Analysing Intercom’s AI strategy

I’m exploring how to implement AI in businesses and today I’m looking at Intercom’s AI strategy.

Intercom — a leading customer support platform — is sharpening its focus, quickly becoming an AI-first company.

While Intercom has been integrating AI in its products for several years, since GPT-3’s release, they have started radically changing many aspects of the organisation.

This analysis provides a glimpse into how Intercom is positioning itself as an AI company, by focusing on 4 key areas:

Product
Pricing
People
Content

Here’s a brief timeline of their AI adoption:

December 2017

Fergal, a machine learning strategist turned VP of AI at Intercom, spoke about the product implications of AI, noting:

Simulating natural human conversation is a challenge. We’re definitely not at the stage where we have a system that’s intelligent and can hold the context of a conversation.

Creating a machine learning system that’s fit for users continues to be tough. Often you can build a very powerful machine learning system that will do 90% of a task well enough, but then you’re left with this remaining 10% that prevents it from entering the wild.

October 2020

Resolution Bot was born.

We had seen the magic of a couple of real end users having their questions successfully answered by a bot, and we knew time spent improving accuracy with real machine learning wouldn’t be wasted

December 9th, 2022

Des and Fergal discuss ChatGPT (launched on 30th November) and its implications for customer support.

Our technology interfaces are gradually becoming more conversational, and we’re just starting to see the quality of natural language understanding get good enough to unlock them.

January 31st, 2023

Intercom unveils its first experiment with generative AI—powering Intercom Inbox with GPT 3.5.

We sketched out a few AI-powered features we thought could be useful, went into production, and put a beta version in front of 160 customers.

The ability of GPT-3.5 to edit and change text makes it very valuable for customer service, and it can already handle tasks such as summarizing text and adjusting tone.

we built a feature and did a couple of rounds of iteration with a summarization feature in the inbox. You could just press a button or use a keyboard shortcut to basically say, “Hey, I want a summary of this feature, put it on my composer so I can lightly add to it.” It’s not perfect. You might need to add a little bit to it, but it’s a huge time saver.

It normally takes 3 minutes to write a summary, but with this new feature, it takes ~10 seconds.

February 2nd, 2023

Fergal and Des hosted another podcast discussing ChatGPT's impact on their industry and its benefits.

GPT may disrupt the customer service industry, but if automation increases the agent’s productivity, it can ultimately unlock capabilities that enhance their value to the business.

They mentioned they were prototyping some new big features.

we have a wave of other features in prototype form that are not quite there yet – big-ticket value things – but we think we see a line of sight to that

March 14th, 2023

Fin was announced.

The big goal, though, was creating a GPT-powered chatbot that could answer customer queries directly. To do this, it needed to be able to harness the power of large language models but without the drawbacks posed by “hallucinations”. Initially, we weren’t sure how long it would take to crack this problem, but now, with the release of GPT-4 by OpenAI, we can reveal that we’ve built a chatbot that can reliably answer customer questions to a high standard. We’ve called it Fin.

We’ve reduced hallucinations by about 10x, building constraints that limit Fin to queries relating to your business, based on a knowledge base you trust.

April 2023 onwards

They produced more content about the impact of AI on customer support roles.

Yes, customer service teams will get smaller. This is inevitable, but it isn’t the doom and gloom people project. Many teams have high staff turnover, and constantly open roles, and this will reduce pressure. It will give managers more time to invest in growing their people.

The customer service job will get much more interesting, and rewarding. Reps will only deal with more complicated questions, or really important customers. The job will still be about human connection, empathy, but also deep problem solving. Reps will have more time to provide the excellent customer service they are striving for.

New types of jobs will emerge in the customer service team. The customer experience that blends AI and reps needs to be designed, to be orchestrated. And it needs to be analysed and improved.

They produced a report on AI trends in the support industry (and have since released a 2024 version)

67% of leaders plan to invest more in AI the year ahead
71% of support leaders believe that customers will expect AI-assisted customer service in the next five years
71% of customer support leaders believe AI gives companies a competitive advantage

Since then, they’ve released a lot of content on building AI products, AI's impact on customer support and its benefits, and industry surveys for insights.

They’ve cemented themselves as an AI-first product company, sharing their learnings and helping their industry.

Let’s see how they’ve done it by focusing on product, pricing, people, and content.

Product

Intercom has been incorporating AI into its products for years.

Their ‘Resolution Bot’, launched in 2020, resolved support queries by using neural networks to detect topics in the Inbox, making support reps more efficient.

In June 2022, they were exploring use cases for generative AI with ‘Smart Replies’—which provided predictive text for support reps to use when talking to customers.

Finding an area that is relatively exploitable by generative AI, but also provides room for imperfection, is the challenge. That’s why our first foray into generative AI was on the agent assist side (Smart Replies). If it didn't work, there was zero downside. If it did work, there was significant upside.

Des - source

They were building the next-gen version of Smart Replies run on in-house models when ChatGPT launched.

“ChatGPT, and text-davinci-003, completely blew us away in terms of quality, so we instantly started to focus on these models, rather than our internal models,”

Fergal, VP of AI - source

So they pivoted.

They used OpenAI’s models to build a bot that talked to end users. This was the start of their AI support bot, Fin.

What's Fin?

Fin is an AI chatbot designed to enhance customer support by resolving up to 50% of customer queries instantly. It's powered by OpenAI's GPT-4 and Intercom's AI tech. It can hold support conversations in multiple languages and sends complex issues to human support teams when necessary.

It can be customised to fit a brand’s identity and choose the information sources, such as your knowledge base or help centre, to increase accuracy and reduce hallucinations.

You can monitor user conversations and see where gaps in your knowledge base are to provide clearer documentation for customers (and Fin).

Fin has responded to over 2 million customer requests.

Intercom has been in support for many years and deeply understands its product(s). This has helped them see the extent to which AI can change it.

After the 2015 chatbot spike on Facebook Messenger’s chatbot platform, it was clear chat wasn't the future interface. Now, the question resurfaces.

“Enterprise software has a history of clunky UIs; one massive question we’ll be answering in the industry over the next two years is: Is text the new UI? And is voice the new UI?” - source

Learning tool interfaces is an ongoing issue that can be solved with the new-era chat interface—as long as you ask the right questions.

Plus voice, real-time audio translation and transcription allow natural language interaction with software.

Multi-modality allows interaction with software through text, voice, images, and video.

It'll be interesting to see how Intercom's thinking translates into their future product experiences.

People

Creating a Skunk Works team

Creating small, dedicated themes is common with companies encouraging employee adoption of AI, which I wrote about here.

Driven mostly by the ML team, we deliberately ran a very small operation with some product and beta support from outside. This reduced overhead of internal communications in such tight timeframes. As the technology was so new, we didn’t want to waste any precious time with a large group ramping up and trying to get on the same page.

Intercom has a culture around shipping fast and learning, which is a major asset in times of technological disruption. Initially, we set out to build a series of features that required either thin or medium depth integrations, so we could deploy them quickly, but that would still deliver real value.

Fergal - source

Two launches later, more teams are surrounding AI product launches.

Adopting AI internally

Intercom has pushed AI adoption across the organisation.

They are creating internal education for engineers to get up to speed on using AI at work, hosting chats, demos, and presentations, and purchasing ChatGPT licences for employees.

“Every software company will need significant expertise in AI and ML to identify the best opportunities specific to every business and to know how to explore and exploit. I don't think it's realistic for every engineer in a tech company to overnight become an AI and ML expert. They will probably all become very familiar with the various APIs that are out there, but I think the research and exploration around opportunities is where there is the most strategic opportunity.”

Des - source

Here’s how Intercom breaks down the ML vs. Product and Engineering teams:

source

Intercom stresses the importance of ML teams having curiosity, applying research to different contexts, and thinking like a customer.

An ML team must be very deep in the culture of R&D, meaning they are used to working with new ambiguous technology, but also care deeply about solving customer problems, and moving and shipping extremely fast. In 2022, we were already deeply working on the relevant customer problems, trying to use other technologies, and so when GPT-3.5 came along, we were really well positioned to move rapidly.

Des - source

The company’s vision is essential for team success as AI drives technological progress. Alignment with the vision created internal excitement and momentum.

Pricing

Pricing in AI companies varies from pay-per-token, traditional subscriptions, to bringing your own API key.

The impact on larger companies is interesting.

Companies like OpenAI charge a fraction of a dollar for a certain amount of tokens, but it can become very expensive when you compound that across thousands of customers and conversations, plus the conversation length.

Early experiments with a controlled number of customers are helpful. You need enough activity to determine the value, but you don't want to overexpose yourself to excruciating costs.

No matter the use case, builders should initially focus on an asymmetric opportunity. And whatever you select, remember that there needs to be a somewhat regular and active use case. With low frequency, you’ll never get a sense if it’s working or not. - source

Once you understand the numbers and cost per user, you can implement different pricing strategies.

Once you got off their ‘startup-friendly plan’, Intercom was expensive. It was a big pain point for young companies who couldn’t afford $500/mo for a support widget—so other players came in to undercut them.

This pricing screenshot is from September 2023, $74/mo for the starter plan.

Intercom now has a lower monthly base subscription fee ($39/mo, $99/mo, $139/mo) and Fin access in all tiers (you only get Fin workflows for the $99 and $139 plans). The big difference is that Fin is charged at $0.99 per resolution.

What’s that?!

This is how Intercom describes it:

Fin usage is measured in resolutions. This ensures that you only pay when Fin does what you care about most – resolving a customer’s question. A resolution is counted when, following the last AI Answer or Custom Answer in a conversation, the customer confirms the answer provided is satisfactory or exits the conversation without requesting further assistance.

Basically, it’s resolving a user's query.

This approach is interesting. Some end users' support queries could cost over $0.99 to resolve, but most are likely well under $0.99 (simple vs. longer questions). The API calls to OpenAI cost a fraction of that, making the pricing work.

Other companies are doing something similar. Zapier has re-jigged their pricing to include a base fee plus a fee for successful tasks.

A task is an action your automated workflow successfully completes. For example, if your Zap has an action to create new Google Contacts, each contact Zapier creates will count as one task.

Content

Intercom has increased content production around AI. They host fireside chats and presentations internally.

But externally, they’ve done a lot.

They’ve hosted podcasts discussing incorporating AI into their products, its place in customer support, impact on support reps, and limitations.

This discussion is likely happening in other companies too, but publishing it externally helps understand their careful consideration of the new tech's implications.

They don’t shy away from talking about the lows and highs. (example)

It educates other customer support leaders on expectations of AI, how their teams may be thinking about it—”do we lose our jobs?”—and how to roll out AI to their teams.

Des has appeared on several podcasts, including 20VC, discussing the implications of AI for Intercom and the support ecosystem, and how the next wave of tech companies should approach building with AI.

Des argues that AI should enable companies to do things that were not possible before, rather than just automating existing workflows.

the disruption comes when actually you would build the entire thing differently in a world post-AI

whoever has the differentiation that no one else can get is the person who can actually charge the highest price".

He distinguishes between a "thin wrapper" on top of an AI provider and a "thick wrapper" which "solve[s] the user's problem end to end fully in a way OpenAI never will".

the areas where AI will deliver most value in startups will be things that just weren't possible before are now possible and I think like you'll see companies like Synthesia will produce massive new things that just literally weren't being done before and they're like new categories and new jobs, new workflows, new capabilities.

They also hosted podcasts with guests to talk about various topics, such as:

This content positively impacts current and potential customers and the broader tech ecosystem.

It again highlights Intercom’s shift to an AI-first company.

Takeaways

To truly shift your organisation to an AI-first one, you need to think about Product, Pricing, People and Content.

Where in your product does AI help the customer?
Where in your workflows does AI help you serve end users?
How can your pricing change to align your business goals with customer outcomes?
Can your pricing be more beneficial to a user?
How are your team using AI internally to help them be more productive?
Does the team feel educated on what AI can do for business operations?
What content can we produce to show our commitment?
Can we share learnings that our customers can benefit from?
How do we earn trust by showing a peek behind the curtain of how we’re building with AI?

‍

Yikes, a paywall!

‍70+ tutorials, courses and case studies wait behind it. No subscription, $150 paid once.

✅ Full course & tutorial access
✅ Case studies on companies using AI
✅ Private community access
✅ No subscription, $150 paid once
✅ Expense it using this template. Or get a team account.

Get all access ->