Defining the generative AI ecosystem

Venture capital firm Sequoia has put the most definitive and well-publicized stake in the ground that we’ve seen to date around creating the category that is generative AI. Sequoia put out a living and breathing market map for the generative AI landscape, which can be found here.

Sequoia organizes the ecosystem, which includes 50-plus vendors, around seven pillar categories: text, video, image, code, 3D, speech and other. Each of these categories has between two and six subcategories. Overall, by the way, this is an encouraging start for generative AI. This map may be a little feature-centric, but it’s encouraging to see there are actual use cases forming, particularly in the subcategories. For example, within “code” there are subcategories for code generation, documentation, web app builders and text to SQL. These are actually real jobs that a customer might be looking to get done! Side note: If I were Sequoia, I might think about using these output/outcome-focused subcategories as the pillars, rather than using the technology (e.g., text versus video) as the organizing hierarchy. You could envision a future where there are speech tools that help with code generation, for example.

This market map isn’t fully representative of this space, but it’s as good of a place to start as any for tracking monetization in this evolving ecosystem. In the sections that follow, we share our impressions of the current state and possible future state of generative AI monetization based on a scan of the players in the space from across Sequoia’s market map.

There’s one big caveat here that we think also bears watching: the Sequoia map and the discourse on generative AI overall are currently focusing largely on stand-alone startups that are offering technology in this space. That makes sense. But we think a much bigger evolution will occur when larger, more well-established companies start embedding these technologies into their existing SaaS products, via either their own development efforts or acquisition. That’s a whole other lens through which to evaluate packaging and pricing evolution and monetization strategy. It’s probably a little early for tracking that dimension, so we’ll focus on the stand-alone players here and cover embedded use cases in a future post.

Generative AI packaging models

Turns out that good-better-best is, well, best — at least when it comes to getting started with a packaging model for generative AI.

Most of the players in this space across categories offer between two and four product tiers. Many offer a free forever plan or a low-priced paid plan as an entry point offer, and then offer a standard single user/baseline plan and one or two more premium plans, typically targeting teams and/or enterprises with variable usage requirements.

Usage of products is gated to create a fence encouraging customers to migrate from free or entry-level to premium paid plans. Usage gates vary based on the category and vendor but typically follow similar constructs as we’ve observed in SaaS overall. Usage is gated most aggressively for free plans, and then usage limits are lifted as the customer migrates to paid plans. Usage is generally defined based on three to five key value metrics across plans.

Overall, it seems the generative AI ecosystem is drafting off the successes of the tried-and-true good-better-best model for SaaS. Packaging appears to be primarily designed to cater to personas based on size, scaling from free and individual prosumer plans to team and enterprise plans for broader departmental or business usage.

There’s nothing wrong with the approaches here — they are recognizable, easy for customers to understand, and clearly presented. We do think, however, that packaging in this ecosystem reflects the ecosystem’s maturity. Most of these companies are building products that are primarily aimed at attracting and engaging generalist early adopter users, who may be trialing these products for their personal side hustle projects and/or for business purposes, or both. The predominant strategy appears to be focused on activating and engaging these prosumer users, and then offering plans that make it easy for these users to recommend these products to their teams for broader business usage.

Generative AI pricing models

Pricing models used in generative AI are aligned to the most common current constructs used by SaaS companies overall. There is greater variability in choice of pricing model versus choice of packaging model across the landscape, with some trends seeming to emerge within each generative AI subcategory.

Across categories, most companies are using monthly and annual subscription models with prices that are based on a flat fee or seats, with usage limits defined by plan. For some companies that don’t price based on users, these usage limits also include a cap on the number of users for a given plan. Pure usage-based pricing is employed by some vendors, although it is not common.

Here are a few examples of the predominant models we’re seeing emerge for pricing generative AI:

• Flat-fee subscription for a defined amount of usage with a clear value metric: This appears to be the de facto model used for generative AI text and writing tools, which is the most mature category in the landscape. Nearly all vendors set a monthly price that is based on the number of words generated by the tool. This typically includes a cap on the maximum monthly words as well as a packaging cap on the number of users. To upgrade for more users or more words, customers must move to the next plan tier.
• Seat-based pricing with usage-based tiers: This model is used more sporadically by certain vendors. For example, GitHub’s Copilot product is priced per user per month, and each plan sets limits on other value metrics such as CI/CD minutes and storage. Similarly, code completion AI assistant Tabnine prices plans per user per month. AI writing assistant Moonbeam prices at $15 per user per month for unlimited access to all its current features. Sales email writing assistant Lavender uses this same model. Several others in the space don’t directly price per user but set limits on the number of users for different plans. As with SaaS more broadly, this model is best for products targeting individuals and/or for products where each incremental user gets differentiated value from using the product.

Generative AI and the evolution to “do it for you” software

We’ve written before about the idea of “do it for you” in SaaS. We first heard of the concept (no surprise) when ProfitWell was acquired by Paddle, and we fully agree with them that “do it for you” will dictate the next major wave of SaaS innovation.

SaaS proliferation within businesses small and large has created tool overload. Many tools were adopted for the sake of having a tool, and this resulted in SaaS sprawl, shadow buying and runaway software spending overall. As companies pivot to contain costs, they’ll rationalize their software footprint and focus more acutely on software that actually makes their lives easier. In fact, this is already happening in response to the uncertain economic outlook for 2023. The software that has staying power will deliver measurable and tangible value by helping complete a job-to-be-done in a way that creates new revenue, secures existing revenue, reduces costs and/or reduces risk.

We’ve hypothesized this market evolution to “do it for you” may have a corresponding impact on how vendors that succeed in that paradigm choose to monetize software. “Do it for you” suggests that the software delivers at minimum a successful output (example: invoices are processed) and, better yet, a successful outcome (invoices are processed with 100% accuracy and 100% timeliness).

In a “do it for you” world, this means you have two options as a vendor: link monetization to the output you deliver (plan XYZ supports up to 100 invoices processed monthly and is priced at a flat fee of $100 or is priced $1 per invoice processed), or link monetization to the outcome you deliver (plan XYZ is priced at X% of invoice volume for invoices collected on or before term).

Here’s where generative AI comes in. We believe generative AI has potential to be a killer app that helps software evolve to actually deliver outputs or, even better, outcomes. Why? That’s the core premise of generative AI as a capability — to generate novel assets, be it text, code, audio or something else, for some defined purpose. Perhaps in some instances today these products are generating assets for generation’s sake (we haven’t done a deep dive feature test for diligence on this, so it’s more of a hypothesis), but the idea is that they’ll evolve over time to tighten focus on target customers and use cases.

It would stand to reason, then, that with that evolution will come monetization models that focus on outputs and outcomes. Stand-alone generative AI software companies certainly have this opportunity but so will other software players that are aiming to embed generative AI to add features and/or services to solve customer problems.

Generative AI is already mostly in “output-based monetization” world today. Text-based platforms such as Copy.ai and Jasper are pricing based on the number of words delivered by their tools on a monthly basis. AP automation AI platform Glean packages based on the number of processed invoices.

If we agree on the “do it for you” trajectory and believe generative AI will be a key conduit for that, how will monetization in the space evolve? Here are a few predictions:

• Per user models will go away over time.
• Packaging will mature before pricing models; as customer personas and use cases are more clearly defined, packages and entitlements will evolve to better align to those needs.
• Output-based models will remain but will mature along with use case maturity and definition – for example, an output of words will evolve to an output of blog posts or social posts or other longer-form writing.
• Outcome-based models are a possibility (e.g., blog posts with XYZ impressions or other engagement) and will emerge for some use cases but be used sparingly. Vendors and customers will find it challenging to define success criteria to support outcome-based models.
• Some of the most compelling packaging and pricing use cases will emerge with generative AI as a feature or product enabler, not as a stand-alone product. Large and established software companies will use generative AI capabilities to bolster how current products deliver outcomes. For example, where you previously used HubSpot to automate the process-oriented elements of your email marketing such as managing lists, creating templates, scheduling sends and tracking results, perhaps HubSpot will embed generative AI to completely take over your end-to-end email marketing by building campaigns, authoring content and self-optimizing based on results.
• These evolutions will make productized services and/or software with a service a popular model for many stand-alone generative AI and other software companies. The combined value proposition of software plus services will enable companies to deliver outcomes. For example, a content marketing agency may choose to employ a smaller writing team and leverage a generative AI writing platform to power its baseline content production. That agency may price to the client based on successful blog posts completed, even if the underlying AI tool prices based on words.

We’re looking forward to watching how this space evolves and how quickly monetization (not venture funding) strategies help turn these technologies into sustainable businesses. We’ll be watching and analyzing as we go, as usual.

Any questions on generative AI monetization? Want to throw ideas around, profile a particular company or deep dive on how these trends might impact your business? You can find me on Twitter at @bbelangerTBR or send your feedback directly to [email protected]. Be sure as always to subscribe to get our content sent to your inbox once it publishes. Thanks for following along!