Allan Malcolm has a wonderfully inconvenient message for anyone hoping artificial intelligence will turn a marketing department into a vending machine. The machine needs a system. It needs decent ingredients. It needs to learn how the company speaks. And, even after all that, someone with taste must remain close enough to pull the plug on a sentence that sounds as if it were composed by a committee of microwaves.
This is not the sermon of a nostalgic copywriter guarding a typewriter. Malcolm is the chief marketing officer of Voxel, an AI company that uses existing cameras in industrial workplaces to produce safety and operational insights. His job is to market artificial intelligence while applying it to marketing. If anyone has an incentive to skip the disclaimers and hurry toward the future, it is him. Instead, he talks about architecture.
“Good inputs equal good outputs, bad inputs equal bad outputs,” he told Sara Faatz on the 10 Minute Martech podcast in June 2026. The sentence will never frighten a philosopher, but it might save a marketing budget. Before the agentic workflow comes the less glamorous audit: Is the go-to-market stack ready? Is the data usable? Has anyone written down the brand voice with enough precision to teach it?
“AI does not replace your team. It’s another person on the team.”Allan Malcolm, 10 Minute Martech
The operator underneath the CMO
Malcolm’s fondness for systems makes more sense when his career is read from the bottom up. Before the C-suite title came customer operations at Cypher Systems Group. Then came eight years at Autodesk, where his responsibilities included demand generation and marketing operations. At Trimble, he became vice president of marketing. By the time he joined Voxel in 2025, he had spent two decades moving between the creative front of marketing and the machinery behind it.
Those companies form a neat, almost suspiciously coherent chain. Autodesk gives designers digital tools for making physical things. Trimble connects digital information with construction, agriculture, transportation and other work in the world. Voxel points computer vision at warehouses, factories and distribution centers. Malcolm’s career has repeatedly brought him to the border where software meets steel-toed boots.
He works from LaSalle, Ontario, just south of Windsor, for a company headquartered in San Francisco. Distance has a way of punishing vague communication. A distributed executive cannot rely on the fortunate hallway collision or on a raised eyebrow across a conference table. The logic has to travel. So does the voice.
First, teach the machine some manners
The fashionable promise of generative AI is immediacy. Ask, receive, publish, congratulate yourself before lunch. Malcolm describes a slower apprenticeship. A company must train internal models on voice, tone, positioning and emotion, much as it would bring a new employee into the culture. The early work may be quick and fluent. It may also be entirely unsuitable for a customer.
He is admirably blunt about the failure mode. Let an untrained system become the copywriter and, he warns, the result can be “a bunch of LLM garbage.” In the interview, he estimated that the first outputs, perhaps even the first weeks of outputs, may belong behind the workshop door. Language can be grammatical and still feel vacant. A model can put every fork in the proper place and still ruin dinner.
Malcolm’s order of operations
- Audit the go-to-market stack before adding agentic workflows.
- Make the underlying data usable and relevant.
- Define the capabilities the team actually wants to augment.
- Train models on voice, tone, positioning and emotional register.
- Put human judgment at the start of the workflow, not merely at approval.
The distinction about human involvement is important. A person stationed at the end of an automated assembly line can correct a typo or reject an especially wooden paragraph. A person involved at the beginning decides what the line should make, what evidence it may use and how its work will be judged. The first role is proofreading. The second is authorship of the system.
Where he lets AI run
Malcolm is not timid about the technology. He simply gives it jobs suited to its talents. Competitive intelligence is one. A purpose-built agent can map a rival’s positioning, apparent customer profile and messaging far faster than a person starting with fifty browser tabs and a rapidly cooling coffee. Ideal-customer-profile discovery is another. Models can sift existing customer information for patterns that deserve a marketer’s attention.
The bars express Malcolm’s stated sequence, not measured performance scores.
Notice the shape of the division. AI begins with breadth: scanning, comparing, grouping and finding the odd detail. Humans remain responsible for consequence: choosing a position, understanding a customer and deciding what the company is willing to say in public. The machine widens the field of view. The marketer decides where to stand.
This is also why Malcolm’s preferred hiring profile is changing. He argues that marketing needs more generalists, people who may have depth in demand generation, brand or content but understand how the other functions interoperate. AI makes production abundant. It therefore makes orchestration scarce. The valuable marketer is less likely to be the person who knows one button better than everybody else and more likely to be the person who understands which buttons should never be pressed together.
“I think the future of our profession in marketing is we all have to have generalist mindsets.”Allan Malcolm
A safety story with an operations spine
At Voxel, these ideas live inside an unusually physical story. The company’s platform works with camera infrastructure that industrial sites already have. It turns video into alerts and patterns around workplace risk and operations. Voxel says it is deployed at more than 250 sites and describes a 90-person team. In October 2025, the company said an Ericsson Ventures investment brought total funding to $64 million.
The marketing challenge is delicate. Computer vision, machine learning and operational analytics can quickly dissolve into a fog of abstractions. Industrial safety cannot. It concerns the ordinary choreography of a working site: a vehicle reversing, a blocked path, a repeated near miss, a practice that looks harmless until the hundredth repetition. The product story has to move from model capability to what a safety leader can notice and act upon.
Malcolm joined after Voxel announced a $44 million Series B. In his public note about the move, he thanked Eclipse investor Lior Susan for the introduction and acknowledged leaders from Autodesk and Trimble who had shaped him. It was a graceful little map of professional connection: the investor who opened a door, the former colleagues who built the judgment carried through it, and a new team trying to apply AI to the industrial workers who rarely appear in glossy automation montages.
His broader record supports the operator’s temperament. He mentored at the University of Windsor’s EPICentre in 2015. He completed coursework in cloud architecture and Industry 4.0. He is a member of the Forbes Communications Council. None of these facts is a fireworks display. Together they show someone persistently curious about the infrastructure around the message.
The useful luxury of saying “not yet”
Marketing technology is sold in the future tense. It will personalize, predict, compose and optimize. Malcolm brings the verbs back to the present. Audit. Prepare. Train. Experiment. Evaluate. His team’s AI practice, as he describes it, permits failed experiments and quick learning. “This is a period of experimentation,” he says, which is both obvious and oddly brave in a business culture that prefers to announce mastery before the documentation has loaded.
The refreshing part of his philosophy is not caution by itself. It is selective impatience. Move quickly on research jobs where the system can create leverage. Take more time where a clumsy sentence can make a company sound identical to everyone else. Hire people who can see across the operation. Make the workflow intelligent before asking intelligence to flow through it.
That view may prove more durable than any particular model. Tools change. The organizational questions remain impolitely stable. What do we know? What are we trying to do? Who is responsible for the result? What should this sound like when it reaches another human being?
Malcolm has spent his career around companies that translate digital information into action in the physical world. At Voxel, he now performs a parallel translation inside marketing. The AI can scan the landscape, help find the patterns and extend the team’s reach. The human still chooses the meaning. For a profession anxious about being automated, that is neither a consolation prize nor a clever slogan. It is a job description.