Stephen Hurrell's Analyst Perspectives

Based on the number of invitations I have received to speak on the topic, usage pricing (also referred to as consumption pricing) is hot. But as I have remarked elsewhere, it is not new. I recall a conversation with a customer over 30 years ago when I was involved with analytic software who asked, “Why can’t we just pay for the software we use?” And this of course makes sense: only pay for what you use. In fact, I assert that by 2027, over one-half of all enterprises will deploy a mixed revenue model that includes subscriptions and usage pricing in addition to one-time sales as enterprises make adjustments to remain competitive.

Going back to the days of the on-premise, perpetual license business software application market, when it came to the split of risk between buyer and seller, it was really caveat emptor—the buyer carried most of the risk. Wind the clock to the turn of the century, and the advent of per-seat, by-subscription and termed contracts shifted the risk to be more equitably shared as the buyer could decline renewal if that particular purchase did not deliver the expected benefits. Wind on still further and there has been more discussion about usage pricing in the last 18 months than in the previous 18 years. I wrote a paper on subscription and usage pricing in September 2020, so what has changed since then? Generative Artificial Intelligence (GenAI) is what has changed.

The initial wave of public AI providers chose to offer free, then paid-for subscription pricing as their monetization model. More recent business offerings have proposed usage models based on individual queries. Salesforce has announced per-transaction charges for their most common agent offerings. Though some providers plan on offering AI-driven features as part of their standard applications and without charging specifically for them, I believe a key driver for the majority of pronouncements on proposed usage-based pricing for GenAI offerings is cost. Training and running models is expensive. And directly linked to this expense, at least on the execution side, is how many AI-based queries or transactions are run. So, in essence, cost-plus for usage charges is a significant share of these pricing models.

And while there are some good reasons for why usage pricing is an attractive option, it does present problems as to how it fits into the way businesses and consumers operate. A major attraction of subscription pricing was that it established a predictable revenue stream that continued until the customer cancelled or in the event of termed contracts, until the contract was not renewed. This predictability is attractive to both buyer and seller. For the buyer, the resulting recurring payments can be projected against fixed, budgeted amounts of expenditure. Likewise for sellers, although the initial amount received is smaller than the amount for a one-time sale, the recurring revenue over time can be projected with relative ease. Contrast this with the unpredictability of a usage-based pricing model, where neither the revenue for a seller or expense for a buyer are known in advance. This is not typically how businesses operate where expenditure is agreed against budgeted amounts of spend. This would be operationally problematic as well; would the AI based activity stop if that month’s budgeted amount of expense had been reached? In addition, processing usage data is potentially a problem to solve as well. As the price to be charged by a supplier is based on how much has been used, there is a need to acquire and ingest transactional data.

So how can software help in resolving some of these issues?

When it comes to addressing the impact of usage on the predictability of expense and revenue for both the buyer and the seller, there are several methods that are deployed. The simplest method is the concept of a digital wallet, that a buying enterprise would draw down against as used and then top up with funds as needed. It effectively operates as a Purchase Order (P.O.) defining how much money has been budgeted or allocated for a set of defined products or services. But this still does not help with predictability as it would be hard to know when top-ups will be needed. Forecasting of usage would certainly help both buyers and sellers and would address revenue recognition concerns for suppliers as they need a concept of an “average” usage pattern to establish a baseline. This would also be helpful information for use cases beyond expense and revenue considerations. Usage data is a record of how your product and service is being used in terms of what, when and by whom. Many subscription and management providers are now offering integrated data ingestion and transformation functionality, sometimes referred to as data mediation, a legacy of their origin in the telco industry. There are also specialist providers in this area, while other enterprises are looking to use general data mart or warehouse models.

Managing usage data requires the collection, filtering, aggregating and normalizing potentially large volumes of data. Pricing these transactions often requires accumulating the data over a period for tiered pricing based on hitting volume targets. For performance reasons, incrementally processing the data may be preferrable. And as mentioned this data has value as source of usage data that can be used for monitoring adoption, identifying profiles for up- and cross-sell and in general information as to what is used, when and by whom.

When considering usage pricing, whether as a buyer or as seller, know that it is not just about how the price is calculated. It also relates to how an enterprise puts in place technology and processes to make sure the promise of usage is fully realized within the confines of budget and expenditure controls and while providing visibility into other ways that the data can be used in terms of understand patterns and projected future consumption. Make sure you ask your provider, whether a point solution, a subscription management and billing system, or platform—or as part of an integrated front and back office—about how their offerings can help not with just pricing, but also as to whether they have both the technology and the processes to also help with enabling the model to work with existing business norms.

Regards,

Stephen Hurrell