WNDYR Blog | AI Transformation Insights & Strategy

The 45-Day Gap: Where Port AI Should Start Paying for Itself

Written by Mike Blakesley | Nov 17, 2026, 11:55:04 AM

A tug finishes a ship assist at 6 a.m. The job is in dispatch by 6:15. The invoice goes out 45 days later.

Nothing went wrong. Before a completed job can become an invoice, someone in finance has to confirm the vessel, the customer, the contract, and the rates and discounts that apply to that specific movement. That information usually sits in several systems, so it gets exported, emailed, looked up, and re-entered until there's enough to bill.

In the port service operations I've worked with, that process often takes a couple of weeks, and complex jobs can take up to 45 days to invoice. The work was finished on day one, but the company is still waiting to be paid for it. For an operator with $40 million in annual revenue, a 30-day gap means roughly $3.3 million in earned cash is sitting idle at any given moment.

The first place AI should prove its value in port services is closing that gap between a completed job and a sent invoice, using data the business already captures. It's a smaller ambition than a new platform or smarter vessel scheduling, and it pays back sooner. Finding it starts with studying the workflow before choosing the technology.

Most port AI attention goes to vessel movement

Most of the conversation about AI in ports centers on vessel movement: arrival timing, berth planning, and tug scheduling. There's good reason for that. A study of roughly 28,000 port calls at Rotterdam found that delayed towage was the largest direct cause of vessel delay, and the researchers traced it to how information moved between pilots, agents, vessel traffic services, and tug operators rather than to the towage itself (Maritime Economics & Logistics).

Fixing that depends on pilots, agents, terminals, and carriers agreeing to share data, which takes time and coordination across many organizations. The same kind of disconnection exists inside a single company, between its operations and finance teams, and that version is far easier to fix because one leadership team owns it.

The days before the invoice usually go unmeasured

Most collection metrics, including days sales outstanding and aging reports, start counting when the invoice goes out. Days to Invoice, the time between finishing a job and sending the invoice, often isn't tracked at all, even though it can be just as long.

The Hackett Group's 2026 working capital survey of the 1,000 largest public companies in North America found that receivables are now the largest and fastest-growing source of excess working capital, up 29% in a year to $773 billion. Days sales outstanding worsened by 2.1 days, and Hackett notes that receivables problems often start long before an invoice is overdue.

Those are large companies with mature finance teams. In the midmarket port operators I've worked with, a delay of weeks often happens before an invoice even exists, and during that time the company is financing work it has already completed.

What we learned by watching the work

We worked with a Gulf Coast harbor towing operator on this problem, and the method we used mattered more than any particular piece of technology.

We started with observational research and one-on-one interviews with people in dispatch, operations, billing, and finance. Each group understood its own part of the customer journey well, but nobody had a view of the whole path from completed job to collected cash. Laying it out end to end gave everyone the same picture for the first time.

Next we built workflow maps and a service blueprint to identify the steps the business actually requires, like verifying the customer and applying the right contract terms, along with the workarounds people had developed to get around their systems, like exports, manual lookups, re-entry, and side spreadsheets. Workarounds are some of the most useful findings in this kind of research, because each one points to something the systems should do, but don't.

The blueprint also showed the dependency that mattered most. Fast, accurate billing depends on tracking each job at the movement level, because who gets billed can change from one movement to the next, and discounts such as flat rates, fuel surcharge discounts, towage discounts, and early-payment discounts can apply in different combinations. If the movement record is wrong or incomplete, the invoice will be too.

From there we used Claude Code to build high-fidelity working prototypes and tested them with the people who'd use them, on real data. Because changes took days rather than months, we could iterate quickly, and the people who do the work raised edge cases that wouldn't have come up in a requirements document.

Much of the reconciliation between collected movement data and combined charges had been falling on the Controller's team, who were doing it by hand, so we designed new tools with them directly. Invoices can now be processed nearly as soon as jobs are completed, with contract and discount terms applied from connected data.

Why the problem grows with the business

Each new customer, contract, and discount structure adds verification work to every movement, so this kind of gap tends to widen as a company grows.

The math compounds quickly. A single movement can carry four kinds of discount (flat rate, fuel surcharge, towage, and early payment), which can combine up to 16 different ways, and every new contract layers its own terms on top.

Manual billing absorbs all of that by hand. Its cost tracks the number of combinations and handoffs, not just the number of jobs, so a growing operator either adds people to a process that slows down as it grows or accepts a longer wait between work and payment.

Starting with the data you have

The usual advice is to fix your entire data foundation before attempting AI, and there's evidence behind it. Gartner predicts organizations will abandon 60% of AI projects that aren't supported by AI-ready data.

Gartner's own definition is worth reading closely, though. It describes AI-ready data as data that's "representative of the use case," including its errors and outliers. Real movement and contract records, with all their inconsistencies, fit that description better than a cleaned-up dataset built in isolation.

Waiting for perfect data isn't realistic for most of the industry anyway. In a recent Thetius and Marcura survey, fewer than one in ten maritime professionals said their organization's data is fully ready for AI. The research and mapping work tells you which records matter for a given workflow, and in most cases you only need those connected and reliable to get started.

AI has also changed what it costs to try. Working prototypes used to take months, which pushed teams to specify everything up front and hope they got it right. Now a team can build, test, and revise in the same week, and that speed makes it practical to design with the people who do the work, not just for them.

What becomes possible after the invoice

Once the movement record is connected and current, it becomes useful well beyond billing.

It supports forecasting, since a reliable history of movements and revenue can project next month's jobs, crew needs, fuel, and cash. It also makes cash flow something you can compete on. An operator that knows its true margin per movement and gets paid faster can Where to startprice more precisely and offer early-payment discounts without guessing at the impact.

It can inform expansion decisions too. Combining detailed per-movement financials with public vessel traffic data for a target port would let an operator model that market before committing to it, building a financial digital twin of the port.

 

For operations and finance leaders at port service companies, start by measuring your Days to Invoice: the average number of days between a completed job and a sent invoice. Most teams don't know that number.

Then follow a single movement from completion to invoice and talk to everyone who handles it along the way, noting each system it passes through and each time someone re-enters it. Separating the required steps from the workarounds will give you a clearer list of requirements than most specification documents, and prototyping a fix with real data and the people who do the work will tell you more than a platform evaluation would.

The money for these jobs has already been earned, and the data needed to bill them is already being captured. Connecting the two is the most practical place I see for AI to start paying for itself.

If you want a second set of eyes, we'll trace one movement with your team, from completed job to sent invoice, and show you where the days go.