How the AI wave could impact the B2B software industry in the next decade

Last year I explained in a video, embedded below, that I believed the B2B software industry was entering a new stage: the SaaS wave was now in its deployment phase, and at the same time, the next big innovation wave, driven by AI, was in its installation phase.

In the past months I’ve structured my thoughts about this topic, and in this post, I will try to expose how I think this AI wave will develop in the next years.

If you’d like to be notified of our next posts you can subscribe to our newsletter.

As a side note, the “bigger picture” is the B2B software industry here. In my article “SaaS” is not to be understood as an industry, but rather as an innovation wave (product and business model innovation) which impacted the B2B software industry in the 2000s and came after the “on-premise” wave.

1. Comparing the installation phase for the SaaS and AI waves

1.1 Challenges faced by startups during the SaaS installation phase (2000–2010)

When you look at the first generation of successful “SaaS first” companies (Salesforce, Zendesk, Workday, Hubspot…), they had to overcome three main challenges to succeed:

  1. Market education
  2. Infrastructure
  3. UI/UX

Market education. A major challenge to overcome during this early phase is market education. The first SaaS companies had to educate customers on the value of the SaaS model, convince them to switch from their on-premise software, and finally to change their habits in terms of software consumption (you subscribe to a service instead of owning a software) and usage (you use the software in your browser).

Infrastructure. If building the infrastructure of a SaaS product seems easy nowadays (with AWS and the myriad of developer APIs available), it was not the case fifteen years ago. Providing a “cloud based” service was really challenging in terms of infrastructure, and the first SaaS startups had to hire large engineering teams dedicated to this aspect only. It was clearly a huge barrier to entry as not every startup could build and maintain such infrastructure (capital intensive + scarcity of talents).

UI/UX. It also seems normal nowadays, but the first SaaS companies had to pioneer new UI/UX back in the days. Traditional software ran on desktop applications and offered complex interfaces. SaaS startups had to explore new interfaces and user experiences that would fit the web browser. It was a total product paradigm shift, and it literally took years of iterations before we got to the polished interfaces we have today.

To sum it up, I believe the winners of this “installation phase” were the companies which managed to overcome these three challenges and which have picked the right early “use case” (sales pipeline management for Salesforce, content marketing for Hubspot, HR for Workdays…). The timing aspect is crucial. Many SaaS companies were “too early” and failed because of the market timing.

Now let’s explore how this installation phase looks like for “AI first” companies.

1.2 Challenges currently faced by “AI first” companies during the AI installation phase (nowadays)

I think that “AI first” companies currently face the exact same problems. I also believe that the winners of this generation will be the startups which will pick on the relevant “use cases” (market timing), and will overcome the market education, infrastructure and UI/UX challenges.

Market education. Whether if it’s replacement or new solutions (AI products which make something better versus the ones which solve a new problem), “AI first” startups have to conduct market education. They have to convince customers of the benefit of this new generation of products, especially that in many cases the “AI” aspect is bullshit, to be honest. Many customers are still dubious and need to be convinced of the real benefit of AI products. A telling sign is when you see that “AI” is used as a value proposition by itself. It was also the case for SaaS at the beginning, SaaS (“as a Service” and available in a browser) was a value proposition by itself, displayed on the homepage of every startup. Now it has disappeared. The SaaS model is accepted by everyone, it’s not a differentiator anymore. The same thing will happen to the term “AI”.

Infrastructure. Similar to what happened for the SaaS wave, AI first companies currently need to hire AI experts to build their product, which is capital intensive and difficult in terms of recruiting (a lot of competition for these talents). Until “AI infrastructure” tools become mature and widely available, this aspect will stay a huge barrier to entry for many startups.

UI/UX. This aspect is probably a bit underestimated, but I think that we’re still at the infancy of native AI interfaces. Startups still need to figure out how to craft products that leverage, for example, voice or image recognition the best way. Chatbots are an excellent example of these struggles. There’s no doubt that in the future we’ll have more bot assistants, but we still need to figure out how to create a great user experience that goes beyond the simple “Text chatbot” adapted from the “human support era”. Same with the voice assistants invading our homes (Amazon Echo…), we’re still at the beginning in terms of what “voice interfaces” can deliver. A lot of innovation needs to happen here.

2. Analyzing the deployment phase for the SaaS wave and drawing hypothesis for the future of AI

2.1 Challenges faced by SaaS companies during the SaaS deployment phase (2013 — now)

Now that the SaaS wave is well into its deployment phase, let’s analyze the challenges currently faced by SaaS companies.

In my opinion the three challenges covered above have transitioned to:

  • Market education challenge -> distribution challenge.
  • UI/UX challenge -> branding challenge.
  • Infrastructure challenge -> integration challenge.

Distribution. Now that the SaaS model is widely understood and accepted, there’s no need to conduct market education anymore (except in very specific industries, but even there, the SaaS model is accepted, it’s more about changing habits). The challenges for many SaaS startups is to distribute their product in crowded channels. It’s not about convincing customers about the benefit of the SaaS model, but how to reach them and to push them to switch from their existing SaaS products :-).

Branding. Let’s be honest, it has become tough to innovate on UI/UX. We’ve come to a point where many of the best practices are shared across startups, and as soon as a product comes with something new in terms of UI, it gets copied fast. The bottom line is, it’s harder and harder to differentiate purely on UI/UX with a traditional SaaS product. The battle has since moved to branding. In a crowded environment with less differentiated products, consumers will go toward strong brands they like. When you look at the recent SaaS “successes” many are stronger with branding than with pure product innovation. Even startups with great products need to build a great brand to break through (see, for example, Superhuman).

Integration. A major difference between the installation and the deployment phase, is the abundance of infrastructure tools available. From web hosting (AWS) to email (Sendgrid) or search (Algolia), there’s no shortage of infrastructure tools available to build a SaaS. What took teams of high level engineers to build and manage this infrastructure now takes a couple of well-trained developers. Infrastructure is not a barrier to entry in most cases (of course you have exceptions). If I had to pick a current tech challenge, it would be the increasing number of integrations SaaS startups have to build and maintain.

To sum up, these three challenges explain why many of the recent fast-growing SaaS companies are the ones which are outstanding with distribution and branding, and why Salesforce, Zendesk and the other first generation winners are still locking the market. They basically have locked distribution by becoming platforms (Salesforce, Zendesk etc… all offer their SaaS app store), managed to build powerful brands (which is hard to copy), and own integration as they have become SaaS hubs.

2.2 Implication for AI first startups in the next years

We’re now entering the “prospective” part of the article :-) Based on the evolution I depicted above, this is how the AI wave could play out in a couple of years (> 6–7 years).

No more market education. As I mentioned already, I distinguish AI solutions that solve an existing problem multiple times faster / better than current software (ex tools that analyze radiographs faster than doctors), and AI products which enable new use cases that were not possible before (ex software to estimate fish population in aquafarms). In both cases, I think it will be clearer what the real early “winner” use cases are, second that we won’t speak about “AI based” solution anymore (the value proposition will focus on the solution and not on the technology — AI), and finally that distribution will be more important than market education in the go-to-market strategy of AI first companies.

Native AI UI/UX. I also expect huge improvements in terms of UI/UX. Founders will create plenty of new ways for us to interact with software whether it’s through voice, movement, typing etc… (and don’t ask me what it will be, if I knew I wouldn’t be here ;-)).

Infrastructure. Probably the most “predictable” aspect of the three. I don’t doubt that in some years developers will be able to use infrastructure services making it easy to integrate real AI in their product. The consequences of this democratization of AI tools will be massive as it will lower the barrier to entry for many founders (they won’t need to hire ten AI specialists) who’ll be able to explore plenty of new use cases and create niche AI applications.

3. Various implications

3.1 During the AI installation phase, data moats are not crucial for success.

Ok, this is probably my most controversial point and is not shared by my colleagues at Point Nine :-). Since the major barriers to entry, according to me, are picking the right use case (market timing), and overcoming the market education, infrastructure, and UI/UX challenges, I believe that data moats are not necessary to build a great AI company in this first phase. Data moats and proprietary data will become key differentiators in the deployment phase when it will be relatively easy to build an AI product. At the moment it’s so hard to build a great team and make a great AI product that delivers real value to customers that data moats are of secondary importance. But don’t say what I didn’t say, if you have a great use case AND data defensibility, it’s jackpot.

Very happy to be contradicted here, please let me know in the comments.

3.2 Building an AI infrastructure company is not necessarily the easiest approach.

It’s during the SaaS installation phase that many of the core infrastructure services were built. The Twilio, Amazon Web Services, Github, Heroku, and other Sendgrid were all started during the SaaS installation phase. That being said, many of these services became “commodities” offered by the major cloud players (Amazon, Google, and Microsoft). Few still operate as independent companies as they faced the competition of players which have the economy of scale on their side. I think it’ll be the same for the AI infrastructure space.

3.3 AI/ML is not a single technology, but a multitude of different technologies with their specificity.

Throughout the whole post, I speak about “AI” as a general innovation wave. I’m well aware that it’s a simplification, and in fact, it covers very different technologies (ex: voice, image or text recognition technologies) which require different infrastructure services and UI/UX best practice before they can be democratized. These technologies have different use cases and will probably develop at different speeds, so be aware of that.