Ask the internet how to drop a Chevy V8 into a 1978 Jeep CJ and sooner or later you land on Novak’s website, often not on a product page at all. A large share of Novak’s traffic goes to its research library: wiki-style write-ups on engines, Jeeps, transmissions, cooling, and just about every automotive system a powertrain conversion touches.

Novak Conversions has been designing and manufacturing Jeep powertrain conversion parts since 1967, first in California and, since 2000, in Logan, Utah. About 20 people work there across a machine shop, a transmission rebuild shop, metal fabrication in steel and aluminum, inventory and shipping, and sales and support. Everything is designed, machined, stocked, shipped, and supported in house, and a single order can run from $3 to $30,000.

All of that knowledge sits behind the sales inbox. When a customer asks whether one adapter will mate a specific engine to a specific transmission, the answer is either in the salesperson’s head or somewhere in 58 years of catalogs, instruction PDFs, and research pages.

A salesperson drafting a reply used to search that catalog by hand to pull the right reference material. Today it takes seconds. They open the conversation, hit Ask AI, and Missive queries Novak’s own system, running on a server in the building, retrieving from Novak’s actual catalog and documentation. Back come the part numbers, the fitment notes, and a starting draft. The salesperson checks it against what they know, edits it, and sends it. What changed isn’t who answers the customer. It’s how long it takes to put 58 years of documentation in front of the person writing the reply.

A few months ago, the same question started with logging out of one webmail account and into another.

Two inboxes, one login at a time

Austin, Novak’s CEO, didn’t go shopping for email software out of curiosity. Customer demand was running high, and rather than hire a salesperson blind, he took the desk himself so he’d know the job well enough to train the next person into it. That put him in the sales inbox every day.

What he found was RoundCube webmail on the company’s own server. Novak ran an info address and a sales address, each managed differently depending on who happened to be running it, plus everyone’s personal address on top. Switching meant logging out and logging back in. Other employees flexed into sales when someone took a day off, which meant working out of another person’s inbox and another person’s system.

Threads got forwarded around so people could ask each other questions, with private notes riding along inside the reply. Two reps could pick up the same customer at the same time without either one knowing. “You can’t really see the full conversation of what’s going on sometimes,” Austin says.

Then the channel count started climbing. Marketing had picked up Facebook and Instagram, so product questions were arriving as DMs. Novak was adding SMS. A chat widget was going onto the website. The reps told him plainly that it wasn’t sustainable.

“I just can’t keep up with all these channels, and the customers are not getting the attention they deserve. I’ll be over here, and then I’ve changed windows, and this guy’s been waiting.”

Why Missive was the last one he tried

Austin searched Google, jumped on forums, and asked friends and coworkers, then started signing up: Hiver, Jelly, Monday (which he decided was more complex than the problem called for) and Front, where he got as far as building out a full account.

Front nearly won it. What decided it was how mail needed to flow for the way Novak works. Austin wanted the company’s email to stay in Novak’s own accounts as well as the shared tool, so that the underlying mail was never dependent on the tool being switched on. “With a two-way sync I could open my old email at any point and still see everything current,” he says.

Missive was the last one he tried. “I tried all these other ones, and then I tried Missive and I’m like, wow, this one’s definitely different, and it’s definitely in line with what I’m looking for.”

It was also a lot at once. Coming from traditional webmail, archive versus close, team spaces, tasks, and chat all landing on day one was overwhelming. So he turned nearly all of it off and kept the one thing he’d come for: a shared inbox any salesperson could open and search. Everything else got added back later, once it had a reason to be there.

The rollout he had to do twice

The first attempt did not go well. Austin added everyone at once and got the same response from most of them: I have no idea what’s going on, this is so confusing. So he cleared it out and started over with the features off.

Ten years of mail re-synced, history archived up to a cutoff point, tasks switched off for everyone until he understood what they were for. Then he sat in the sales seat a couple more days, worked out how he wanted to run a day, and went person by person to set each of them up by hand.

The work was never really the software. Everyone had spent their career doing email their own way, in private, where it never had to match anyone else’s method. Some people file everything into folders, some never archive anything, and neither habit is wrong until two of them share a queue.

A team inbox asks out loud what most teams have never had to answer: what archived means here, who owns a thread once it’s claimed, and when something is finished. Somebody has to make the call and then hold everyone to it.

Austin made himself that person. “That’s why I was the one doing it. Hey, you’ve got to listen to me, this is the way it’s going to work. It might be confusing at first, but it’s going to be better in the long run.”

A research tool in the sidebar

The AI build ran in parallel, which is why the whole move waterfalled together. Novak’s IT lead took the catalog, the instruction PDFs, and the research library and indexed all of it, so the assistant retrieves answers from Novak’s own documentation rather than guessing at them. It runs on a local server in the building. From there they connected it to Missive as a custom MCP integration, with Claude doing the drafting.

“Missive will go talk to our MCP, get the information about our company, and then bring it back and formulate an email,” Austin says.

Austin is direct about where the line sits. The assistant is a research tool, not a rep. It pulls the relevant specs out of the documentation and assembles a draft; a technical salesperson verifies the fitment, adjusts it to the customer’s actual build, and decides whether it goes out. On a $30,000 conversion, or a $3 part that has to be right, that judgment isn’t something Novak is looking to hand off.

“The AI doesn’t talk to our customers. It finds what we already know and puts it in front of the guy writing the email. He’s still the one who decides if it’s right for that build.”

Before that, using their own AI meant copying the question into another window and pasting the answer back. Now the assistant is in the conversation and already has the thread in front of it. The search step collapsed. The review step didn’t move.

The real payoff is on hiring. A new salesperson has the whole catalog behind every reply on day one instead of building that recall over years, which is exactly what Austin needed while training someone into a product line going back to 1967.

Chat, cart, and buying history in one thread

The newest piece is a website chat widget, also built in house, that drops conversations straight into Missive. Next to each one, a custom sidebar shows what the customer has in their cart, which page they’re reading, and what they’ve bought before, matched on their email address. Novak’s SMS conversations are next to land in the same place, so the reps stop hopping windows to see who was waiting.

Fifty-eight years of Jeep powertrain knowledge used to be spread across a research library, a stack of instruction PDFs, a couple of long-tenured people, and two webmail logins. It’s now in the same window as the customer asking the question. “I don’t think I’d ever be able to go back,” Austin says.