On the effects of AI on the enterprise software industry (Part I: The Demand Side)
A AI-motivated follow-up to The Never-Ending Cycle of Disruptive Innovation in Enterprise Software
This is Part I of a series of two (?) posts I'm writing on how the latest advancements in AI technology will affect the enterprise software market. Here, I'm focusing on the demand-side of the discussion, i.e., on how AI will affect customer expectations and on how incumbents x new entrants will be able to fulfill these new expectations.
How will the latest advancements in AI technology affect the enterprise software market?
We are still in the earliest innings of understanding how these new technologies will affect enterprise software. Nonetheless, I'd like to take a stab and make some predictions and provoke a few avenues of further investigation.
I'm especially interested in looking at enterprise software market's competitive dynamics as it relates to the ongoing fight between incumbents and new entrants. We can dive deeper through the following angles: a) understanding how these said advancements will affect the disruption dynamics that are very present in the enterprise software space and b) understanding how access to data will affect who wins a given battle between incumbents and new-entrants. I also propose a few questions that may help us assess the potential effects of AI in a company's future prospects.
The Never-Ending Cycle of Disruptive Innovation in Enterprise Software
Enterprise software applications, from the bedrock (CRMs, ERPs, HCM suites such as those made by Salesforce, Oracle, et al.,) up to the more mission-oriented systems (such as those made by early Hubspot, Segment, Resend) are subject to Never-Ending Cycle of Disruptive Innovation in Enterprise Software, a textbook application of Clayton Christensen's Disruptive Innovation theory.
In most markets, customers' abilities to absorb (i.e., use and value to the point of paying more for) product improvements grows at a slower pace than the product improves [1]. This causes segment after segment of customers (from least needy to most needy) to at some point become over-served.
As I wrote in the above-linked article [2]:
Enterprise software is an incredibly fertile ground for disruptive innovation. Let’s take the example of Salesforce to illustrate that assertion.
Salesforce’s CRM offering goes much further than what a “simple” CRM is expected to do: aside from storing prospective customer information and allowing for a pipeline of deals to be managed, it has all sorts of bells and whistles, such as industry-specific reports, complex workflow automation, AI [3], an app store, and LOTS of flexibility that cater to all different sophisticated customer needs; it costs a lot and takes long to be implemented. And Salesforce keeps adding incremental sustaining innovations to the product every year and causing a bigger and bigger portion of the market to become over-served.
In Christensen's theory, his over-serving happens because incumbents tend to focus on the needs of their highest-end customers - i.e., the customer segments that are willing to pay the most - when prioritizing their roadmaps. The highest-enders not only pay more but also demand more: they are the most demanding in their needs for product improvements, generally along the lines of more and more bending of the software to their specific idiosyncrasies. That inevitably creates more complexity and thus makes the software harder to use - and eventually more expensive - for all the lower-than-the-highest-end customers, who'd be increasingly happy, if we were to follow along the highest-to-lowest-sorted list of customers, with a simpler, easier to implement, cheaper solution if only one were available.
In sum, we could think of it all like this: With every new feature Salesforce introduces (i.e., a product “improvement” through the eyes of, e.g., Nike, which is happy to pay more for those “improvements”), a new cohort of customers (e.g. some hypothetical John Doe & Sons agricultural implement retailer from Iowa) falls over into the over served bucket, because it gets incrementally frustrated with the pricing and/or complexity of the product, and thus becomes fertile ground for Pipedrive's entrance into the market.
Prediction I: LLMs will delay the onset of disruption, favoring incumbents
So how, if at all, will evolving AI infrastructure (LLMs becoming ubiquitous, available via easy to use APIs, etc. etc.) affect this cycle?
A new layer of AI capabilities (e.g., Microsoft's Copilot in case of Microsoft Office customers) is a sustaining innovation for incumbent enterprise software offerings; that is, one which improves the product in a dimension important to said company's most demanding customers.
These dimensions are things like flexibility of customization, convenience of adoption, ease of use, etc.
My tentative thesis is that AI will add a new vector of product characteristics aligned with the current vectors in place, so as to in practice bump up every customer segment's innovation adoption curve and thus delay the onset of "overservedness" along this line (the highest-to-lowest-end), therefore delaying the inevitable onset of disruption-by-disruptive-innovation by potential new entrants. AI could slow the cycle of disruption by introducing a new dose of functionality | convenience | customization that's relevant to these customers. No large feat of imagination is necessary to post that AI could help products gain new functionality.
My tentative thesis is that LLMs will favor incumbents by delaying the onset of disruptive innovation.
If that's the case, products that were about to be disrupted could gain new leeway. From another angle, segments of customers that were about to become over-served could take longer to become so.
This could mean AI is potentially disproportionately advantageous to incumbents as opposed to new entrants. But not all incumbents benefit equally.
Data seems to be key
Another interesting angle to look at this problem is data. Many argue that AI models are going to become commoditized. If that's so, the logical conclusion is that those who hold differentiated data corpora will be the able to leverage baseline models and offer AI functionality that's differentiated.
…we could think of it all like: With every new feature Salesforce introduces (i.e., a product “improvement” through the eyes of, e.g., Nike, which is happy to pay more for those “improvements”), a new cohort of customers (e.g. some hypothetical John Doe & Sons agricultural implement retailer from Iowa) falls over into the over served bucket, because it gets incrementally frustrated with the pricing and/or complexity of the product, and thus becomes fertile ground for Pipedrive's entrance into the market.
For an enterprise (customer) the most differentiated data corpus is its own (sitting on Workday, Salesforce, Workday, Oracle, SAP, Hubspot, etc.). Therefore, those same business applications holding proprietary company data are best positioned to benefit.
Oracle's ERP, for example, has a company's history of financial data, and most probably its competitors'. It could build on top of open-source LLMs by training them on company-specific data, therefore being able to offer highly unique AI solutions to its customers. No rocket science here.
They could, thus, offer AI capabilities that are hard for anyone else to replicate. The question then becomes “will AI capabilities become key for solving customers’ jobs-to-be-done?” That's still an open question.
Other types of data are also important, although a bit less. One type is industry-specific data (for example, that held by Bloomberg). There're players that house both proprietary and industry-specific data, such as Workday's Peakon product - and Qulture.Rocks’ climate/engagement surveys product - which allow customers to measure their own historic employee engagement data while also comparing said data to that of an anonymized group of relevant industry peers’).
Delayed onset - but onset nonetheless
Even though I think there's a significant chance that AI will delay the onset of disruption for most enterprise software incumbents because, in part, data, I still think the Never-Ending Cycle will go on for the foreseeable future.
On the shorter term, the most over-served customers are still very unhappy, and will jump to simpler, cheaper solutions if they become available, even if their AI capabilities are still nascent.
On the longer term, it will most likely happen at scale when new entrants, especially startups, figure out an AI first way to leverage these new capabilities. A parallel is ads [4]: when the first newspapers got online, they served ads on a webpage just like they served ads in print: banners that would be placed on top of or besides articles. Then mobile happened, and Facebook figured out an entirely new way to place ads in front of customers: along the news feed. A new paradigm that was optimized for the new reality of mobile. The same will eventually happen to AI and enterprise software: today we're layering AI on top of an 'old' software paradigm. What will the AI-first enterprise software paradigm be? Conversational?
A framework to evaluate the impact of AI
Building on top of the work of Sills, B. et al (2023) [5], I've given their framework my own spin. For every positive-effect question, add the stated number of points to the company's scorecard. Alternatively, for every negative-effect question, subtract the stated number of points. Add the points in the end for an overall AI-impact score, the greater the more positively affected the company should be.
Positive-effect questions:
Does the company's products benefit from network effects, especially in relation to data? (+ 4 points)
Does the company's products house an important corpus of customers' own data? (+ 2 points)
Does the company's products house an important corpus of customer's industry-specific data? (+ 2 points)
Can AI capabilities expand the customer's TAM, or accelerate TAM penetration? (+ 1 point)
Do the use cases and end-customer bases of the company’s offerings have more stringent requirements in terms of data governance and compliance? (+ 1 point)
Negative-effect questions:
Does AI have the potential to reduce the company’s total addressable market via commoditization? (- 4 points)
Does AI have the potential to reduce the company’s total addressable market via seat count compression? (- 3 points)
Notes
[1] According to Christensen, product and service improvements come from three core dimensions: functionality, reliability, and ease of use (ease of use being itself composed of convenience, customization and price).
[2] The core of the article can be summarized by the following sonnet, written by our friend ChatGPT:
In enterprise domains, the giants stand tall,
Their products refined, answering the call.
Sustaining innovations fuel their might,
While in their shadow, disruptors take flight.
A simpler, cheaper product takes the stage,
For overshot customers, it's all the rage.
The giants, they scoff, ignore this new threat,
In high-end markets, their sights are firmly set.
But the disruptor, it does not stand still,
With steady improvements, it starts to fill,
The market gaps left by the giants' retreat,
Its growth in power is no small feat.
"Rationality" the giants' doom does seal,
And thus, disruption turns the enterprise wheel.
[3] At least when this was written, Salesforce's AI offering was very simple, so I don't think it counts as “AI” in the sense of current advancements in LLMs, etc. etc.
[4] This analogy is still very poorly structured, but serves to illustrate the point.
[5] “Navigating the next frontier of enterprise software – AI Primer” by Brad Sills and team at Bank of America Securities.