It's The End of the (ARR) World and I Feel Fine

So much changes - but probably for the better if we are honest with ourselves

There’s been no shortage of recent media coverage and Twitter pearl-clutching around how native AI companies simply aren’t measuring ARR correctly or are intentionally misleading investors with their numbers. Add to that the endless online chatter about charging for outcomes or deliverables as opposed to user licenses — and the inevitable question of what the hell does that actually mean and how do you do it right? Everywhere you look, there’s a new expert writing about how to do this basic fundamental thing differently in this new era.

What I think everyone is missing is the high-level reality: ARR was great for the SaaS age, but it may be as relevant as a Commodore 64 for the AI software era. And if so, whoa — that means a lot of change for how we think about running, evaluating, or investing in software companies. Let’s get into it.

Recurring Revenue Used to Mean Software Support and Updates

Way back when I left Siebel Systems — the CRM leader of the on-premise world — and joined salesforce.com in 2003 as their first Head of GTM Ops, I witnessed the creation of the concept of ARR and even TCV. In the vein of “back in my day,” software used to be sold upfront for a large initial fee, and then customers were charged 8-10% annually for support and the right to receive updates. The software was usually shipped on a CD — the “golden master.”

That world wasn’t terrific for customers. On top of those software fees, they’d pay up to 10x that cost for databases to store their data, hardware to run the software, and armies of consultants from Andersen Consulting (what became Accenture) to install it. There were many other customer-unfriendly elements of the on-premise model — namely how much you’d spend reimplementing your customizations every time you installed an upgrade. At least CFOs could amortize the software spend over multiple years.

On-Demand / Cloud Started with Renting Software — It Really Did

With “on-demand” (that’s what it was called then), customers just signed up for software via their browser and paid on a monthly rental basis. The software was primarily charged per seat, with higher prices for editions with more capability. Pretty quickly, while that was a sweet deal for customers, month-to-month rentals or MRR made it damn hard to run and self-fund your own growth as a software company. Without a guaranteed stream of customer revenue you could reliably count on, it was hard to invest in enough GTM opex to drive the growth VCs all clamor for and attract the best salespeople who were used to fat commission checks — especially when in the early days salesforce.com was buying its own Sun servers to run in Equinix data centers (no, there were no hyperscalers then).

The Rise of Annual Contracts — What a Long Great Run It’s Been

Salesforce, and many other SaaS companies funded since then, quickly gravitated to the concept of annual contracts — whereby customers, for a specified discount, signed up for an annual contract with some minimum monetary commitment they couldn’t leave except for contractual breach. That was soon followed by annual payments upfront and multi-year agreements, where customers committed to an annual amount for 2-3 years. In fact, salesforce.com’s early compensation plans were 1/3 ACV, 1/3 cash, 1/3 multi-year — to drive this need for revenue the company could count on and the cash it needed to fund its business (remember that SFDC raised less than $20M from seed to Series C).

As a result, what mattered most to investors about these companies was the level of committed revenue and the relative growth in that committed revenue over time. This gave rise to the classic SaaS efficiency metrics — the Magic Number, the Cash Burn Multiple, NRR (net retention rate), and even UFR (up for renewal rate). The latter became important to truly understand customer health when companies started including the locked-in second and third years of multi-year agreements in their numerator and denominator for NRR (disguising potential churn problems). With this, we saw the creation of almost an entire new function in customer success managers whose role was to ensure existing customers were “happy” and would definitely renew — with the sometimes flawed notion that these brave souls could make that happen regardless of the product’s quality or how much sales oversold the product’s promise.

The Positives for Software Providers

• First and foremost, these contracts deliver the predictable recurring revenue you need to properly plan for incremental opex investment and the cash to pay for it, assuming you can forecast churn correctly.

• Second, far higher go-to-market efficiency per dollar of revenue — assuming no existential churn, the majority of your sales and marketing dollars are spent on adding net new ARR since account execs are never paid commissions on renewal dollars. This means a good rep becomes a cash ATM for a company by somewhere between year 2 and 3.

• The ability to, as long as your customers don’t churn in droves, stay in the game long enough to compound your way over time to 10 and 11 figures of ARR and — assuming you’re not heavily investing in growth — turn into cash cows ripe for PE purchases.

• The ability to amortize the cost of your large upfront commission payment to a rep over the length of a multi-year agreement, thus making your sales efficiency numbers look far better.

• The time you need to get your product working well for your customers during and post-implementation to ensure they’re happy with their purchase.

• The time to build deep relationships with that customer, have them attend your annual shows, “make them famous,” and have them tie their own success to the use of your software.

The Negatives for Software Providers

• It’s damn hard to displace competitive solutions at prospective customers who are under contract for a longer period — the traditional rip & replace strategy. Those contract end dates dictate when you can sell your competitive solution to the incumbent.

• The amount someone will spend on your solution generally starts at what they spent on their prior solution, regardless of how many times better yours is.

• Most importantly, it creates a real crutch for the product team. They don’t have urgency around resolving problematic issues or staying competitive with new upstarts — it’s the definition of becoming a legacy provider. PMs also are far more influenced by what Sales wants to sell for net new ARR vs. what Success wants to retain it — it’s just how it is.

• Additionally, over time, many SaaS orgs have had to invest more and more of their annual S&M spend in maintaining/expanding customer revenue and often turn into some type of seven-headed IBM client account team with reps, SEs, PS folks, CSMs, AMs, renewal managers where effectively no one really owns the customer relationship.

The Pros and Cons for Software Buyers

• On the pro side, buyers can predict their operating costs with relative ease. The annual commitment also forces them to put in the required time and energy upfront to make the software successful. Similarly, given they know how much vendors care about preventing churn, they can put pressure on the orgs to address their issues.

• The reality, however, is that they can’t really demand providers fix their issues on the timeline they want, nor have the vendors fill in key product gaps that then force them to buy a slew of additional products they also have to lock into for a year or more. This leads to a lack of organizational flexibility — they’re often limited by the end dates of their respective software contracts to drive effective process change. Their opex spend on tools is locked up until a renewal date — and god forbid if that’s several years out. And let’s hope their vendors don’t, as some unimaginative large SaaS vendors have done, decide to raise prices for the same products at renewal far higher than the cost of CPI that was heavily negotiated in the originally signed contract.

The AI Age — Hiring, Not Buying, Software

As I’ve discussed before, we need to think of AI software as products that actually help customers do their work — either assisting humans in their jobs to be done or actually doing some of those jobs entirely, obviating the need for a worker to do that work. Therefore, in that world, we’re not actually “buying” software. You’re “hiring” it. This purchase, especially when AI vendors are charging you far higher prices (than normal POS or “plain old SaaS”) for agentic “labor replacement” solutions, should be viewed through the same lens as hiring a human assistant — albeit one that’s super smart, always learning, works 24/7 (9-9-6 is for ninnies), and never rants on Glassdoor about your abusive leadership style.

When you start using that analogy, the concept of annual contracts and ARR and, god forbid, TCV seem like sheer lunacy. For example, if you were to hire a new employee that cost you $200K a year, would you pay that employee all of their $200K upfront? Hell no. Would you sign them to a one or multi-year agreement that, even if they significantly underperformed, you’d keep them in that role until the contract was up? Of course not. So why would it be any different for agentic software that claims to replace said humans? If we buy this argument, here, in my view, are the top implications for SaaS companies going forward, in no particular prioritized order.

What Contract Terms? You Want Cash When?

SaaS companies have built their complete operational models on the back of annual contracts with cash payments upfront. We even convinced many non-freemium buyers that we’d invoice them the moment we provision their service and not when they go live and start seeing value. Established SaaS vendors are still pushing for annual contractual commitments upfront — but I think this will be a losing battle, especially for any systems of action AI software providers that charge far more than traditional CRUD systems of records.

Customers are hiring this type of software to do work, and as such they’ll wait until they have enough history of high performance before they’ll consider exchanging some combo of contractual terms and non-cash-in-arrears payments for preferential lower “per-unit” pricing. Even then, they’ll be far more dubious about locking themselves in for multiple-year contracts and especially about paying a year’s worth of cash upfront.

Committed Recurring is the Exception, Not the Standard

There’s been a ton written around what to do with pricing and packaging in the new agentic world. So much of the content is centered around paying for usage and/or outcomes. Which of course means that predicting what revenue might be for a vendor or cost might be for a buyer is no longer a relatively fixed and understood number.

In my mind, what will emerge as the new normal is a combination of a base platform access fee (which should not be more than 25-30% of the roughly monthly cost of using the service) — which would be effectively committed monthly recurring revenue — and then a set of highly variable usage/outcome fees tied to the specific systems of actions you hire/use from a vendor.

To play devil’s advocate: if it’s usage-based, is this really that different than how email marketing, SMS messaging, or platform infra vendors charge — which is effectively a per-piece charge that customers can decide, once they have some baseline usage predictability, to pre-buy a block of usage for a lower per-piece rate? It’s not — but I think this becomes de rigueur as opposed to only for certain categories of software.

As for outcome-based, this is a whole new ballgame that while inherently interesting (customer only pays for value received, vendor is able to charge perhaps more per outcome over time than what they may have charged previously), also brings a ton of friction into the agreement between the two parties. Regardless, I predict that you will not have a repeatable committed monthly revenue amount as the predominant component of your overall revenue that you can rely on to run your business.

It’s Easier For a Customer to Decide if It’s Time to Switch Vendors

The report card to evaluate AI products versus a SaaS product is far less opaque and much easier to use when deciding if it’s time to make a switch. SaaS products don’t do work — they’re repositories and workflows that support you doing the work — so deciding if it’s time to move on requires a larger introspection process. Was it more a reflection of our team, our use of the product, or the actual product itself?

In the AI world, we skip all of that. AI products are here to help us do the jobs to be done and are often measured by measurable output and/or whether they could replace what we used a human for before just as well, if not better. As such, there’s real purity here — these products either work or they don’t, and they either deliver the desired results or they don’t. It’s much easier to assess whether you should be shopping for a new AI product — especially if you didn’t commit yourself to a long-term contract.

The Ironic Case for Why Agentic Products May be Stickier than We Think

In the pre-AI world, when you purchased and used SaaS apps, they were repositories and workflows to support your know-how and your business processes. The licensed users were the practitioners of this craft — the expertise for the given function was owned by the customer.

With agentic use cases, however, the race is on to move as many low and mid-level tasks to agents as fast as possible — with the notion that you’ll manage and direct these agents at a high level. These agents determine on their own the best paths to get a given directive done — so in this world, you’re actually outsourcing the craft and the expertise to them. Your institutional memory, if you will, is getting erased over time. So if you decide at some point not to use the agentic offering, what will you roll back to? That’s going to be really hard.

So while you may swap out one vendor’s agent for another, you’ll never go back to doing nothing here because no one in your org knows how to do that full job anymore.

If you want to prevent your agent from getting swapped out for another vendor’s agent — then context, memory, and reinforcement learning are key. Your solution should be one in which you’re constantly learning what works for that given customer as well as prompting the “managers” to inject know-how and training (i.e., FEED ME) such that, unless all of that context is portable (customer will want it, vendors should resist it), a customer will rightfully worry that a replacement agent will take way too long to get up to speed comparably with the incumbent agent.

The Best Products (with Great Marketing) Will Likely Be the Big Winners

In the AI world, I believe that while amazingly loud and catchy marketing works (AI marketing seems more consumer-inspired than B2B), the best actual products may just win — the product either does the job better than others or it doesn’t — which buyers can evaluate much easier than ever before. As a result, PMF becomes less of a question mark. You either have PMF or you don’t. Your product either delivers on the task you say it does or not — what I call product purity.

If people don’t buy it and stick with it, then you cannot pass GO — you will not be able to invest in and accelerate your GTM. You will not show the level of crazy rapid revenue growth that investors are looking for to make king-making investments. From an investor perspective, the upside is that, phew, no more friggin zombie SaaS companies that take forever to die — forget NRR and GRR, you won’t have R. No amount of sweet swag, LinkedIn brags, or G2 reviews will change that.

Increasingly, as we get through this hacky “perhaps easily replicable” AI products and “curious George researcher” buying phase, the best products will need to have inherent stickiness beyond some clever AI workflow hacks with APIs that access your customers systems of records. If anyone watched the OpenAI Dev Day video, good luck with that as Altman and team push for control of the user experience definitely at home and potentially at work.

The New Product / Design / Dev Iterative Troika Cares Far More about Customer Needs

In the agentic world, where there are far fewer UX screens to build and no longer the six-month struggle to listen, design, code, test, alpha, and release capabilities — the customer’s feedback on both how well you’re doing on the jobs to be done they’re using and the additional jobs they want you to do becomes gold. As I’ve written before, those companies that iterate around these pains faster and more insightfully than others will win — hard stop.

In the past, SaaS product managers would often scoff at customers’ feature requests, deriding them as “poor product managers.” Customer success teams had to force themselves into the product roadmap sessions to ensure customers’ top needs were being heard. In an agentic world, that’s no longer the case — the voice of the customer is gold because they don’t tell you feature requests, they tell you what their pains are, and it’s up to you to imagine how you can deliver a solution they couldn’t pre-imagine existing and blow them away.

In a world without a large amount of pre-committed revenue, the intentional over-focus on new features that sales can sell to win more new customers gets rotated back to a very healthy balance — if your existing customers can more easily just move on from you, you’d better do whatever it takes product-wise to keep them. No amount of customer love or customer success touches will overcompensate for that. In the early days of SaaS companies, the expression was often that they love the customer success organization and the company better than the product because it takes time and money to get the product up to snuff. That approach is flipped on its head in the agentic age although great success people are worth their weight in gold.

POCs & Onboarding Are Now Key Strategic Levers — Customer Success Starts Mid-Funnel

In the pre-AI SaaS world, these three words were some of the most dreaded to hear during sales forecast calls. It was almost anathema — either we had a defined trial period for our product or we didn’t. We never wanted, god forbid, for the potential customer to have to deploy and use our product and see the value of it before agreeing to a contract — much less commit dedicated resources from the company to deploy and babysit the customer without guaranteed revenue.

Well — in the AI world, au contraire, mon frère. Whether you call it a live production trial or a proof of concept, prospects want to not only see this newfangled agentic AI product in action but they want to get their hands on using it with their data before agreeing to be a customer. When you hear the phrase “forward deployed engineer,” this is just a fancy term for having a technical/implementation person help a customer use your AI product and see value in it before committing to being a customer.

While this may seem like a pain in the ass, the reality is the customer, if you do this right, is already deployed and raring to go at the time of signature — thereby greatly reducing the chance (unless your product hallucinates a lot and asks users to join their new cult) that the customer will bolt. Remember, absent an annual contract, they surely can and will.

Similarly, as I’ve written previously, as the use of AI is such a personal rabbit hole for each individual to jump into and more than 800M people have a preconception around how it works, onboarding users — whether they’re using an assistant or managing agents — is something you must nail. If they don’t get it quickly and make it their new habit, they’re likely to move on to look for other solutions that are simpler to understand and use, both at an individual and organizational level.

Finally, I hope it doesn’t need to be said that customers are not going to put up with lengthy or costly deployments of these products and they surely are not going to agree to start paying until they’re live — so invest super heavily in in-product onboarding and agentic configuration and deployments. Do NOT say you need Accenture to get this to work. Product led and Engineering led for the win!

Funding Future Growth from Within is Far Harder

This is likely the most impactful change to the traditional SaaS business model. Without committed ARR and high retention rates, you simply can’t afford — short of a ton of super patient venture capital dollars — to invest aggressively in GTM growth considerably ahead of seeing the results. Alternatively, to invest in that growth must mean you need a far more cost-efficient GTM model.

Good news is that if you have an agentic product that very clearly does a given job well, the sales process should be, as I’ve previously written, far less complicated and one that requires far fewer people to close a deal. Bad news: if your retention isn’t guaranteed, you might, god forbid, have to compensate account executives on renewals. The second you do that, you really start to break the ATM machine of good account executives. But don’t worry, you should need far less opex in general to hit your goals as two-thirds of the people you would have hired are now agents (as long as companies don’t charge you an arm and a leg for that labor replacement value).

VC Funding May Become Stratified into Winners

Except for the absolutely crazy breakout companies, this implies that if you don’t have a ton of committed ARR and a lot of upfront cash payments, you’ll need to fund some of your growth the old-fashioned way — from the balance sheet. In that world, I actually take a contrarian view to the point that there will be so many more software companies and it will require less funding for them to be super successful because of AI.

I believe it may be more of a winner-take-all approach at some point in each market, as those companies with the most funding are likely able to invest the most GTM opex ahead of growth. That being said, it’s becoming harder and harder to assess the durable revenue of start-ups — everyone likes to claim an ARR number. But for obvious reasons, your last month’s revenue × 12 does not = ARR in a world where many AI deals are short pilots, usage-based contracts, or test runs — not committed, long-term contracts, not to mention the high degree of variability on the outcome/usage-based pricing each month.

In addition, determining if a company has true PMF is much harder to evaluate as LLM platforms keep moving up the stack, new AI competitors can pop up and match your capabilities in a few months, and customers are doing a lot of “trying shit out” to look good to their bosses. Sometimes it feels like AI B2B founders are almost in B2C land as they need to find and re-find product-market fit every six to twelve months. This is probably why we’re seeing all of these king-making VC rounds, often announced just months after a company’s prior raise, that at first glance don’t completely foot/pencil out. Investors are picking their winners in categories and providing them the capital to invest considerably ahead of traditional ARR economics, and in some cases, gross margins be damned.

The PE Industry Will Have to Re-Invent Its Model for the Post-ARR World

This was announced by Vista a few weeks back that I suspect very few people took note of. I did. Vista, arguably one of the best private equity investors in the SaaS industry, understands that the existing PE model is dead — the model that says buy a company, slash its opex to get to Rule of 40 with generally a smaller investment in R&D, and count on NRR to hold up long enough such that you can then flip that company to someone else.

In the agentic world, every new AI incumbent in every horizontal and vertical market is coming for the “King” with systems of actions that add far more value to getting work done than the vast majority of SaaS companies owned by PE companies. As such, PE has to change and move quickly: a) leverage agentic to add far more utility to their underlying assets — the majority of which are systems of record — the enterprise strikes back, and b) leverage agentic UXs to actually deliver product integrations and not PowerPoint integrations across a set of acquired assets. Let’s hope they do, as many startups who realize an IPO or super strategic acquisition isn’t in the cards rely on private equity buyers as their natural forever home.