As hype around AI continues to give way to real-world implementation, the challenge for businesses is no longer about experimentation but working out how to make the technology work at scale.
By automating routine tasks and strengthening customer interactions, businesses are moving from isolated use-cases to embedding AI across entire operations.
So how can scaling companies adopt AI solutions without complicating an already complex tech stack?
This question was at the forefront of discussion at our most recent Sifted Talks panel where experts unpacked how to collaborate with AI, strengthen customer and team connections and how the technology can be used to scale with confidence.
Our panel of experts included:
- Monique Koster, head of small business EMEA at video software platform Zoom
- James Mulligan, head of strategy & programs at AI simulation software company PhysicsX
- Emma Burrows, cofounder and chief technology officer at AI prototype developer Portia AI
- Harriet Allardyce, client success manager at used car sourcer Sellmyride
What makes a company ‘AI-first’?
AI-first companies don’t always have the flashiest of tools, says Koster. In fact, they are usually the ones who have cracked AI adoption the best.
“At Zoom, we don’t have AI bolted on,” she says, “it’s built into how we work. When I join a meeting, my AI companion has already prepared context for me and then captures what matters. That’s what AI-first actually looks like not adding more tools to your stack.”
The companies that succeed start by identifying a real problem and then let AI become part of the solution, she adds. “The ones who struggle are the ones still stuck in pilot mode.”
Burrows says teams need to identify a “crunch point” in workflows and craft the right tool around it.
“The failure mode lies in companies who are not going far enough in understanding that if they solve a problem, where is that bottleneck going to go?,” she says, “What does that mean, not just for the problem they have today, but the problem that they'll have in a month."
That makes the difference between a team experimenting with a tool for a short period of time and having an AI strategy that has longevity.” - Burrows
Importance of data
Clear guardrails can be useful in avoiding access to incorrect or sensitive data, says Burrows, but heavy-handed policies can slow adoption.
“A better approach is to think about the kinds of ways that people are informally sharing best practices within the company which is more effective than specific top-down guidelines.”
Large language models — the underlying technology powering AI tools like OpenAI’s ChatGPT and Anthropic’s Claude — have been trained on millions of data points lifted from the internet, Mulligan says. But PhysicsX is training models purely based on physics.
“That requires us to find and create our own data. If we're going to predict the outcome of a numerical simulation then we need a whole stack of potential simulations to train that model,” he says. “The more data we have, the better the model becomes."
There is a world where you have foundation models for physics and chemistry that can predict the aerodynamics of a vehicle, for example, without even having seen that initial design. Getting this kind of data can sometimes be a bottleneck.” - Mulligan
Avoiding skepticism
Employees at PhysicsX have overcome skepticism by utilising AI without formal approval from IT departments, in what Mulligan calls “shadow AI” applications.
“Shadow AI is the strongest signal for us,” he says. “All of our engineers are using AI personally and can explain the value of it to their actual job.
“They can then communicate that as champions to the rest of the organisation. I would embrace shadow AI if I was looking to adopt AI at scale at other organisations.”
When first rolling out the technology, PhysicsX also had an “experimental budget”, Mulligan adds, where employees could be given $200 to be used on a Claude Max account for example.
As someone in a customer-facing role, there is often an element of uncertainty around how AI can be used to help, says Allardyce. The company has been using the technology in providing feedback from client calls.
“Everybody who's worked in sales has had an experience where a client they felt connected to has said they want to part ways. With AI, I'll use data and analytics to ask for constant feedback.
If I feel that I've had a positive client call, I will factcheck it with AI that doesn't have bias. It's a fantastic training tool and it can help you more accurately forecast relationships.” - Allardyce
Measuring and proving the value of a product
The majority of Sellmyride’s clients and employees are based in the US, adds Allardyce, meaning a lot of work risked being lost when working across different platforms and teams.
“When I first joined, we didn't have a centralized CRM and everybody was using AI here and there which wasn’t integrated. Choosing something that could maximise how many people were benefitting from it rather than teams working in silos was really important, especially when going to the CEO with something that could meaningfully scale our business.
“I'll sit at meetings and my summary will be shared with colleagues in the US who can go and pitch to their sister businesses based on those findings. It takes out a lot of verbal communication as it's all backed in data.”
Measuring and proving the value of an AI tool can be done easily by first bringing to light the pain point or bottleneck to a senior leader, adds Koster.
“I see in small businesses there's a lot of pain points around time and resource savings. Connecting AI to a pain point in an organisation is what makes the technology so valuable and it can often help to explain its worth to employees internally.” - Koster
The Zoom Scale-Up Summit is a virtual event bringing together EMEA business leaders to explore how modern work is evolving. Join customers and industry experts to discuss practical strategies for leveraging AI in collaboration, communication, and customer engagement. Click here to learn more.
