lifetime value, and customer churn risk. We hear mixed feedback on their accuracy today, but I’d expect them to improve over time. Meanwhile, it might be useful to check them out yourself and use them if their results pass your sniff test.

How do you manage work-life balance?

I love to read, travel to new countries/ places, experience art and cook. 

Honestly, though, I don’t think I’m particularly good at this balance, especially with the start-up’s growing pains over the last few months. Thank you for the timely reminder! 🙂

Much love,
Andy

Email: [email protected]
Twitter: @emaillove

Email_Love__3_-removebg-preview

Sign Up for
Curated Inspiration

Join over 6,000 other Email Geeks that receive actionable insights, new tactics, fresh design inspo, and profiles of your favorite #Emailpeeps.