‍

At a fireside chat hosted by Eight Roads, Johan van der Poel (founder of Northlane) sat down with Manny Medina, founder and former CEO of Outreach and now founder and CEO of Paid, a platform that helps companies price and bill AI agents and track margins and ROI in real time. The premise of the conversation: pricing has become one of the hardest problems in software, equally for SaaS businesses pivoting to AI and for AI-native companies trying to capture the value they create.

Four ideas kept coming back throughout the evening. The value metric is the whole game, and packaging, model and price level are secondary. Seat pricing is breaking, because agents replace roles and customers now demand seat reductions. Cost-plus and per-minute pricing is a trap, because the input provider will always undercut you. And the transition demands real organizational muscle, from sales compensation to pricing governance, which is easier to build while you are still small.

Here is the full conversation, distilled.

Seats are breaking

Software has moved from perpetual licenses to subscriptions to, now, agentic delivery. Agents do not assist work, they perform it, and they replace roles. That puts seat pricing in an impossible position: customers demand seat reductions as agents take over headcount, and the seat model has no answer for that.

Manny framed the moment as two fundamental changes happening at once. First, our ability to capture value has fallen behind our ability to innovate, because everything is still locked to seats. Second, for the first time in roughly fifteen years, software pricing carries a real variable cost component in the form of tokens and inference. Today's margin relief from cheaper or self-hosted models is temporary. Compute and data-centre capacity are getting more expensive, so the window to capture value is now.

Johan added a third dimension: value delivery itself has become bigger and more diverse. Horizontal platforms serve many use cases, which means many candidate value metrics, at exactly the moment the tooling to monitor and act on usage-based pricing is finally maturing.

The value metric is the whole game

Pricing has three levers: packaging and structure, price model, and price level. The value metric trumps all of them. That is where both the opportunity and the urgency sit.

The method is to map the customer's value chain from outcomes to outputs to activities to access, and price at the point where your solution's attribution to value is still defensible. Complexity can come later, when you need to differentiate segments.

There is a bigger unlock hiding here. "Jobs to be done" has always been used to build products, never to monetise them. Agents change that. You can now price the job itself: what is the human-equivalent worth of the work, how many people would the customer need to do the same thing? Two tests tell you whether a job is priceable. Does it genuinely matter to the customer, and does the ticket size clear your LLM cost hurdle?

Cost-plus and per-minute pricing is a trap

Customer experience is the live case study. Sierra and Intercom's Fin moved the market by charging per resolution, with no charge if a human touches the ticket. Most other CX vendors charge per minute and resist outcome pricing because outcomes are "hard" to define.

The danger with input-based pricing is structural: the lowest-cost input provider can enter your market and undercut you. Manny's example was ElevenLabs. They own the voice minute. If they enter the CX market, they can offer a cheaper minute at equal quality, and per-minute pricing gets commoditised overnight. That is a race to the bottom you cannot win.

Outcome pricing forces you back to first principles: why do customers buy you, and what outcome do they actually expect? Define it precisely. A resolution is not just a closed ticket. It is a ticket that does not reopen within a week, or an escalation a human can close quickly. Getting that definition right is exactly what separates you from the input providers.

Trust, transparency and control

When monetisation hinges on perceived value, customers need two things before they will accept it.

Control: kill switches, wallets, spend limits and caps. Transparency: every credit charged maps to visible agentic work and ROI, drillable down to the individual action, which ticket, which email, which purchase order.

The old vanity metrics ("look how many active users you have") are meaningless in this world. Agents finally let you evidence the value you deliver, and trust matters more than it ever did.

Separate the metric from the model

Choose the value metric first, then the charging model. These are different decisions. You can price on a usage metric but wrap it in an upfront entitlement or commitment, which matches how buyers budget, with clear overage and next-tier rules. That eases the transition for nervous buyers.

Tokens and credits are just a mechanism to fold multiple usage metrics into one currency. They are only worth the abstraction if you have two or more meaningful usage drivers. The real question never changes: how do the AI components add value?

And know your buyer. An attendee building software for the trades raised exactly this point: their buyers will not naturally value a "token". Traditional-industry buyers may never intuitively grasp credits. Do not over-invest in educating the market unless you intend to be the one who changes it.

Pricing differently is a wedge, but not a free one

If you price the way buyers expect, so do your competitors. You all sound the same, and the conversation falls back to features, speeds and brand.

Pricing on outcomes makes you sound like a different category. You move out of the software budget and into the much larger headcount budget, and you fish for bigger fish. Buyers are curious, and new pricing opens a new conversation.

It is not free, though. Migrating to outcomes puts current revenue at risk and is transformational rather than incremental. Sierra treats its outcome catalogue as proprietary IP and publishes no pricing. The winning posture: make promises you can keep, and keep the how proprietary.

The transition playbook for SaaS

For SaaS companies making the move, the playbook discussed on stage has two steps.

Step one: meter, don't charge. Put meters on the variable value elements and show customers their monthly or quarterly consumption, zeroed out so they are not billed. This normalises the variable component before money is attached to it.

Step two: convert at renewal. Move seats to a platform fee with unlimited seats, and bake roughly 70% of expected annual consumption into the deal as credits. The 70% is deliberate: a big enough commitment to matter, but small enough to trigger an in-year expansion event.

For nervous buyers, cap usage for the first six to twelve months, then show them how far they blew through the cap. That builds goodwill and renewal leverage at the same time. Play the multi-stage game rather than maxing out year one.

Don't underestimate sales comp and org muscle

The longer you stay on non-scaling pricing, the harder it becomes to change seller compensation. Top reps expect large upfront checks from multi-year seat deals. Shift to variable or outcome-based comp and you risk losing your best people, especially the coin-operated seller profile common in US sales teams.

The lesson: introduce it early, while you are small, so you hire the right seller profile from the start. Manny's read: this is why incumbents like ServiceNow, Workday and Salesforce are fumbling agent monetisation.

Pricing is also multidisciplinary. Product, sales, customer success and finance all have a stake, which is why it needs a "pricing dictator" who gathers input and then decides. And it is a muscle, not a one-off decision: the right model at seed is not the right model at Series A or beyond.

Johan's rule of thumb: find the lowest level of complexity that gets the job done. Start with the single usage metric best correlated with value and make that your first commercial lever. Do not try to meter and monetise everything on day one.

Top-line versus cost-saving AI

Two dimensions decide how you capture value: autonomy and attribution.

Top-line AI always has an attribution problem, because too many things contribute to revenue. The pragmatic default is to charge for tasks while making the top-line-unlock argument in the sale, accepting that attribution stays hard.

Cost-saving AI, human-equivalent AI and risk-avoided AI are easier to put a number on and make a cleaner ROI case: this costs you $100 today, it costs $10 with us. It is the defensive play, less risk and less upside. Top-line is the growth play, more upside and harder to prove.

Expansion does not stop once you have replaced a team. The value catalogue keeps growing: new workflows, error-rate reduction, net-new value that did not exist before. Manny's example was a GTM-automation agent whose value is not only the human savings but the errors avoided by stitching platforms together.

Embrace the complexity

Full value and outcome pricing means more dispersion in deal outcomes and more freedom in sellers' hands. Deals start to look custom, which feels uncomfortable next to the old fixed-box model.

But at scale, as Manny put it, the top 20% of customers already have custom deals, drifting toward 50% over time, which twists CPQ, CRM and billing into a pretzel anyway. So build systems around one source of truth, what did the agent do and what value did it deliver, and treat everything else as negotiation. That, incidentally, is Paid's core thesis.

On pricing the agentic layer specifically: do not anchor on your cost, or you will train customers to interrogate your token bill. Value first, then cost guardrails to avoid bleeding on power users, and t-shirt-size your outcomes by difficulty. A simple lookup and a multi-database, retry-heavy task belong in different buckets. Marketplace or committed-spend pass-through can be an easy on-ramp, but it anchors on cost, so treat it as the start of a journey toward value.

What to watch

The SaaS migration is do-or-die. Manny dates the inflection to the March 2026 "SaaSpocalypse": emergency board meetings through spring, PE owners such as Vista pushing portfolios off seats and publishing CEO AI-growth leaderboards. Roughly half of Paid's business today is SaaS companies making this move.

AI-native companies are growing fast on new logos, but expansion pressure appears above roughly $50m ARR, especially on all-you-can-eat deals. Agents are spreading everywhere: manufacturing, supply chain, chemicals, law. Manny dates the real capability inflection to a frontier-model step-change late last year, followed by the SaaSpocalypse. And for agent builders, one tactical note to end on: enterprises are "buying one of each" right now. Even if a competitor is already in the account, keep selling.

Northlane advises AI-native and traditional software companies on Pricing strategy. If you are rethinking your model for the agentic era, get in touch.