Skip to main content

AI’s Latest Victim: The Per-Seat Pricing Model

Pricing is the most important, most under-discussed element of the software industry. In the past, tech founders could get away with giving pricing short shrift under the mantra that the best product will ultimately win. No more.

In the age of AI-enabled software, pricing and product are linked; pricing fundamentally impacts usage, which fundamentally impacts product quality.

Therefore, pricing models which limit usage, like the predominant per seat per month structure, limit quality. And thus limit companies.

For the first time in 20 years, there is a compelling argument to make for changing the way that SaaS is priced. For those selling AI-enabled software, it’s time to examine new pricing models. And since AI is currently the best-funded technology in the software industry - by far - pricing could soon be changing at a number of vendors.

Why Per Seat Pricing Needs to Die in the Age of AI

Per seat pricing makes AI-based products worse. Traditionally, the functionality of software hasn’t changed with usage. Features are there whether users take advantage of them or not - your CRM doesn’t sprout new bells and whistles when more of your employees log in. It’s static software. And since it’s priced per user, a customer incurs more costs with every user for whom it’s licensed.

AI, on the other hand, is dynamic. It learns from every data point fed into it, and users are its main source of information. Usage of the product makes the product itself better. Why, then, should AI software vendors charge per user, when doing so inherently disincentivizes usage? Instead they should design pricing models that maximize product usage, and therefore, product value.

Per seat pricing hinders AI-based products from capturing value they create. AI-enabled software promises to make people and businesses far more efficient, transforming every aspect of the enterprise through personalization. We’ve seen that software tailored to the specific needs of the user has been able to command a significant premium relative to generic competitors. For example, Salesforce offers a horizontal CRM that must serve users from Fortune 100s to SMBs across every industry. Veeva, which provides a CRM optimized for the life sciences vertical, commands a subscription price many multiples higher, in large part because it has been tailored to the pharma user’s end needs.

AI-enabled software will be even more tailored to the individual context of each end user, and thus, should command an even higher price. Relying on per-seat pricing gives the buyer an easy point of comparison ($/seat is universalizable) and immediately puts the AI vendor on the defensive. Moving away from this pricing structure allows the AI vendor to avoid apples to apples comparisons and sell their product on its own unique merits. There will be some buyer education required to move to a new model, but the winners in the AI era will use these discussions to better understand and serve their customers.

Per seat pricing will ultimately cause AI vendors to cannibalize themselves. Probably the most important upsell lever software vendors have traditionally used is tying themselves to the growth of their customers. As their customers grow, the logic goes, so should the vendors’ contract (presumably because the vendor had some part in driving this growth). 

However, effective AI-based software makes workers significantly more efficient. As such, seat counts should not need to grow linearly with company growth, as they have in the era of static software. Therefore, tethering yourself to per seat pricing will make contract expansion much harder. Indeed, it could result in a world where the very success of the AI software will entail contract contraction.

How to Price Software in the Age of AI

Here are some key ideas to keep top of mind when thinking about pricing AI software:

1. Start by using ROI analysis to figure out how much to charge (but not how to charge). 

This is the same place to start as in static software land. (Check out my primer on this approach here.) Work with your customers to quantify the value your software delivers across all dimensions. A good rule of thumb is that you should capture 10-30% of the value you create. In dynamic software land, that value may actually increase over time as the product is used more and the dataset improves. It’s best to calculate ROI after the product gets to initial scale deployment within a company (not at the beginning). It’s also worth recalculating after a year or two of use and potentially adjusting pricing. Tracking traditionally consumer usage metrics like DAU/MAU becomes absolutely critical in enterprise AI, as usage is arguably the core driver of ROI.

While ROI is a good way to determine how much to charge, DO NOT use ROI as the mechanism for how to charge. Tying your pricing model directly to ROI created can cause lots of confusion and anxiety when it comes time to settle up at year-end. This can create issues with establishing causality and it sets up an unnecessarily antagonistic dynamic with the customer. Instead, use ROI as a level-setting tool and other mechanisms to determine how to arrive at specific pricing.

Identify a usage-agnostic metric that both grows as your customer grows (volume-based) and that your product impacts meaningfully and measurably (value-derived).

2. Use a volume-based pricing mechanism which doesn’t disincentivize usage.

A best practice in static software has been to price discriminate using two axis pricing models. The first being volume, traditionally set on a per seat basis, and the second being features. This two axis framework still applies in dynamic land, but the trick is to find a volume-based lever that doesn’t restrict usage.

Specifically, identify a usage-agnostic metric that both grows as your customer grows (volume-based) and that your product impacts meaningfully and measurably (value-derived). A good example of this is Textio, an AI-based augmented writing service that helps recruiters fill open roles faster. Their pricing is volume-based, tied to the number of hires the company plans to make, with unlimited usage across unlimited users. This encourages the buyer to deploy Textio to every recruiter and hiring manager and use it for every hire they make. In turn, this allows Textio to maximize data collection and, thus, product value.  

Use this metric as the denominator of your volume-based pricing axis (e.g., per person to be hired in the case above). It’s best to create buckets so that pricing falls into a reasonable denominator range (e.g., the price is $X per bucket of n to n+1,000 people to be hired) instead of using granular increments. Layer on the feature-based pricing axis, ideally positioned to mitigate any potential contraction issues with the volume axis.

3. Use pricing to improve product. 

Most AI vendors are already ingesting each customer’s historical data upon onboarding to ensure recommendations are tailored. The breakout AI vendors of the future will find ways to share insights between customers. Pricing can be a critical lever towards jumpstarting these data networks. 

Indeed, Textio used strategic discounting early on to build its proprietary data set. They offered customers the opportunity to participate in their Data Exchange program which entailed discounted pricing AND the benefit of insights from the broader network of users. Over time, the value of being a part of such networks grows and discounts may no longer be necessary.

Real estate is another industry where data networks are being built with strategic pricing. Companies like CompStak and Crexi gives brokers credits as they submit data to the system, creating a virtuous flywheel of proprietary commercial real estate data. 

One important note: In order to build the trust critical to underpin such networks, AI vendors will need very clear policies on what metadata is being shared and to invest heavily in bulletproof infosec practices. Those that succeed will have amongst the strongest competitive moats (we call them Coaching Networks) in an era in which static software continues to commoditize. 

    While it may be too early to dictate exactly how pricing for AI software should be set, I’d urge vendors to experiment with value-derived, volume-based models which don’t restrict usage. Regardless of exactly which model they land on, I believe the most successful players will start to hop off of the per seat ship before it sinks them.