“Ultimately, you're the one who has responsibility for the work.”

Kong Aik Lee, Covarage

Tell us a little about yourself, your background and what led you to build Associum.

Wes Barnes

My name is Wes Barnes. I've lived in Asia for 26 years, across Japan, Hong Kong, and Singapore, with about 14 years in Singapore.

I'm a co-founder of Associum.

We help finance and consulting firms deliver their work, whether that's a slide deck, report, or spreadsheet, from the raw data all the way through to the final product they can share with clients.

I started working on this because after more than 10 years in private equity. I saw that we could use data, data science and AI to improve our processes and ultimately invest better.

As we developed those ideas, we realized that what we were doing in private markets applied much more broadly, to anyone working in finance or consulting.

The workflow is fundamentally about research and analysis, culminating in some kind of analysis or deliverable that gets shared with the other stakeholders involved in the decision.

Kong

A lot of people might look at that and say, "It's Excel and PowerPoint. Why can't I just use ChatGPT? Why do I need Associum?"

Wes

The short answer is that our results are better and more cost-effective.

Associum is what you call an AI harness, focused on professional knowledge work. It coordinates the different models you're using, the data coming in, the data you already have and the agents processing it.

When you use Associum, you can see the difference in the quality of the output, both in terms of the visual representation and its adherence to the examples or templates you've provided.

There're also detailed citations and the ability to go back and see where all the information came from.

That transparency is important because it reduces the risk of hallucinations and other problems that can occur with AI systems.

We're also able to choose different models for different activities within the harness, which allows us to manage costs more effectively.

Kong

You mentioned being able to trace information back to its source.

Another founder I interviewed recently runs a legal AI platform. He made an interesting observation: AI being 98% correct can sometimes be more dangerous than being obviously wrong because the output looks trustworthy.

Do you agree?

Wes

I think we have to change how we work.

We're very used to applying critical thinking when we're dealing with other people. But AI appears so powerful that it's easy to simply take what it gives you and proceed with your work.

We shouldn't do that.

We need to treat it as though someone has given us a report. We need to read it thoroughly, apply our own judgement and go back and understand the sources behind it.

I like to compare using Associum to working with a team of analysts or associates that you manage.

They're smart people. You ask them to conduct an analysis and they come back with a great report.

As their director, you wouldn't simply take the report and immediately give it to the client.

You would read it. You would ask questions. You would ask, "Where did this come from?"

You're relying on your team to ensure the information and citations are correct, but you still verify the work.

AI should be treated similarly.

If Associum gives you a 40-page report in 20 minutes, it's still your responsibility to read it, sense-check it, and go back to the original sources to make sure you're comfortable with them.

Ultimately, you're the one who has responsibility for the work.

Kong

And Associum helps the user do that by maintaining those sources and citations?

Wes

Yes.

One of the ways we minimize hallucinations is by grounding the information first in the documents or data sources you've provided, rather than relying initially on the model's training data.

We also put citations throughout the documents and spreadsheets that are produced, so you can cross-check where the information came from.

Finally, we have an agent that runs after something has been created to audit and double-check the work before giving it to the user.

But even with those safeguards, I believe the team still has a responsibility to go back and verify the work.

I think one of the largest AI risks within enterprises is people simply taking AI-generated work and passing it directly to clients.

We've already seen high-profile cases where that has created PR problems.

But potentially worse than the immediate PR issue is undermining your credibility with your clients.

Kong

Let's move away from Associum for a moment and talk about your broader experience.

You've been on the investment side and now you're a founder yourself.

What do experienced operators and investors understand that first-time business owners usually don't?

Wes

When you're founding a business, you have to be an optimist. You naturally think, "It's okay. We're going to be fine."

As an investor, you tend to be a little more conservative because you've seen so many different outcomes.

Today, the information barrier around starting a company is much lower. You can find enormous amounts of information about how to start a business, market it, build a brand or develop a product.

One thing someone coming from an investment background may bring, however, is the habit of thinking about what you're building from a value perspective.

Whatever I'm doing today, how is this increasing the value of the business?

When I look one or two years ahead, what tangible or intangible value will I have created?

As an investor, when I make an investment in a company, I have to think about how it goes from valuation A to valuation B. What actually creates that increase in value?

I think that's useful for entrepreneurs too.

Even in a downside scenario where things aren't going as well as you'd hoped, you can still look at what you've created that has value.

That might be a product customers love. It might be licenses or infrastructure that enable the company to take the next step.

There are many possibilities.

But I think having that mindset of continually creating value is important.

Kong

Looking across your career, what's a business risk you've either experienced personally or watched another company underestimate until it became expensive?

Wes

I think one of the largest business risks is hiring the wrong people and then not acting quickly enough to correct the situation.

You bring someone in and perhaps they're not bad at what they do. They're simply not the right fit.

Sometimes they're just good enough that everything appears to be working.

But they're still not the right person to take the business to three times its current size.

I think that's a fundamental risk that almost every business faces.

The first way to manage it is through selection and making sure you're bringing in the right people.

Then you need to give feedback quickly and try to coach the person toward where you need them to be.

If that isn't working, you eventually need to have the more difficult conversation.

Kong

Let's zoom out.

Over the next 12 to 24 months, what risks do you think Singapore businesses are collectively underestimating?

Wes

A lot of people are focused on very large-scale risks.

The conversation might be around geopolitics or AI catastrophes.

More practically, we're entering a period where AI models are becoming extremely capable.

Those capabilities can also be used to hack systems, create fraud or deceive people.

We're in a transition period where our systems are being hardened to protect against increasingly powerful models, but we're not there yet.

I wouldn't be surprised if we see more incidents involving scams, businesses being misrepresented or malicious attacks against software infrastructure.

From an AI perspective, I think that's one of the largest practical risks we face.

As business owners, we need to make sure our systems are hardened against it.

Kong

There's also another side to AI risk, though. Businesses themselves are rapidly adopting these tools.

Wes

Exactly.

We all need to think about implementing AI.

But as we implement it, we also need to be careful about what we're allowing AI to do internally and what we're allowing employees to do with it.

Autonomous control by AI agents can be incredibly useful.

But businesses need to understand how to control and manage those agents.

These tools have made it extremely easy for people to build things themselves, which also makes it easier to make mistakes.

What information are you giving the AI access to?

Which applications can it control?

Can it send emails on your behalf?

These are the kinds of questions businesses need to think about.

Some of this is common sense and will be managed well.

Other parts won't be.

And if something goes wrong, ultimately the business bears that risk.

Kong

You raised another interesting point before the interview: if businesses don't provide employees with appropriate AI tools, employees may simply go outside the company and use their own.

That creates another category of risk.

Wes

Yes.

Obviously, employees shouldn't be putting confidential information into external tools, and businesses need to communicate that very clearly.

But even if the information isn't confidential, employees may still be conducting part of the company's business outside the organization.

Ideally, you want that work happening inside the business.

Imagine every employee is conducting their general research through their own personal ChatGPT account.

That research is being done for the company. The company is paying them to conduct that research.

That information should also contribute to the company's knowledge base rather than remaining only inside an employee's personal knowledge base outside the organization.

Even some AI platforms deployed internally still keep information primarily at the individual level.

That's less of a risk because at least it's happening internally, but it may still not provide the visibility and organizational knowledge that the business ultimately wants.

Kong

That's particularly interesting because the instinctive response to AI risk might be to restrict employees from using it.

But if you don't give them appropriate tools, they may use AI anyway, except now it's happening beyond the company's infrastructure, controls and visibility.

Wes

Exactly.

Businesses need to provide people with the right tools while also establishing the appropriate controls around how those tools are used.

Kong

Let's finish with one question.

If you could warn a first-time entrepreneur about one risk before they started their company, or pass on one lesson, what would it be?

Wes

Be action-oriented.

Don't wait.

Always be moving.

Everything will move more slowly than you want it to, so you can never really move fast enough.

Keep taking action and eventually you'll get there.