**Steven Forth** is CEO of Ibbaka. Connect on LinkedIn

Annual Recurring Revenue (ARR) is the north star metric for most SaaS companies. It is a key target in most annual plans, management gets compensated on attainment, and many investors value companies based on multiple ARRs. Ownership of a company is a claim on future revenue streams. ARR is an indicator of future revenue, so it makes sense as a valuation metric. Or does it?

Is Annual Recurring Revenue ever really recurring?

That would depend on churn. ARR does not factor in churn, which is why many companies fixate on Net Revenue Retention (NRR) as well as ARR. NRR = Current Revenue + Expansion Revenue - Churned Revenue. Ibbaka does an annual survey together with PeakSpan Capital, the Ibbaka | PeakSpan Net Revenue Retention 2024 Survey Report.

How predictable is churn? If churn is too high or unpredictable, then annual recurring revenue is suspect.

Let’s take an extreme and unfortunate example. AI SDR company 11X has been in the news to date for all the wrong reasons.

Financial Reporting Concerns:

Inflated ARR Calculations: 11x allegedly included revenue from short-term trials or contracts with break clauses in its ARR calculations, potentially misleading investors about its financial health.
Manipulated Growth Figures: Former employees claim that 11x manipulated growth figures by including payments from former customers in its annual revenue.
High Customer Churn Rate: Employees reported a high churn rate of 70-80%, indicating significant dissatisfaction among customers.
Sources here.

Assuming no expansion revenue, that would give an NRR of 20% to 30%.

In conversations with investors, we have been told that an NRR under 90% puts ARR in question.

One of the findings in our 2024 NRR survey is that year-to-year NRR performance is becoming more erratic; in other words, ARR is becoming less predictable and less useful as an indicator of future revenue streams and enterprise value.

Revenue streams are diversifying

This is not the only strike against ARR. The other trend is the emergence of hybrid pricing models. According to Maxio,

67% of SaaS companies now use usage-based pricing, up from 52% in 2022.

High-growth companies (40%+ YoY) overwhelmingly favor hybrid pricing.
Transparent, flexible billing leads to lower churn and higher expansion revenue.

Modern SaaS and agenticAI companies have multiple pricing metrics and revenue streams. If one takes a conservative approach and limits ARR to subscription revenue (per user, per process, per model), one captures only part of the revenue. And given the rising levels of churn, it is not clear how predictable ARR actually is. Usage based pricing, capacity access pricing, outcome based pricing … modern SaaS companies have many revenue streams, and this will be even more true of agents and the agent economy.

From per user to → usage

From per process to → processes executed

From per model to → model instantiations

Outcome based pricing is an even bigger challenge to conventional thinking because it is contingent on performance. Outcome based pricing only works when

The outcome can be clearly defined and measured
Attribution of the causes of the outcome can be agreed on
The outcome is somewhat predictable in advance

With advances in AI, causal machine learning, and predictive analytics, there will be more and more cases where these conditions can be met. A popular example of this is how Intercom prices its FinAI agent at $0.99 per resolved conversation. See Comparing the Value Model and Pricing Model of Intercom’s Fin AI Agent.

Annual Predictable Revenue (APR)

Is it time to retire ARR as the north star SaaS metric and replace it with something new?

I believe the answer is yes. We need to be focused on …

Annual Predictable Revenue.

Recurring revenue is of interest to investors because it is thought to be predictable. Its importance lies in the ability to predict future revenue streams and not in the fact that it is based on subscriptions.

What makes revenue predictable?

That will depend on the business model and on the mix of revenue streams.

Hybrid pricing models are built up in four layers.

Fixed subscriptions - where the buyer makes a long-term commitment for some unit (many pricing metrics are possible here other than users; Ibbaka for example, is priced on models)
Access - where one pays for assured access, like a retainer
Usage - where one only pays for what one uses
Outcomes - where payment is contingent on success

We need an integrative metric that brings all four of these together and does so in a way that is both predictive and acknowledges uncertainty.

Am I making this too complex? Probably, but as Einstein said, we need to make things “as simple as possible but no simpler.”

Pro forma APR for early stage SaaS company

APR only works if you have a way to predict revenue across the different revenue streams. Let’s look at some options.

Three Approaches to Building an APR Predictive Model

1. Component-Based Revenue Forecasting with Confidence Weighting

This approach segments revenue streams by type and applies different forecasting methodologies and confidence levels to each component.

Segment revenue streams into distinct categories (subscriptions, access, usage, outcomes)

Apply appropriate forecasting models to each segment:

For subscriptions: Time series analysis or cohort-based retention modeling
For usage: Regression models based on historical usage patterns
For outcome-based revenue: Probability models based on historical performance

Assign confidence levels to each revenue projection based on:

Historical data reliability
Contract terms (length, termination clauses)
Customer health indicators (such as Value to Customer, Customer Satisfaction Scores or Net Promoter Score)

Calculate weighted APR by multiplying each revenue prediction by its confidence percentage

2. Machine Learning-Enhanced Churn and Expansion Prediction

This approach leverages advanced analytics to predict customer behavior at a granular level, focusing on customer retention, expansion, and usage patterns.

Build a churn prediction model using machine learning:

Analyze customer usage patterns, support interactions, and behavioral data
Create risk scores for each customer or segment
Apply the probability of retention to each customer's revenue contribution

Develop expansion revenue models based on:

Historical upsell/cross-sell patterns
Product usage trajectories
Customer maturity stages

Integrate usage forecasting:

Time series analysis of usage patterns
Correlation of usage with business cycles, seasonality, and growth

Combine predictions into a unified APR forecast with confidence intervals

3. Hierarchical Forecasting with Integrated Confidence Scoring

This approach builds forecasts at multiple levels (customer, segment, product, total) and then reconciles them for maximum accuracy.

Create bottom-up forecasts at the customer level:

Contract-level revenue projections
Customer-specific usage forecasts
Propensity-to-expand scores

Develop top-down forecasts based on:

Market trends and seasonality
Historical growth patterns
Competitive dynamics

Apply middle-out forecasting for key segments:

Industry-specific projections
Customer size cohort analysis
Product adoption curves

Reconcile forecasts using statistical techniques to resolve discrepancies

Calculate confidence intervals for each level of forecast

What this means for Pricing

Hybrid pricing models are not going away. They are working well in the real world (see the Maxio data) and give us the flexibility we need to price for a volatile environment that is being disrupted on many levels:

external elements like tariffs (see Pricing in a Time of Tariff Uncertainty: A B2B SaaS Survival Guide)
technology change from AI
business model disruption from agents
buying process evolution (again from AI)
pressure from SaaS procurement management platforms (like Vendr)

We need all the tools we can get our hands on to develop flexible and adaptive pricing models. Subscriptions, access fees, usage fees and outcomes all need to be part of our cookbook. But on their own, they are not enough. We need better ways to measure results. That is where Annual Predicted Revenue, or APR, comes in.

The tools to build predictive models are now well understood and easily available. AI is standing by to help any company build and understand a predictive model. It is time to move from ARR to APR.

For pricing metrics this means we need to take predictability into account when we design pricing. Willingness to Pay, Value Capture Ratios, and Load Balancing all have their role, but predictability needs to be promoted to a first-level goal in pricing design. Pricing should include predictive models as a core part of the design.

contact ibbaka to get a value model you can use to win in a generative AI world