Michael Delgado, Co-Founder and CEO at Canals – Interview Series

Michael Delgado, Co-Founder and CEO at Canals, is a former corporate lawyer turned entrepreneur who has built a career bridging legal expertise, product development, and operational technology. After starting at top-tier law firms including Cravath, Swaine & Moore LLP, he moved into startups, taking on leadership roles at Willing before co-founding Vested, which was later acquired by MetLife. He went on to found Canals in 2022, applying his experience across law, operations, and product to tackle inefficiencies in traditional industries, particularly by leveraging AI to modernize complex business workflows.

Canals is an AI-driven platform designed to automate critical back-office operations for wholesale distributors, including sales order processing, accounts payable, and procurement. The company focuses on transforming unstructured inputs such as emails, PDFs, and handwritten documents into structured, actionable data that integrates directly into existing ERP systems. By continuously learning from user interactions, Canals reduces manual data entry, minimizes errors, and accelerates operational workflows, positioning itself as a practical execution layer for businesses rather than a purely analytical AI solution.

You transitioned from a legal background at firms like Cravath, Swaine & Moore LLP into startups, eventually founding Canals after your experience building Vested. What specific breakdowns in distribution workflows pushed you to start the company, and how did your earlier roles influence that decision?

My wife runs a distribution business, so it was through her that I first started visiting warehouses, talking to distributors, and learning the industry.

As I spent more time immersed in distribution, what stood out most was a process called “sales order entry.” Orders come to a distributor through a wide variety of channels in a wide range of formats, and each one needs to be reviewed and manually keyed into an ERP. It’s time-consuming work that falls on teams of sales reps—people whose jobs are supposed to be driving revenue and building relationships.

The more conversations I had with distributors, the clearer it became that this was not a small inefficiency. Sales order entry is a core workflow in a massive industry that technology had historically failed to serve, in part because traditional software couldn’t handle the variability. I’d spent years building software and following the advancement of AI, so I was well equipped to see a large market, a real pain, and a new way to solve it. Canals grew from there.

For readers new to this space, what does Canals actually do inside an organization on a day-to-day basis, and how does it interact with existing systems like Enterprise Resource Planning (ERP)?

At a high level, Canals takes the inputs distributors, contractors, and manufacturers deal with every day–emails, PDFs, spreadsheets, even handwritten notes–and turns them into structured data that can flow between system and power end-to-end workflows. It then uses that data to automate downstream actions, whether that’s generating a sales order or submitting an invoice, before pushing clean, validated data directly into an ERP.

The ERP remains the system of record, while Canals acts as the operational AI keeping it accurate and up to date.

Industrial distribution still relies heavily on emails, PDFs, and phone calls to manage orders and invoices. Why has this level of manual work persisted for so long, and what has prevented meaningful automation until now? 

The problem is that traditional software depends on rigid rules and standard templates. That works in environments where inputs are consistent, but construction and distribution are not like that. Documents come in a wide variety of formats, and there are dozens of different names, shorthand terms, and field slang that all describe the same product. At a certain point, the number of edge cases becomes unmanageable. You can’t realistically define rules for every variation, so the process falls back to manual interpretation.

The will to introduce more efficiency has always existed, but until recently the technology couldn’t keep up, making earlier approaches difficult to implement and impossible to scale.

A core challenge here is turning unstructured inputs into structured actions. How does your platform interpret emails, attachments, and documents, and convert them into usable data and workflows?

It’s a challenge that requires two steps to solve.

The first is parsing. Canals identifies the relevant documents in a user’s inbox, pulls out the key line items and fields, and extracts the data.

The second is matching. This is where that extracted data is resolved within the system. In some cases, that means mapping line items to the correct SKUs, handling variation in how products are described, and normalizing units. In others, it means reconciling documents, such as matching an invoice to a purchase order and receipt, aligning line items, and identifying discrepancies.

The result is structured, contextualized data that can drive an end-to-end workflow.

You’ve supported workflows tied to over $2.1B in payables. At that scale, what patterns emerge around inefficiencies, delays, or errors that most companies don’t even realize they have? 

There are some obvious efficiency gains. On the accounts payable side, for example, our customers automate 96% of their invoice processing on average, which removes a significant amount of manual work.

What’s more interesting though is how that manifests beyond cost savings. In order entry specifically, speed directly impacts revenue.

In construction timing is critical and staying on schedule is the priority. If a contractor is requesting quotes from multiple distributors and one responds in ten minutes while the others take hours, the job is usually going to the one who responded first even if it’s not the lowest price. Getting the material on time matters more than saving a few dollars.

That dynamic has a direct impact on revenue. Automating sales order entry increases how often a distributor is the first to respond, which increases how often they win business. For one of our customers, that translated into 57% of their transactions turning into orders, compared to a previous average closer to 20%.

Legacy systems like ERP platforms are often rigid and difficult to modernize. How do you approach integration without forcing companies to rip out their existing infrastructure?

ERPs are deeply embedded in how a business runs, so the real constraint isn’t just integration, it’s how fast and how cleanly you can integrate without adding overhead. If implementation is slow or requires heavy involvement from internal IT, it becomes a disruptive blocker.

Our approach has always been to invest in making our implementation fast and frictionless. We have dozens of pre-built integrations with a large team of engineers to support custom deployments, and we prioritize getting customers up and running quickly without creating an ongoing maintenance burden.

We’re seeing a shift toward more autonomous systems across industries. How far can automation realistically go in distribution workflows before human oversight becomes critical again?

There are plenty of things AI can’t do. It’s not going to make complex business decisions, manage customer relationships, or operate in the field. What it can do is remove a lot of the repetitive administrative work that sits underneath those processes.

In most industrial workflows, the right model is human-in-the-loop where AI handles the bulk of the work while leaving people in control of the exceptions. When something is straightforward, it can be automated. When something is ambiguous, high-value, or carries real risk, that’s where human judgment is critical.

The goal isn’t 100% autonomy. It’s to automate the tedious, manual, and routine parts of the workflow so people can focus on high-value decisions and exceptions.

One of the risks with automation is losing institutional knowledge from experienced operators. How does Canals ensure that expertise is captured and reflected in the system rather than replaced? 

One of the key advantages of AI over traditional software is that it can learn over time.

When an experienced operator reviews something, makes a correction, or handles an exception, the system can capture those decisions and intelligently apply them going forward. As usage increases, it starts to reliably reflect those patterns instead of relying on a fixed set of rules.

That means institutional knowledge is no longer tied to a single person. Instead of living with individuals, it gets baked into the systems used to run the business, so it’s applied more consistently across the organization. When experienced employees leave, their expertise stays captured within Canals. When new employees start they’re working within a system that already reflects how the business operates, which helps them ramp faster and execute more consistently.

The surge in data center construction is putting real pressure on supply chains. How is that demand changing expectations around speed, accuracy, and coordination for distributors? 

The race to build data centers is accelerating with $700 billion being poured into construction, putting immense pressure on contractors and distributors to keep up.

What that demand changes is the tolerance for delay. Workflows that were manageable at lower volumes—like manual order processing and document reconciliation—start to break down at scale. As projects get larger and move faster, gaps between quoting, purchasing, and fulfillment become more visible and more costly on both sides of the transaction. The lack of accurate, up-to-date information undermines coordination and can result in unexpected delays and sudden work stoppages.

The teams that can operate with speed and real time visibility have a clear advantage. At that point, automation isn’t just about efficiency, it becomes a requirement to keep up with the pace and complexity of demand.

Looking ahead, how do you see AI reshaping procurement and supply chain workflows over the next five years, particularly as systems move from assistive tools to more agent-like decision makers?

It’s hard to say with any degree of certainty, but what’s becoming more evident is how AI is being applied—narrowly, in specific workflows where there’s a lot of repetition and a clear path to reliability. In procurement and supply chain, that shows up in execution-heavy processes. These workflows are tied to real dollars and real relationships, so the bar for autonomy is high. The near-term shift will be less about agent-driven decision making and more about expanding what can be handled reliably, with people staying closely involved where it matters.

Thank you for the great interview, readers who wish to learn more should visit Canals.