Augie now connects to MasterMind TMS. Learn more.

Augie now connects to MasterMind TMS. Learn more.

Augie now connects to MasterMind TMS. Learn more.

Augie now connects to MasterMind TMS. Learn more.

Deep dive

Building voice AI for supply chain & logistics

Check calls are one of the hardest tasks in supply chain to automate because conversations are complex and accuracy matters.

Harish Abbott

CEO, Augment

Building voice AI for supply chain and logistics

Companies in all parts of the supply chain make many thousands of check calls every day. These calls are the essential confirmation of where goods actually are, which trucks are on schedule, and when things aren’t going according to plan. Check calls are also one of the hardest tasks in supply chain to automate because conversations are complex and accuracy matters.

Automation saves time, if it works

Most of the time operators spend on a check call is time spent waiting. About 4 in 10 dials never reach a person. The median phone tree wait when you do reach a human is 34 seconds. We want operators spending their time with drivers and customers, not on hold.

When someone picks up, they might be driving, on a break, or in the middle of an unload. Many of them are not native English speakers. Operators must work around these exceptions on every call just to get the information they need.

Phone tree graphic showing complex phone menus and long holds

Complex phone menus, long holds, and often you still hit a voicemail

Voice agents can handle these calls, but only with a different approach than the off-the-shelf frameworks built for casual conversation or normal customer service. A check call is a special case because every answer is a claim about the physical world: where a truck is, when it will arrive, how cold a trailer is, and so on.

To get the update without frustrating drivers, agents should triangulate information as much as possible before making a call. We designed Augie, our AI teammate, to do just that. Augie calls a driver only when critical information is needed. On calls, Augie aims to receive and verify the necessary information with minimal frustration for the person on the other end of the line.

Trust but verify

A good dispatcher doesn’t interrogate a driver. They take sensible shorthand at face value. But they also know the load details, so nothing gets past them.

We built Augie to work the same way. Augie understands the way normal people talk: For example, “Louisville” is Louisville, KY (unless the load isn’t anywhere near Kentucky). “Just arrived” tells you what happened and when without demanding a time. Everyone on these calls is busy and every extra question opens up the possibility of frustration.

Augie also checks every answer against what we already know like route, appointment, and time of the call. We then only ask for what we need and only press when the answer is critical.

We know this is Gainesville Virginia because it’s right on the route. No need to confirm the state and no risk of placing the load in Florida or Georgia.

The driver doesn’t know which state Ashton is in (Maryland, but fair to be confused since it’s only a few miles from both Pennsylvania and West Virginia). But Augie does.

When the agent committed to a location in the calls we audited, it was right 96% of the time. We would rather ask one good question than write down a wrong answer.

Active confirmation

Good dispatchers repeat things back as they hear them. Active confirmation makes drivers feel heard and also gives them a chance to correct us before a wrong answer causes an escalation. We built that playback behavior into every call.

Both our transcriber and human reviewers have trouble understanding this temperature, but a quick playback gets the right value.

Typical human voice characteristics matter just as much. Pauses, fillers and short acknowledgments all help determine whether a driver stays on the line. At Augment, we calibrate voice characteristics through testing on real calls, not in a studio.

The natural back-and-forth adds a little time, but feels more natural and gets more reliable answers.

Cases that no transcriber gets right

Non-native speakers. Heavy accents. Small towns. Five cities with the same name. Load and PRO numbers in every format imaginable. The range of possible answers is enormous, and even frontier general-purpose speech recognition has no way to narrow the possibilities down.

Augie has knowledge of the load, the route and the stop, and that context allows us to parse through speech for the information operators need.

Even a great transcriber won’t guess “Pomona” here, but they can get close enough that we can narrow it down and confirm (close to a stop and matches the route timing).

Why a good prompt is not enough

It’s tempting to think a good model with a clever prompt can get check calls right on its own. But a good agent has to be tuned to handle very messy inputs and offload key validations where the sensible-looking first pass of even a very smart model can be wrong.

At Augment, we build our voice solutions on real road audio. We replay production calls every time we update the agent. We tuned our location checker on many hundreds of tricky labeled answers from real calls. And we calibrate how often and how intensely we follow-up. Ask less and bad data gets through. Ask more and drivers stop picking up.

Augie patiently waits for stalled readouts of identifiers that callers often get confused.

Augie knows what’s close enough and when a precise location isn’t necessary.

The bar for voice AI

The supply chain runs on phone calls. Every call carries a claim about the physical world. A voice agent earns trust when it knows the context before it dials. Augie asks only for information it needs and confirms what it hears.

We will continue to tune Augie on real calls, because the road always brings a new edge case. Every person on the line should feel heard. Every operator should get time back for the conversations that need a person. Our work on voice has just begun.

Meet Augie, the AI teammate built for supply chain.

Augie handles the daily grind so your operators can focus on service and scale.

PRODUCT

Freight Brokerages

FTL Carriers

LTL Carriers

Private Fleets

An AI teammate that sees every instruction through to completion.

Play

FOLLOW US

PRODUCT

Freight Brokerages

FTL Carriers

LTL Carriers

Private Fleets

An AI teammate that sees every instruction through to completion.

Play

FOLLOW US

PRODUCT

Freight Brokerages

FTL Carriers

LTL Carriers

Private Fleets

An AI teammate that sees every instruction through to completion.

Play

FOLLOW US

PRODUCT

Freight Brokerages

FTL Carriers

LTL Carriers

Private Fleets

An AI teammate that sees every instruction through to completion.

Play

FOLLOW US