DELIVERY: A GovTech Podcast - Chris Fechner, CEO, Digital Transformation Agency

So I'd always wanted to work at the DTA, it was one of my career ambitions, and I was a little bit disappointed to start with, to be really honest. that we weren't doing some of the key service delivery functions and that we were progressively handing them off to other parts of agencies that were doing delivery? So what I actually saw when I came in, the first really big problem was a lot of the orchestration of the D T A was about still being a service delivery organization.

But we didn't quite have the right balance of skills and capabilities to transition into a policy. assurance agency. So in the first couple of years we had over fifty percent attrition. So lots of turnovers. So people going to places like Services Australia to work. And when you've got really high attrition, you really can't get a lot of internal momentum going on because you just start and then it drops.

We've also had a couple of pretty critical A and A O reviews around some of our own internal procurement, which was really damaging our reputation about our whole of government procurement capabilities. And probably the third and final aspect was we'd just had a change in government from a government that had created the DTA or a persuasion of government that had created the DTA.

to one that had not necessarily always agreed with its process. So there was a lot of things that I think were sort of working to keep the DTA operating and agencies were on a good side, agencies were relatively supportive of the f new functions of the DTA.

But we really had a big cultural, technical learning challenge to actually change from a service delivery agency to a policy agency. And we had to build all of our artifacts like so the investment oversight framework, our assurance framework, our benefits. So staff changing processes not being right for the type of people we had, the technologies we were using being largely manual.

And me coming in as someone that honestly didn't have a whole lot of federal government experience were the things that kept me awake. So I mentioned just before, if I last till October I'm five years through the term. in the first three months that I was there, I was thinking that my greatest success could be that I could last till the end of the year. So it was really challenging and

Staff were unhappy and we really needed to reset ourselves. So the first year, two years was constantly about resetting and building up the base of credibility to continue to move forward.

a lot of its work did really focus on that. And then the first iteration of the DTA was really about trying to become a scaled service delivery organization without a scaling function. And I think that was its downfall, relying too heavily on the market, not building up its own capabilities. And competing with other agencies. In this third iteration, by not doing service delivery, suddenly our advice on something could be taken seriously because

The outcome of negative advice is not, oh, we'll do it for you. It's there's got to be another way. So if we structured our work purely around what advice we could give to government, elected government, as well as the advice we could give to the public service. And we could do that without creating an own internal conflict.

then we're going to do much better. And I think really we're one of the few completely horizontal agencies. We don't have a core delivery capability outside of digital government support.

And our existence is really contingent on the quality of the advice we provide and the value that the public service thinks that they get from us. So we spend a lot of time doing retrospectives and going through our processes

and evaluating how they actually support the agencies being successful in their role because ultimately we don't want to be a critical agency. We want to be the agency that helps other agencies be successful because it's only through their success that our success is guaranteed.

technology components is actually a decision that Cabinet made to maximise reuse. So in our recent digital experience policy, one of the new standards came out was the digital access standard. And the digital access standard says stop creating new portals for government. Think about reusing the ones we do. And we work really closely with Services Australia and the MyGov portal to actually move people away from creating their own individual siloed entry points.

into that one. So we've seen really great success in people moving more collaboratively across government because they now one of the things that we do is we actually can connect the right people. So the DTA, as it says, not delivery, but if we can connect ATO and Services Australia, then we can actually make sure that they streamline their approach because they've got the same audience.

But it does also create complexities. The minister has a special place in her heart to have a higher degree of diligence on her own portfolio than maybe she's inclined to have on others'cause she wants to set that high standard. So often we're in the tough situation of having to talk to Services Australia about whether the proposals are ready for government yet. And that can be often complex because you've got briefs going up to the same minister that may say slightly different things.

But you do want to have an understanding of these things.

So I think one of the holes in our environment at the moment is things like artificial intelligence really have exposed the fact that you can't have a two year history or you don't necessarily have financial sustainability things on a startup. We are really actively looking to see how we can get those lower risk procurement thresholds reduced so that we can get this innovation back into the government purchasing. But when we get into that small to medium business

Well, a very large amount of the contracts by number actually go to small to medium businesses now. And with the recent changes as identifying Australian businesses through the Department of Finance We've now got mechanisms to actually ensure that buyers know where companies are starting from, who they are owned by, and where the money's going as well.

And we actually have affiliations with industry associations like the Tech Council and the Australian Information Industry Association where we try and discover new buyers and find opportunities for them to engage in our marketplace. As I said before though, I think we're hitting on this complexity that government needs a certain amount of pre qualification to work with any vendor, and we think that that's actually too high for some of the smaller, lower risk thresholds.

And new and increasing requirements like cyber insurance are really, I think, tough for a lot of new market entries to actually meet the government criteria. So we're looking at ways where we can sort of lower those thresholds.

bring in more innovation into the system and also support step up arrangements where you can start small and without having to go back out to market, step up a little bit higher until you're actually at the point where you can compete on an equal footing with existing panel members, for instance.

actually put it as one of their attributes that we're connecting to them to go back and do those things. I think it'll take a little bit of time for it to flow through. And it also doesn't change the world of free trade agreements. We can't actually use it as a vehicle to exclude somebody else except for the certain carve out conditions that Department of Finance have put in.

We need to be a much stronger value add economy in Australia. And I think government engaging with Australian businesses is is a really good way of doing it. I think we could change the settings somewhat to actually increase that participation. In fact, the single sellers review that we did about some of our really large whole of government contracting arrangements.

did actually recommend, and we're taking it up, a recommendation how we could lower the threshold for who can be in a single seller arrangement based on them being an Australian company and growing those capabilities locally.

'Cause it's simply unaffordable, it's unsustainable. Where GovAI is working as part of the Australian AI government plan is really about having some tools to support our people in that space. So one of the key outcomes from GovAI is Gov AI chat, which is something that allows us to use generative AI on sensitive government information without the fear that it's going to link into one of the big, large language models somewhere.

It also provides a safe platform environment to actually support people experimenting inside of government. And when we talk about government, government's got some of the most sensitive information. Information that can permanently damage government, permanently damage people. So we have a higher duty of care than most people to make sure that that data is appropriately used with artificial intelligence. And the third one is

Living in the shadow of RoboDote, we know that technology can cause permanent harms to people. And the controls that the DTA puts in the policy environment is one really strong vehicle to actually support people adopting AI, but working very hard to make sure that people truly understand the risk of applying AI in their particular context.

especially if it can create harms to people, especially harms that are irreversible. So GovAI is the let's go really hard at building something that people can experiment, try and use. to figure out how it's going to change their policy, their service delivery and their internal capabilities.

Whereas the DTA's work is to help figure out what's coming next, technical standards on things like agentic AI, policy work, risk frameworks so people can assess the degree of risk that it has inherently and that it has with control. And also one really key area that we're working in is the market.

How do we discover all of this brand new marketplace around AI? And how do we ensure that contracting through with this brand new marketplace for AI doesn't actually give the farm away when we actually do those engagements? And it goes back to that point. We won't just have

one of the hyperscalers or one of the foundation models or the frontier model providers doing the whole thing for government. That simply won't work. Because we also want to make sure that if we need to, that we can do all this stuff independent of those. So thinking about open standards, thinking about interoperability with other governments, thinking about how we can create the ecosystem behind it, are all underpinning our decisions on where we go with AI and government.

generative AI, we started seeing new things coming down the pipe. That first it was chaining of AIs and then there was the biggest and most complex models and then there was training data and intellectual property rights and all those sort of things. Now we've got Agentic coming through. What we're applying is a principle based approach, so still fundamentally meeting the twenty nineteen Australian AI ethics principles.

And then looking at the policies that are less changeable and then adapting the standards as we go. And then the last part in my mind is keeping up with the technical standards. And one thing that we're currently working on is with the OECD on a government framework for the application of agentic AI. So we're smart enough to know that we don't want to do this on our own.

We're also smart enough to know that we don't just want to get locked into a single vendor in these particular capabilities as well. So our job is to go fast enough. that we can avoid the harms or the massive amounts of duplication that could happen if we don't do it. And part of that will also be cautiously applying the brakes

before we get too deep into any one thing because if we'd have gone deep into one thing straight away, it would have been the wrong thing. And I think that's going to be the case for a while.

forty, nearly fifty years later. And we've had more of a Lego bricks approach to technology. We just keep building on top of it and on top of it. And at a certain point that legacy impact starts to slow you down in what you want to do next and we've definitely at that point and artificial intelligence has really amplified that

sort of gap between what we want to do and what we can do because of the technology legacy that we've got. We've got between forty and sixty billion dollars potentially of legacy tech debt and about sixty percent of the application portfolio is in that category of being old and needing to be remediated.

There's no chance we're ever going to do that in a traditional approach. So we need to think about what are the things we no longer need to do in systems that we used to do in systems. What's the data? that we actually need to do it because one of the principal benefits of systems used to be they create the data. Now we're actually getting that data from potentially many other sources.

some of which aren't government controlled. So how do we do the things we need to do? How do we grow policy? How do we improve service delivery without creating a skyrocketing debt? And what we do know is

hundred million dollar, billion dollar implementation projects hardly ever deliver exactly what they say they're going to. How do we think about two or three hundred AI based things that can be put together in a composable way to create long service chains where we used to think about it just vertically.

We can't address the legacy debt, either in security, in resources or in technology, in the way that we've actually created that technology debt. We have to do it differently. And I think that's going to be a really big challenge for the government. especially in a tight fiscal arrangement as we're entering into, we simply won't get the investment. So when people say, Does it go into healthcare or does it go into digital? It's a pretty easy answer from a an elected government perspective.

But what we're really trying to deal with is how do any of those other policy environments get to move forward at the speed they need to without the appropriate underpinnings of data, digital and cyber?

that don't require a lot of rationally redesigning the processes. But I don't think that it's going to be the answer. Information gets served to you quicker. You can write code quicker, but you don't actually get that really transformative change. If you think about what generative AI set out to do, which is actually mimic human intelligence.

I think we still got a long way to go before we can actually apply that to legacy. But we're certainly going to get lots of opportunities to speed things up. Then I think we actually also need to have a two speed one that thinks about what can we do completely differently. If we think one of the greatest I think

challenges was the Whippit program, which was replacing an old social services system that was thirty years old at the time, forty years old now. And one of the things that it saw was there's so much complexity in it that we spent over a billion dollars doing that work. And a billion dollars later we still have a lot of that same functionality operating in that old system. So we need to be sure that as we're investing, it's not these big single bets, it's hundreds and hundreds of smaller bets.

that we're consciously thinking about how they work together.