News / Shippers and 3PLs 'paying for AI experimentation'

Photo: © Violetkaipa

2025 and 2026 are years of “total experimentation” for the way pricing for SaaS and AI workflows is being passed to the customer, but “the budgets are there”.   

On the AI Pivot panel at last week’s Manifest 2026 conference, Krener Komoni, CEO of Tive, described a “year of total experimentation”, with models ranging from outcome-based, to email-based, to SaaS-like monthly subscriptions. 

Others on the panel, which included Pablo Palafox, CEO of Happy Robot, and Mathew Elenjickal, CEO of FourKites, described “credits” or “actions” pricing: charging by the units of work an agent performs, such as reading emails or extracting data from documents. 

Mr Komoni said: “The budgets are there, and everybody is willing to spend on AI to see the outcome. I think in 2027/2028, that pricing model, and who pays what, is going to get normalised.   

“But I would say today, shippers and 3PLs are paying for this experimentation.” 

That experimentation is being driven by a basic mismatch between traditional software contracts and what buyers now want from AI. SaaS licences are typically priced around seats, modules, or usage tiers, assuming relatively stable functionality and predictable demand.

Agentic systems, by contrast, are designed to perform work. Their cost base can rise and fall with the volume and complexity of tasks pushed through large language models, and the value delivered is often measured in time saved or exceptions avoided rather than features unlocked. 

Mr Komoni said shippers and logistics providers were allocating meaningful spend already, citing a statistic from a recent supply chain conference that “40% to 50%” of next year’s budgets referenced artificial intelligence. But while willingness to invest is high, the mechanisms for charging, and for proving return, are still being tested in live deployments. 

Mr Palafox argued that pricing should follow “time-to-value”, reframing “build versus buy” as “buy and build” in an AI world where enterprises will create their own digital workforces on top of an infrastructure layer. Because Happy Robot positions itself as infrastructurem rather than a managed service, Mr Palafox said it could not easily “anchor to an outcome”, instead it “evaluated work” through a credit system that tracked what the AI workforce actually did. 

FourKites described a similar logic in more granular terms – billing by “actions” consumed to automate a task.  

Mr Elenjickal explained such an action might be “reading an email” or “extracting data from a document”, with intensity-based pricing reflecting underlying compute costs, for example, processing a 50-page document requiring more resources than reading a single page.  

One speaker said this approach aligned pricing with the economics of AI, where marginal cost was no longer close to zero, and where usage spikes could be immediate, once customers begin automating across functions. 

Mr Palafox suggested 3PLs may move faster than shippers because they were under pressure to reduce “cost to serve”, and could justify early experiments through quick-win use-cases.  

He advised customers to start with “low-hanging fruit” to demonstrate ROI, then expand automation more broadly once stakeholders become comfortable paying for work done by software, not just software access.