Disrupt yourself, or be disrupted: Intercom, Cleo, and Kive talk AI beyond the hype
Unless you’ve been living under a rock, you’ll have heard all about AI and how fast it’s moving - and the pace of change is only accelerating. Looking at the latest YC cohort, almost every single startup is some version of “AI for X”. Which means it’s high time, if you’re not already, to start thinking about how it impacts your business — or someone else will do it for you.
I have found that there are three reactions to this latest wave of AI innovation. First, there’s business as usual: ignoring what’s happening and pretending you can wait or that you don’t need to worry about it. Then there are the companies bolting on some AI features and slapping an “AI Inside” sticker on it. Or you’re going back to a blank sheet of paper: rethinking your product and business from the ground up to better solve customer problems using these amazing new tools — to the point where it’s so seamless, your customers aren’t even sure you’re using AI.
That’s how the panelists of my recent portfolio event, ‘Beyond the Hype: Leveraging Generative AI in your product’, are approaching it. We invited Des Traynor, Co-Founder of Intercom, Anita Woods, VP Product at Cleo, Olof Lindh, Founder and CEO of Kive, and Pietro Casella, Chief Architect at EQT Digital/Motherbrain Labs to share how they are rethinking their products and processes with AI.
Let’s dive in.
And then there was ChatGPT
‘We built our first AI tool in 2018. It was useful, but not transformative. And then ChatGPT happened,’ said Des. ‘Fergal, our Director of Machine Learning, who’s been in ML for like 12 years, played around with it for a few hours. He said he’d never seen anything like it in his life.’
Intercom saw the transformative potential of ChatGPT almost immediately. However, it’s one thing to recognise it, another thing to act on it. But act on it they did. Five weeks after ChatGPT went live, they had built AI assistance into their product, and they’ve recently released Fin, a full end-to-end support agent powered by GPT-4 and Intercom’s proprietary ML tech.
Des emphasised the importance of facing up to the magnitude of change required and the need for rapid adaptation. ‘The scale of opportunity can be scary, even if it’s good news. We had to change the org chart, product roadmap, finance plans…and we had to do all that in weeks if we wanted to capture the opportunity.’
Putting together a “code red” team is crucial — a dedicated, cross-functional team that figures out how generative AI will transform your product and industry. You also need to rethink your org chart. If you want to build fast, partner the best AI thinkers in your business with the best product people and get the right stuff on the roadmap. You can’t silo your AI talent. This approach helped Intercom quickly pivot their long-term product strategy and business model from selling support tools to selling a bot that does the support, effectively replacing their direct customers.
So you have to face the wider business implications, even if these are detrimental in the short to medium term. ‘These API calls are not cheap. And that’s a whole new thing,’ said Des. ‘But job one is to stay alive.’ To play in this space right now, you either have to build your own model — which is a big expense in itself — or accept that you’re going to be making a lot of API calls. Initially, this might result in tighter margins, but it gives you time to figure out what to do longer term. The priority should be using the models and tools on hand, and then develop your own if it makes business sense. As Pietro pointed out, some businesses are starting to train smaller, more affordable models with these larger foundation models, so they can run these cheaper models in production — a promising area for innovation and product development.
And when it comes to pricing, you’re not chasing cents on the dollar; you need to think big picture. Look at the value AI is generating for you across your value chain. In Intercom’s case, AI displaces the time spent by human support and does things near instantly rather than with a 25-minute delay. You need to take all those first- and second-order effects into consideration when thinking about price vs the value you provide your customers.
The AI-problem fit
Once you’ve responded to the immediate AI wave, how do you keep up with the rapid pace of development?
Part of it is thinking about the new problem and solution space that AI opens up, said Pietro. First, you need to determine whether an AI-based system fulfils the specific requirements and expectations of your problem domain. In other words, does it genuinely help you solve the problem at hand? Then you can evaluate how generative AI’s abilities — such as classification, text and code generation, reasoning, execution — and new UI syntax might improve or transform your offering.
Take one small example. Right now, we think about autocomplete in terms of finishing a word or sentence. But generative AI means we can jump from that to hitting tab to automatically execute a full set of tasks. Accepting an email invite, organising a calendar event, inviting the right people, and sending an AI-generated transcript after the meeting — all by pressing a single key. ‘Good designers have always tried to anticipate what the user wants to do next. AI is the superpower for that,’ said Des. This paves the way for the development of autonomous AI agents like AutoGPT, which we’re only now seeing the beginnings of.
There are also new AI building blocks in the tech stack and emerging design patterns to consider, such as vector databases, “models-as-a-service”, and multi-input LLMs. You simply have to keep your ear to the ground and get hands-on with them. Understanding these new elements will enable you to leverage them effectively to augment or transform your product.
But, as Pietro and I both stressed, the old playbook still applies. Any great product or company, AI or not, needs to be built on solid tech, customer-centricity, and great UX. AI is no shortcut to success. And it’s such a dynamic environment that being the first mover is not much of an advantage. The winners will not be those who move first, but those who keep moving to continuously rethink their products in this new paradigm.
Des summarised, ‘If your product involves imagery or text, voice or audio — and soon video — and it involves anything that is in any way guessable or predictable, the chances are it can now be done cheaper and faster. It can now be done by more people in more ways. And because of that there’s a huge opportunity for new startups here to overtake the old generation. I really suggest you don’t waste that opportunity.’
Cleo and Kive are certainly making the most of it.
Stories from the portfolio
Cleo — Becoming the conversational AI for your relationship with money
Cleo has been focused on AI for financial health for six years. But that didn’t stop Anita and the team from seeing the new opportunity posed by generative AI. However long you’ve been building with AI, these new tools open up completely new possibilities.
‘Actually, the easy part was getting buy-in,’ said Anita. And it’s easy to see why. Today, Cleo’s chat AI can handle 11,000+ non-data-driven intents (e.g. what’s a good mortgage rate?) and it has 4,000+ data-driven responses (e.g. how much have I spent on McDonalds this year?). With ChatGPT, Cleo can go from that 11k+ hardcoded responses to tens of millions of answers that are generated probabilistically, while also adding more depth and context to conversations. In short, generative AI for Cleo means radically more intelligent chat at scale across different domains. Customers get a conversational AI with personality that gets to know their habits and can remember past conversations. It can be the copilot for their entire financial life.
This development has real promise for financial literacy too — one conspicuous gap in formal education. Imagine if kids could chat with a bot that helps them navigate their financial journey, adapting to their life goals and habits.
‘Early on, it’s about having the remit to play around with this stuff, and seeing organically what the opportunities are,’ said Anita. ‘The hard part is thinking about how we actually make this a real thing. Getting the right resources, monetisation, what new hires we need, new business opportunities…Those are the conversations we’re having now.’
It also underscores the importance of chat-first UI in a generative AI world. As Des noted, most people know what they want to do, but in an unfamiliar product they usually don’t know how to do it. The ability to express your intention in natural language could be incredibly advantageous, making chat-first UI an essential component for future products.
Kive — Get your hands dirty!
‘We knew Generative AI would have a significant impact on our business,’ admitted Olof. ‘Would people even save their visual assets if they could just generate them on the fly?’
Instead of despairing, Olof and the team at Kive embraced the potential impact and started getting their hands dirty. They built a proof of concept using a diffusion model in just four hours, demonstrating that they could rapidly create a functional product with generative AI. Next, they organised a generative AI hackathon for the whole team, leading to the development of AI Canvas — a collaborative generative AI creation tool. ‘It was like Figma meets Midjourney,’ said Olof. A mere 10 days after the hackathon, AI Canvas was released on Product Hunt. Now it’s Kive’s native edit function.
The proof of concept and hackathon spurred learning and boosted company-wide confidence, proving that the challenge was not insurmountable. ‘Getting your hands dirty and experimenting is essential if you want to identify the opportunities that AI presents to your business,’ said Olof.
He also stressed the importance of considering the full customer workflow. Most of Kive’s customers used the existing product during the inspiration and pre-production stages of visual asset creation. But could they use generative AI to help their customers leapfrog the entire production process and directly create the finished asset instead? This idea led the team to product photography. Taking everything they’d learned in their experiments with diffusion models, they trained models on products and discovered they could generate photos of products in any environment. ‘We can put any product in the Alps, the desert, the rainforest…We can produce custom visual worlds in a thousand different locations, without travel or emissions — and for a fraction of the cost,’ explained Olof. AI is more than just a feature; it revolutionises the landscape of problem-solving and product development.
And you can start small. Kive deliberately partnered with customers who wanted small-scale solutions, allowing the team to understand how this new technology functioned on a smaller scale before integrating it into their core product. This strategy enabled them to continue developing on their core product, while innovating on the side.
Anno dominAI
AI is a rapidly changing landscape, and there are a lot of moving pieces — the big tech players competing in the foundation model space; regulatory and copyright issues; specialised vs generic models; the development of multi-modal models; AI agents and digital employees.
But that doesn’t mean you can wait to see where the chips will fall. The companies who succeed will be those who get their hands dirty now, who start experimenting with the tools as they come out in order to rethink their products and find new opportunities to better solve customer problems — and be ready to throw it all out and start again when the next thing comes along. Intercom, Cleo and Kive are all testament to this lesson.
As I said in my introduction to the event, ‘every company is an AI company now.’ If you’re not thinking about how AI disrupts your industry, someone else is. So get building.
