DeepFabric lands Kenco as It scales supply chain agents

“He knew the kind of work we were doing with other 3PLs and other supply chain teams, and he knew the philosophy with which we operated and how we built the platform,” said Kalyan Kommineni, founder and CEO of DeepFabric, in an interview with FreightWaves. “He felt that we were a good fit for how Kenco was looking at AI and how Kenco deploys new technology into their environments.”

Starting with a humble proof of concept

Successful AI agent implementations start from humble but manageable beginnings. Start small and move quickly.

“Our goal is to take an agent into production as fast as possible with the minimum possible risks,” Kommineni said. “Before we even think about bringing on someone as a customer, we sign the NDA, we take their data and show how the agent works with their data. That gives them a feel for how it runs in production.”

The proof-of-concept stage exists to find out what breaks before it lands on a live customer account. It’s an important step, given the risks of beta-testing AI agents in live operations rather than in a closed, separate virtual twin. That approach minimizes potential disruptions while the agents are fine-tuned.

“Now I will say agents fail all the time, but we make sure that is happening in a non-production environment,” Kommineni said.

Kenco absorbed none of that failure on the way to production. All six agents went live without disrupting service to a single customer, according to Chief Operating Officer David Caines.

Why automating audit work was a massive win

The 3PL’s own structure created the opportunity. Kenco operates 141 distribution facilities and 43 million square feet of warehouse space across 33 states and Canada. Every handoff in that network is a document somebody has to check.

“There’s a lot of work — manual labor — that happens in audit,” Kommineni said. “Again, this is not good labor. This is something that operators do not want to do, that they would want to punt off to an AI agent. So we’ve gotten really good at the entire audit piece. And the audit is not one thing. There are different kinds of audits working with different parties.”

In a 3PL, that means the warehouse validating the carrier, the carrier validating the customer and the customer validating the warehouse, each one a separate reconciliation with its own paperwork.

“Every connection between these links is a failure opportunity, and we do everything it takes to make sure that these agents are not failing in production,” Kommineni said.

The freight auditor is one of DeepFabric’s three most widely deployed agents, alongside the proposal manager and the inventory manager. The company reports audit-spend reductions of 45 percent and request-for-proposal response times cut by as much as 30 percent.

Setting the goalposts on supply chain agents

Benchmarking is where most buyers struggle to find an easy answer. The question of what to measure comes before what to automate. Kommineni frames it as an effectiveness question over an efficiency question.

“As a 3PL, as a company, or as a team within a company, what are you trying to accomplish here?” he said. “Are you trying to do more work with the same people, or are you trying to sometimes even add more people but then do a lot more work? And then there are sometimes, ‘Hey, are you trying to reduce the labor hours?’ So there are different lenses that customers take. Depending on that, the KPIs are driven.”

The measurement changes with the agent: The proposal manager is scored on how many more bids a team pursues without adding headcount, and an audit agent is scored on manual hours spent per 100 invoices.

“KPIs are agent-dependent, but we help our customers understand what the pre-agent KPIs are, what the baseline is, deploy the agent, understand what it is after the agent is deployed, and then measure the results,” Kommineni said.

Without that pre-agent baseline, nothing can be measured effectively.

Somewhere in the middle on models

The other question buyers must figure out is which model to build on. The answer Kommineni gives involves taking a balanced approach.

“So there are two different dichotomies here. There is, ‘Hey, there is one model that does it all.’ There is, ‘You know, 20 different customized models that I’m going to deploy into my environment for my use cases.’ The answer is always probably somewhere in the middle,” he said.

DeepFabric routes each use case to whichever model fits it on capability and price, and the company carries the exposure that comes with that.

“We take the risk of managing all those. We take the risk of managing the token economics. We take the risk of the execution as well, so that our customers don’t have to think about the thing that changes every day,” Kommineni said.

Kenco draws its own line between the two categories of AI investment it is funding: agentic AI, which completes tasks inside defined guardrails, and generative AI, which produces content and finds insights. DeepFabric powers the agentic layer, and Kenco’s teams keep approval authority over agent output through inspection and override paths built into the platform.

Kommineni traces the execution-first posture to what his team could not deliver in a previous job. As a former consultant, he said execution was always a challenge despite giving clients a plan.

“But with DeepFabric, we actually took that to heart in terms of execution. We actually will help you execute what we would have suggested as consultants so that you don’t have to think about, ‘Hey, how am I going to execute these agents in production,’” he said.