Today, 23 August 2026, marks three years since VAi, our insolvency co-pilot, went live.

Three years.

In the world of AI, that feels like rather a long time.

When we launched VAi on 23 August 2023, ChatGPT itself had been publicly available for less than a year. Most accountants, company directors and insolvency professionals were still trying to work out what generative AI actually was, never mind how it might fit into their working lives.

We decided to find out.

Not by talking about it. By building something and putting it to work.

Why did we build VAi?

The original idea was actually quite simple.

I’ve worked in insolvency for more than 40 years. During that time I’ve seen thousands of directors arrive at roughly the same point.

Their company is struggling.

Cash is tight. HMRC is pressing. Suppliers are chasing. Perhaps the bank is becoming nervous. Staff need paying on Friday. The accountant is worried.

And the director has a head full of questions.

Can I carry on trading?

Am I personally liable?

Can I pay this creditor but not that one?

What happens to my employees?

Can I sell the business?

Do I have to liquidate?

What should I do tomorrow morning?

Often the biggest problem isn’t that professional advice isn’t available.

It’s that the questions don’t conveniently arise between 9am and 5pm.

They arise at 11 o’clock at night when somebody can’t sleep.

Or on a Sunday morning.

Or five minutes after they’ve opened a letter from HMRC that they don’t properly understand.

That was the gap we wanted VAi to help fill.

Not to replace an insolvency practitioner.

To help somebody understand the position well enough to know what to do next.

That’s an important distinction.

Three years later, I believe in that idea more strongly than ever

AI has moved extraordinarily quickly since August 2023.

The models are vastly more capable. They can analyse documents, interrogate numbers, research issues, draft correspondence, test arguments and hold conversations that would have seemed remarkable only a few years ago.

But something else has become increasingly obvious.

Having access to AI isn’t the same thing as knowing how to use it well.

That’s where a lot of the current debate goes wrong.

People talk about AI replacing accountants, lawyers, insolvency practitioners and all sorts of other professionals.

I don’t think that’s the most interesting question.

A much better question is:

What happens when somebody with decades of professional experience learns how to combine that experience with AI?

That’s where things become genuinely powerful.

AI can process information very quickly.

But professional judgement still matters.

Context matters.

Experience matters.

Knowing which question to ask next matters.

And in insolvency, understanding the human being sitting on the other side of the table matters enormously.

Insolvency isn’t really about balance sheets

That might sound an odd thing for an insolvency practitioner to say.

Of course the numbers matter.

But after four decades doing this work, I’ve learned that the numbers are usually only part of the story.

Behind almost every insolvent company there is a person.

Someone who started a business with an idea.

Someone who employed people.

Someone who took risks.

Someone who perhaps hasn’t paid themselves properly for months because they were trying to keep everyone else paid.

And sometimes somebody who has simply left things too late.

That’s human nature.

When the news is bad, people sometimes avoid opening the envelope.

Technology doesn’t change that.

But it can make getting that first bit of help much easier.

You don’t have to make an appointment.

You don’t have to explain yourself to a receptionist.

You don’t even have to admit, initially, that you’ve got a problem.

You can simply start asking questions.

And sometimes that first conversation is enough to get somebody moving.

We’ve also learned what AI shouldn’t do

This matters just as much.

VAi was never intended to tell directors:

“Don’t worry, everything will be fine.”

Sometimes everything won’t be fine.

A company may have reached the end of the road.

Equally, it shouldn’t automatically tell somebody that liquidation is inevitable simply because cash flow is difficult today.

There may be funding options.

A business sale.

A restructuring.

A Company Voluntary Arrangement.

An Administration.

A solvent sale followed later by an MVL.

Or simply some difficult but manageable decisions that need making quickly.

Good insolvency advice has never been about pushing every company down the same road.

It’s about understanding the circumstances first.

That’s why we’ve continued developing the way VAi asks questions and gathers context before trying to provide answers.

In other words…

Don’t diagnose before you’ve listened to the patient.

It’s a principle that applies just as well to AI as it does to people.

The biggest lesson has probably been this…

AI works best as a co-pilot, not an autopilot.

That’s why we chose the description in the first place.

A good co-pilot helps you think.

It checks things.

It spots what you may have missed.

It challenges an assumption.

It helps prepare the paperwork.

It can take some of the repetitive workload away.

But somebody still needs to know where the aircraft is going.

That distinction has become clearer to me with every passing year.

I’ve now spent a considerable amount of time working with AI, testing it, arguing with it, correcting it and finding new ways of using it.

Sometimes it produces something genuinely impressive.

Sometimes it gets something wrong.

Occasionally it gets something spectacularly wrong.

And that’s exactly why professional judgement remains so important.

The skill isn’t simply getting an answer from AI.

The skill is knowing whether the answer makes sense.

There has been another unexpected benefit

Building VAi has changed the way I think about my own profession.

When you’ve done something for 40 years, a great deal of what you know becomes instinctive.

You stop consciously thinking about why you ask a particular question.

You just ask it.

But if you’re trying to teach an AI system how an experienced insolvency practitioner approaches a problem, you have to unpack that instinct.

Why does that fact matter?

What question should follow?

What are the warning signs?

What information is missing?

When should somebody stop trading?

When should professional advice be sought immediately?

Trying to put that knowledge into a system forces you to examine your own thinking.

I’ve found that fascinating.

So what happens next?

VAi will continue to develop.

The technology certainly will.

But the objective remains much the same as it was on 23 August 2023.

To make high-quality insolvency knowledge easier to access.

To help directors understand their options earlier.

To help accountants and other advisers work through difficult situations with their clients.

To help insolvency professionals use AI sensibly rather than simply because everybody else seems to be talking about it.

And, perhaps most importantly, to help people ask the difficult questions before the difficult situation becomes a crisis.

Because three years of working with AI hasn’t changed one of the oldest lessons I’ve learned in insolvency.

The earlier you deal with a financial problem, the more options you usually have.

Ignore it for another three months and those options have an unfortunate habit of disappearing.

Three years… and we’re still learning

So today isn’t really about claiming that we’ve cracked AI.

Nobody has.

It isn’t about claiming that VAi can replace decades of professional experience either.

It can’t.

It’s about recognising three years of experimenting with what happens when the two are brought together.

Experience on one side.

Technology on the other.

And somewhere in the middle, hopefully, something genuinely useful.

When VAi went live three years ago, none of us really knew where this technology would lead.

I still don’t.

That’s part of what makes it interesting.

But I do know this…

AI is going to become part of professional life. The question isn’t whether we use it. It’s whether we learn to use it well.

Three years into that journey, I’m very glad we started when we did.

Happy third birthday, VAi.

There’s plenty more work to do.