How to Price Your AI-First Product

Why we’re seeing a rise in output and input based pricing models relative to SaaS

AI-first companies are no longer automatically opting for software as a service (SaaS) pricing, but are evaluating whether there are other transactional ones that are more attractive. This topic has been particularly top of mind in the tech industry lately with the proliferation of "AI enabled services" or "Service as Software" businesses - companies that are leading with services and looking to replace traditional labor spend by selling work instead of technology. Below we will discuss our perspective on the pros and cons of each, when we think one model may make more sense, and make a further distinction within usage based pricing - between charging for inputs vs. outputs.

The Trend Towards Transactional Pricing

Labor is a much larger category of spend than software at almost every company. The draw of tapping into this spend is clear - larger contracts, an easier sell (potentially) since labor is generally seen as more mission critical than software, and if the margin story plays out with AI, similar margins to traditional software. As LLMs are increasingly able to automate entire workflows, AI-first companies can sell into these labor budgets while automating some or all on the back end.

SaaS has long been lauded as the best model – in our analysis SaaS companies have historically traded at ~1.5-2.5x higher revenue multiples in the public markets than transaction models since 2020. Despite this, we are seeing companies increasingly adopting transactional models (vs. fixed recurring ones) in order to tackle these larger TAMs which have been traditionally categorized as labor.

Companies are not just evaluating subscription vs. usage but in fact are considering three types of pricing models – fixed cost (per seat, location, etc), input based (usage) and output based (outcomes or success based).

Types of pricing models

Below we explore the different pricing models, walk through illustrative use cases, and provide concrete examples from specific companies including outlining the benefits and tradeoffs.

Fixed Cost

The SaaS model (charging per seat or location) has many attractive elements – cost predictability for customers, forecastable revenue for the company and often a compelling cash conversion cycle with contracts paid upfront. It also has drawbacks - it is a challenging value proposition to sell if usage will not be consistent every week or month, if the software is replacing something that is “free” like email or Excel, or when the ROI is not perceived to be strong enough to warrant a recurring fixed expense (which often is a high bar). In addition, with fixed SaaS models there is always a risk that the customer’s top line scales faster than its need to add additional headcount especially in an AI first world. This means companies may not capture as much of the value they’re creating, and therefore scale revenues, as their customers grow.

Fixed seat-based models work best when customers have an existing software budget allocated for this use case. This model is also often most compelling for products that are used frequently, where software is enabling workflows, or when there are clear productivity or efficiency gains. Subscription models are less suited to situations where AI solves a very specific “wedge” use case very well but is done by customers less regularly such as annual planning or IT spend rationalization. In such cases, customers may be way less willing to pay a recurring flat rate for software they won’t use on a regular basis.

For example, Ivo charges a per-seat SaaS fee for its AI-powered contract review solution. They primarily target in-house legal teams who use the product weekly and tend to have fixed software budgets but see a clear path to seat expansion because contract negotiation workflows are inherently collaborative with other functions like procurement or HR. In financial services, Hebbia, for example, helps finance professionals parse documents to extract key information from data rooms, earnings transcripts, press releases, SEC filings, and other public and proprietary data sets. Hebbia also charges per seat because it is a daily use case and customers are accustomed to paying fixed software licenses for data and analysis products.

Input Based

Input-based models charge on consumption – for example how many terabytes of video, the number of queries or the number of pages that are being processed. This model is closest to the conventional usage-based pricing we’ve seen from software companies – how much customers are utilizing in storage, compute, or API usage determines their bill at the end of the month.

Input-based models work best in use cases where usage levels are relatively predictable or knowable in advance and customers can translate inputs into cost savings or ROI. For instance, Trialkit, a company that sells discovery software to law firms charges based on the volume of inputs (GBs of data) it has to ingest, analyze, and store because its compute and storage costs will vary depending on whether the case is just a few hundred documents or has thousands of files various file formats like video, documents, and audio recordings.

Skarbe is another example in the AI-powered sales productivity space. Skarbe allows customers to “pay for value, not for seats” so they sell credits customers can use to process specific CRM batch uploads, automate meeting followups, or generate lead overviews. Customers can quantify inputs and determine the value and cost savings of each action to make a ROI based decision on when (and when not) to leverage Skarbe in their workflows.

Pricing for value in this way is more challenging if the input metric is not tracked or is less predictable (i.e., document size may vary, or amount of data indexed might grow exponentially). Companies would need to provide a way to help customers track the input if it’s not already a KPI they track internally. It is imperative to do this well otherwise customers may feel like they’re being overcharged, or that they’re dealing with nontransparent and complex vendor pricing.

Output Based

Output-based business models charge for an outcome – such as the number of units of work produced, tasks completed, or summaries written. This model translates most directly to replacing human labor (instead of a human outputting a body of work, AI is), and works well for companies particularly focused on throughput.

Certain industry structures like legal and consulting are well aligned to this model. For example, most attorneys bill clients hourly so clearly calculating the cost of output for a specific case allows them to bill the client for the work whereas allocating the cost of a monthly software license across clients is more challenging. For instance, Casemark, a platform that generates summaries for legal documents like depositions, medical records, contracts, and litigation updates, charges per summary generated. Summaries were previously generated by paralegals so this unit of work is well understood and paralegals are now freed up to work on higher value tasks. Summaries are also clearly tied to a specific case which allows attorneys to pass on Casemark’s cost directly to clients.

The output-based model is well aligned with industries like healthcare where payers and providers are incentivized to drive efficiency and improve patient outcomes such as in value-based care. They may be more open to adopting solutions that structure their business models to only charge on success. For example, Synthpop.ai, a defy portfolio company, is an AI first platform that streamlines healthcare workflows and automates up to 80% of administrative and operational tasks. Synthpop charges only when its product successfully completes a task such as assessing the accuracy of a patient referral packet, completing intake forms or calling a payer to confirm insurance coverage. Customers primarily buy Synthpop’s services to reduce labor costs, to alleviate healthcare professional burnout, and to reduce errors in processing, but additional benefits include increased revenue because by processing patient information more quickly their customers are able to serve more patients while also improving the patient experience. Synthpop sells its customers on increased efficiency while improving patient outcomes, which is central to many of their customers’ missions.

Hybrid Models

One challenge we’ve seen with input- and output-based models is retention and repeatable top of the funnel conversion – subscription models tend to have more favorable retention metrics on a month-to-month basis, particularly in the enterprise. Once a customer signs up, the monthly cost is fixed for the life of the agreement regardless of the frequency of use. For usage models, retention is not only less certain but also less clearly defined since business needs and therefore volume could vary drastically month to month. For this reason, many AI companies that employ input or outcome based models have introduced hybrid pricing models with tiered subscription or platform fees with usage minimums. Tavrn, which generates AI-powered demand letters and medical chronologies for attorneys, has experimented with tiered pricing while retaining a usage-based component. The company signs annual commitments with its customers who commit to a minimum number of medical chronologies generated, and charges for additional case load beyond that threshold.

Which model is right for you?

Each pricing model has its own sets of tradeoffs which are important to consider as you assess which monetization strategy is right for your business. A few factors we suggest you take into consideration:

1. Frequency of use: How often do your customers use your product? Is it used daily/weekly/monthly or is the use case more sporadic? A fixed monthly price may be easier to articulate upfront as you build more product outside of your initial wedge; however, retention and conversion metrics could suffer if there aren’t other factors driving stickiness and clearly illustrating ROI.

2. Customer metrics and KPI alignment: What are the most important metrics your customers care about? Are they already measured and tracked? How do customers think about ROI and what are they optimizing for? Aligning to what your customers track and care about is key. It’s important to consider the nuances in your customer segment.

3. Pricing simplicity: Make sure it is clear to customers how you charge and what they will be billed for using your solution. If you’re using input or output pricing, make sure customers can quickly calculate how much it will cost for a specific task so they are not surprised when the bill arrives.

4. Cost structure: If your costs will vary significantly based on the complexity of the specific task and the information you will need to process to perform it, a usage based model may make more sense economically if your customers are willing to pay for inputs or outputs. AI model costs should continue to decrease so in our opinion this should only be a significant consideration if the cost is so high it makes your unit economics challenging today.

5. Labor efficiency mindset: While it may be tempting to compare your AI-first software cost relative to humans doing the same work and price as a percentage of labor savings, not all companies want to evaluate ROI in this way. Your customers may still think of software and labor spend as different until they have seen the results. Align your model and value proposition with the pain points you hear from your customers. The benefit of AI driven work may not be to replace humans but to allow them to focus on more strategic work. In other companies it may be more about reducing errors in industries that are short staffed or have higher turnover. In these situations customers may be more open to input and output based models because it directly aligns with their biggest need - labor.

Conclusion

The importance of selecting the right pricing model cannot be overstated - it can have far-reaching implications on retention and engagement, willingness of customers to purchase, perceived value add and ROI.

While usage based models are en vogue right now, their sturdiness in terms of long term multiples remains to be seen. In fact, the term “usage” is really just a newer term for transactional or non-subscription based business models. While we’ve seen transaction-based business models in the past get valued at SaaS-like multiples at certain times (Uber, Lyft, Robinhood, Doordash, etc), there was often a correction over time as investors realized the tradeoffs of these business models compared to SaaS - lower revenue predictability due to its non-recurring nature, greater lumpiness in revenue, difficulty in smoothing out revenue if spikes or volatility in demand occurs, etc. Now, with AI startups, we are seeing a similar premium in valuation for many companies with non SaaS business models, both input- or output-based. On the other hand, one could argue that these higher valuations can be justified by the step change in tech innovation, pace of growth, opportunity to tap into labor spend, the potential margin as software replaces humans, and the larger resulting TAMs. Are these valuations justified or will multiples follow a similar trajectory as in the past? Time will tell but for now we are excited about the innovation and business impact these new models represent.

Are you building verticalized software with SaaS or usage pricing? If so we’d love to hear from you!

Thank you to all the people who provided input and thought partnership - Julia Maltby at Flybridge, Min-Kyu Jung at Ivo, Pedro Paulino at Tavrn, Scott Kveton at Casemark, Elad Ferber at Synthpop, and Mikita Martynau at Skarbe.