13 Garbage Routes Just Disappeared at Casella. The Fix Was an Invoicing System.

THE CRAFT

Thirteen garbage routes disappeared from Casella Waste Systems this summer, and the trucks that ran them and the crews that rode them went with them. What made it possible was a back-office system: one path for taking an order, dispatching a truck, billing the customer, and collecting the money, finished in the second week of May.

Twelve weeks after it went live, Ned Coletta told analysts that Casella had “already eliminated 13 routes and the related trucks and labor from the business.”

Thirteen routes will not headline anybody’s AI conference. It is a real number, though, and it has a property the conference numbers usually lack: you can see exactly what produced it, and what produced it was an invoicing and dispatch system rather than a model.

Casella has AI. It is running in-cab cameras across the fleet that coach drivers in the moment, a workers’ compensation triage program that has been expanded, automation work in the back office, and it is moving toward AI-assisted routing. Every waste company with an investor-relations page has an AI sentence ready for you this year. What Casella also has, and what is rarer, is a calendar you can lay the sentences against. Systems integration finished in May. Route consolidation began immediately after and is still running market by market. The routes came out in the summer. The AI is arriving on top of all of it, now.

A route is not a thing a model can remove for you. Somebody has to know which stops on two trucks are three streets apart, whether the customer on the corner has a local arrangement nobody else has heard of, which of two conflicting records for that customer is the one that gets billed, and where the driver and the truck go instead. Before May, a dispatcher in the Mid-Atlantic trying to answer those questions was working across businesses Casella had bought that each kept the answers in their own shape.

That is the story I want you to have this week, and the reason it matters to a company that has never owned a garbage truck. Casella is a consolidator, and consolidators are the purest version of a problem almost every mid-sized business has in some form: the work is being done several different ways under one roof, and the technology you are being sold assumes it is being done one way.

THE OPERATOR

The situation

Casella collects, hauls, transfers, disposes of, recycles, and processes non-hazardous waste across the eastern United States for households, businesses, towns, schools, hospitals, and industrial customers. Trucks make stops. Transfer stations consolidate what the trucks bring in. Landfills and processing facilities take it from there. Recycling and organics turn a portion of it back into a service or a commodity someone will buy. The company did $1.84 billion of revenue last year across three geographic regions and a Resource Solutions business.

The economics are stubbornly physical, which is what makes the company a useful test of grand technology claims. Money in this business is made on density. A route with the right stops in the right order beats the same route with empty miles between them. A transfer station earns its cost when enough material moves through it. A landfill is worth more when the company can feed it profitable tons out of its own collection system as well as from third parties. None of that responds to enthusiasm. It responds to whether the truck’s day was built well.

Now add the way Casella grows. It buys local waste businesses, five of them so far in 2026, worth roughly $165 million of annualized revenue, and it is not alone: the largest companies in the sector spent more than $2.7 billion on acquisitions in the first half of this year. Casella’s Mid-Atlantic region, the one this brief is about, did $341 million in 2025 and is substantially assembled out of purchases.

Here is what a consolidator actually buys, and it is the part the purchase agreement does not itemize. It buys customers, trucks, people, facilities, and revenue, which is the visible half. It also buys a different way of taking an order, a different way of scheduling a pickup, a different way of writing an invoice, a different way of chasing a payment, and a set of local arrangements that live in the memory of the people who made them. One acquired operation knows a customer by a route number. The next knows the same kind of customer by an account name. A third has a billing rule that made sense to a family-owned hauler in 1998 and has never been written down since.

You can own the network on paper and still be running several smaller businesses underneath it. The trucks are yours, the routes are yours, and nobody can safely combine two of them, because combining them means being confident about which record is true.

The customer feels this before the CFO does. A pickup gets missed. An invoice does not match the service that was delivered. A payment needs a second phone call and a person who remembers the account. Inside the company, every one of those looks like somebody’s mistake, because that is genuinely how it presents: someone entered the wrong code, someone at the local operation handled it the old way. The customer is dealing with one company. The company is dealing with several versions of itself.

There is a clock on fixing this, too. Coletta has been pointing at what is happening to disposal capacity in the Northeast: an incinerator that sits, in his words, “right in the middle of our market area” has announced its closure at the end of 2026, and Albany has announced plans to close its landfill and build a transfer station instead. More tons are going to be looking for fewer places to go. Density is about to be worth more than it is worth today, and the company best placed to capture that is the one whose routes are already built well.

The move

Casella started with the least photogenic part of the whole business.

The name for it is lead-to-cash, and stripped of the jargon it means the path a customer request takes from the moment it arrives to the moment the money is collected. Someone takes the order. Someone schedules the pickup. Someone puts it on a route. Someone drives it. Someone invoices it. Someone collects. Someone untangles it when the invoice and the work do not agree. In a business built by acquisition, each of those handoffs is a place where an inherited exception can quietly become permanent policy.

In the second week of May, the Mid-Atlantic customers moved onto one of those paths instead of several. Coletta is plain about what that cost in human terms: “a lot of training, a lot of work with our teams down there,” across dispatchers, drivers, operations staff, and the customers themselves. An integrated customer payment portal came with it, a phone app followed in May, and a rebuilt casella.com landed in July. Different pieces, one purpose: the same customer, the same order, the same route, the same invoice and the same payment, visible to everyone who touches them, including the customer.

Only then does route consolidation become an option rather than a gamble. Casella’s CFO, Brad Helgeson, put the sequence in one sentence for analysts: “We’ve completed our systems integrations and are well into route consolidations.” Coletta described the second half as market-by-market work running through late June and into July, and said it would be “a 5-plus month process.” Nobody threw a switch here. It was dozens of small decisions about which stops move to which truck, each one needing somebody confident that the customer on the corner will still be served on Tuesday.

By the first week of August, 13 routes were gone, along with the trucks and the labor assigned to them. Some of the remaining trucks were converted to automated collection while the work was open.

The AI comes next, and Casella is not shy about it. In-cab technology is deployed across the fleet, coaching drivers in real time toward safer behavior. The workers’ compensation triage program has been expanded. Coletta has told analysts the company is putting money into AI-enabled tools and the data infrastructure underneath them, expecting the back office to need fewer hands over time, and the routing itself is heading the same direction.

Look at where that AI is landing. AI-assisted routing on top of a region where each acquired business described a customer differently would be a very expensive way to produce confident nonsense. AI-assisted routing on top of a region where every stop, every service agreement, and every invoice now sits in one place is a multiplier on work that has already been made true. Casella did not buy the multiplier first.

Worth being precise about what actually changed in May. The migration was an operating decision that happened to require software. What Casella rebuilt was how an order becomes a truck’s Tuesday, and the system was the thing that made the new version of that work possible to run.

The result

The cleanest result is the one that changed the plan: 13 routes, with their trucks and their crews, no longer exist as work Casella has to cover. Nothing about that is a forecast. It is a smaller cost base in a business whose cost base is trucks, fuel, and the hours of the people driving them.

Management has attached targets to it. Helgeson framed the opportunity as taking cost out by running the business with fewer routes after the integration, sized at $5 million in 2026 and another “$10 million over the next 2 years.” Those are targets, and I would hold them in a different hand than the 13. The routes have happened. The dollars are what management expects the routes, and the ones still to come, to be worth.

The quarter around them is doing several things at once, and it is worth seeing the whole picture rather than the flattering slice. Revenue rose 16.9% to $543.7 million. Of that increase, $46.2 million came from acquisitions and rollover, and $32.2 million, or 6.9%, from same-store growth. Solid-waste pricing was up 5.5% while solid-waste volume slipped 0.6%, so price was carrying growth while the trucks themselves were slightly emptier. Landfill tons ran 8.4% higher, about 86,000 additional tons, split between waste Casella brought in itself and third-party volume.

Profit tells a more complicated story than revenue. Adjusted EBITDA rose 12.5% to $123.2 million, but the margin on it fell 0.8 percentage points to 22.7%. Two named causes account for more than the decline. Fuel recovery fees covered the dollar cost of higher fuel and cost about 0.4 points of margin percentage in the process, which is arithmetic rather than a problem. Resource Solutions took another 0.7 points against a strong prior year. Set those two aside and the underlying business expanded margin by 0.3 points. The Mid-Atlantic savings had not yet reached the line in the quarter that was reported. Management expects them in the second half.

The number that gives all of this its stakes is the comparison. Waste Connections earned a 32.8% adjusted EBITDA margin in the second quarter. Republic Services earned 32.1%. Casella earned 22.7%. That gap is roughly ten points of margin on a business generating over half a billion dollars of revenue a quarter, and it is the honest reason to care whether Casella can turn purchased routes into a denser network. A smaller consolidator with more integration still ahead of it has somewhere to go that the giants do not.

Now the honest take.

Casella is running five programs at the same time: buying companies, raising prices, pulling more tons into its landfills, integrating systems, and consolidating routes. The quarter is the sum of all five, and no line in it separates them. Revenue of $543.7 million beat the roughly $526.4 million analysts expected, by about $17.3 million, and that beat belongs to acquisitions, price, and landfill volume. It is not evidence about the route work, and reading it that way would be flattering the story.

The route result itself is strong operating evidence rather than a clean experiment. The Mid-Atlantic is where the sequence, the systems change, and the outcome line up in time, and that is the best available cohort. It is also a region with new acquisitions arriving, trucks being converted, local volume moving, and a new operating team under an incoming COO who brings thirty years in the industry. Any of those moves a route count on its own.

And the AI evidence specifically is earlier than the operating evidence. The one AI-adjacent number Casella has published is a key OSHA metric improving 34% year over year, which Coletta attributes to the in-cab coaching together with the expanded triage program. That is a genuinely good result, and it is a safety result with two causes. It is not proof about route economics, and Casella has not claimed it is.

Which is the part I find most useful about this company right now. The proof it has is the proof it earned, in the order it earned it.

The Craft of AI read

Almost every AI conversation a mid-sized company has in 2026 starts from the wrong end. Something arrives to be bought: an agent, a copilot, a platform with a dot-ai in the name, sometimes a very good one. It arrives before anyone senior has sat with the people doing the work and understood how the work happens, and the understanding is supposed to catch up later. Usually it never does, and the tool ends up pointed at a process that runs on local names, private spreadsheets, phone calls, and a few people who remember which exception applies to which customer.

You know what happens then. The tool works exactly as designed. It produces a faster, more confident version of the same uncertainty, the pilot returns a polite result, and the conclusion in the room is that AI is overrated in this industry.

Casella ran it the other way, and the reason it could is worth naming, because it is copyable. The company had a result available that did not require any AI at all. Put the customers of a group of acquired businesses on one path from order to payment, and routes that overlap become visible for the first time. Thirteen of them come out. The trucks and the hours come out with them. That gain pays for itself in fuel, maintenance and labor, and it would still have been worth doing if the word AI had never been mentioned in the building.

That is the test I would apply to any groundwork project sitting in your plan right now. If the only justification for it is that a future AI initiative will need it, you have not built a business case. You have built a dependency and called it a foundation. Casella’s version paid on its own terms first, and then it happened to leave behind exactly the thing that makes routing software worth buying: one true record of what the work is.

There is a contrast here worth holding onto. The twenty years of maintenance history Pella had to argue for keeping was a story about a company that had been too disciplined to write anything down. Casella’s problem is the opposite shape: everything was written down, several times over, by businesses that had each been right in their own world. It is the same failure in the end. Neither company had work a system could read, and in both cases the fix was an unglamorous argument somebody had to win internally before any of the interesting technology could do anything.

Casella still has plenty to prove. It carries debt at 2.7 times, a large acquisition pipeline, and a margin gap to the two biggest companies in its industry that will take years to close if it closes at all. But the sequence is right, the first gain is real, and the AI is arriving at the moment it can actually do something. Most companies are running that in reverse.

Things to consider

• In a roll-up, what you bought was several versions of the same job. The purchase agreement lists customers, trucks, facilities, and revenue. It does not list the several different ways those businesses take an order, price an exception or chase a payment, and that is the part that decides whether the savings you underwrote ever show up. If you have acquired anything in the last three years, ask how many distinct ways your company currently records the same kind of customer. If nobody knows the number, that is the answer.

• Ask what you could stop doing if the records agreed. Casella’s first visible result was a subtraction: 13 routes, gone. That is a more useful shape of evidence than a new capability, because it lands in the cost base immediately and nobody has to be persuaded it is real. Find the equivalent in your business, whether that is a manual reconciliation, a duplicate approval, a second call to the customer, or a truck roll that should never have been scheduled. The thing you can stop is worth more than the thing you can add.

• Keep targets and results in separate columns. Casella’s $5 million for 2026 and the $10 million beyond it are management’s expectations. Thirteen routes is a fact about the operating plan. Both belong in the conversation, and they do not belong in the same sentence. The quickest way to lose your own board on an AI program is to let its forecast start doing the work of its evidence, because the first time the forecast slips, you have spent the credibility of everything under it.

• Name the assumption the savings live or die on, and put one person on proving it. Casella’s is specific: that integrated records will reveal enough genuine overlap to remove routes without a customer noticing. That sentence can be tested, and a named person can be held to it. Go through the initiatives in your AI budget and see how many have an assumption stated that plainly. The ones that do not are not initiatives yet; they are intentions with money attached, and they will burn slowly for a year under a part-time owner.

• Make the groundwork pay before the multiplier arrives. The Mid-Atlantic migration did not need an AI business case, because fewer routes, cleaner invoices, and a customer who can pay online without a phone call are worth real money on their own. Build your first layer so it clears its own cost in fewer corrections, shorter handoffs, or less truck time. Then whatever intelligence you put on top of it is upside rather than justification, and you are not betting the case for the foundation on a technology decision you have not made yet.

THE WORKBENCH

Do this tomorrow

An hour, one transaction, and somebody who actually works it in the room with you.

Build a workflow truth sheet. Pick the one transaction your business cannot afford to get wrong: the order, the claim, the quote, the schedule, the repair, whichever one carries your revenue. Take one real instance of it, not the procedure document, and walk it across a whiteboard from the moment the request arrives to the moment the money is collected, or the customer has what they asked for. At every step write three things: which record the person is working from, who decides what happens next, and what happens when the record and the reality disagree.

Then mark every point where the same customer or the same transaction changes identity, leaves the main system, or depends on a person remembering an exception. Every mark is a translation your business is paying somebody to perform, over and over, every day, and it is exactly where Casella’s routes were hiding before May.

Price three of them before you leave the room. Take the three biggest marks and put this year’s numbers on each. How many hours a week do your people spend reconciling two records that should have been one? How many invoices get corrected, credited, or chased a second time because the service and the billing came from different places? How many trips, visits, calls, or shipments happen twice because two teams could not see the same thing? Multiply each by loaded cost and write the annual figure at the top of the board.

That figure is operating margin being absorbed right now, this quarter, by work your company cannot see as one whole transaction. No vendor and no budget cycle is holding it up, and it is the honest number that tells you whether your next dollar should buy intelligence or should first buy agreement about what is true.

THE QUESTION

Ned Coletta did not open with a model. He opened with the migration of a group of acquired customers onto one path from order to payment, in the second week of May, at the cost of a lot of training and a lot of disruption for the people who had to learn it. Three months later, 13 routes were no longer work his company had to do.

So take your most important workflow, the one that carries the customer, the cash, or the schedule. If you followed a single transaction through it tomorrow morning, start to finish, how many times would it become a different thing in the hands of a different team?

Be honest about the count rather than answering the way you would answer it in a board meeting. Each of those points is a place where your people are doing the translation that your systems could not, and they are good enough at it that the business runs anyway. That is exactly why it has never shown up as a problem anyone owns.

It is also why the AI you are being pitched this quarter would arrive with nothing reliable to read.

Casella spent the spring making its work true before it spent anything making it intelligent. The trucks came off the road first.

---

Want this done for you?

You just built the whiteboard version: one transaction, an hour, and a count of the places your business translates itself.

My Ground-Up Workshop is the rigorous version of the same exercise, run on your business with the small group of your people closest to that workflow. Every AI strategy you’ve been sold starts with software. This one starts with the people who do the work.

One to two days in person, then a synthesis week, and you walk out with a plan: a ground-truth map of how your core workflow actually runs, the one where a rebuild would release real operating margin, a target operating model with AI built in from the ground up, and a 30-60-90 day plan your own team can start on. I do a small number of these a quarter, for $20,000.

I’m a Partner at geniant. Since 2003, our craftspeople have done one thing: understand how work actually happens, design how it should happen, and build software that works the way people do. When a workshop gives you the roadmap to execute or turns into a build, they’re who does it — one senior-led team, layered on top of the systems you already run, value in weeks rather than months.

Book a discovery call →

— Grant K. Baldwin