TAINA Talks: AI / GenAI in Tax Operations

The role of  AI / GenAI in Operational Tax

Artifcial Intelligence (AI) and Generative AI (Gen AI) offer financial institutions a wide array of opportunities to streamline and enhance their Operational Tax (or Tax Operations) processes, especially in areas like due diligence, tax reporting, compliance, and risk management. 

For our next edition of TAINA Talks, we called on industry thought leaders and tax professionals to share their insights on the role of AI / Gen AI in Tax Operations in a series of 4 polls exploring the follwowing questions;

• Is your organization prepared to integrate AI/Gen AI into its tax operational processes?

• What aspect of tax operations in your institution would benefit most from AI/Gen AI automation?

• What is the biggest challenge your organization faces when adopting AI in tax operations?

• Do you believe AI/GenAI can help financial institutions achieve better compliance with international tax regulations like FATCA and CRS?

Watch the below podcast to see the poll results and an engaging discussion around AI and GenAI in Operational Tax with the TAINA team including Rich Kent - CTO,  Shaun Boughey - Infrastructure Lead , James Sweetman - Head of Product and Sarah French - Chief Revenue Officer.

Transcript

September 26, 2024

Sarah Cooper 
Hello, I'm Sarah French and Sarah here at Taina and joining me today is Rich Kent, our Chief Technology Officer, James Sweetman, Head of Product, and Shaun Boughey, our Infrastructure Lead.
We'll be discussing a very current topic around AI Generative AI, more specifically in the regulatory compliance and operational tax space and the opportunities it may bring if it's really leveraged. As you know, we issued 4 poles and today we'll be discussing openly the results of those.
Anything to add before we start looking into these results team.

Shaun Boughey   
No let’s jump in it.

Rich Kent   
Yeah absolutely.

Sarah Cooper  
Is your organization prepared to integrate AI Gen. AI into its tax operational process?

• Using AI tools extensively - 8%
• Considering AI for future - 8%
• Exploring AI solutions - 67%
• No AI plans for tax operations - 17%

I guess first and foremost were any of you surprised by the results that we that we have here?

Shaun Boughey   
I don't think I'm shocked it's been such a big thing in the last kind of year, right?
But I think everyone went for a phase of kind of seeing if it was a bit of a hype trend and now it's kind of look like it's got a bit more longevity and was a bit more kind of to it. You're starting to see businesses look at it and kind of looking at that adoption process.

So it kind of makes sense, people are at the exploration stage and trying to work out what it could do. What are the business benefits where it could assist? And I I think the the poll perfectly shows that right is a majority of people are in that kind of exploratory phase.
And rightfully so. Right is you've got to find that business need to kind of meet the usage of it instead of going the other way round. So makes sense from my opinion.

Rich Kent   
I'd actually expects the number that have no plans at all to be a little bit higher because it the specific question was around tax operations. Take the large FIs, they've all got innovation departments and sections, but I think maybe our question specifically in tax OPS was interesting.
Like I have no doubt, most organizations are looking at AI, but it's clear it's not filtering through to the to the field of tax OPS. I really would have expected that to be a little bit higher, I think.

Sarah Cooper 
Do you think maybe that's because I think the general assumption is that tax OPS always has to be manually paper driven or and no one can see kind of a way or you know kind of a way of moving on from that.

Because they are so stuck in the world of paper in in with tax documentation, especially in spaces like tax reclamation. It's very paper heavy .I do wonder if that's why. It's a change, isn't it? An evolving change.

Rich Kent   
Possibly. I've noticed historically that these certainly larger organisations rely on.
Sort of innovation coming in from outside organizations.
So that whilst they might not have AI experience in house, you know using a supplier like TAINA, that's how you get innovation, like by buying it in essentially.

James Sweetman   
Yeah. And I think this, I think to your point, rich, especially if you're looking wider, is that you know the prevalence of AI tools which could be used for you personally.
I know it's within tax OPS, but like you can start exploring personally is quite amazing.

In so far as that, you know, I personally have a ChatGPT like account which I use and I kind of play with and like you know I built like a tax bot on it just to see what it could do. If you see what I mean.
So I think that because you can easily go and do that, you can easily create an account. It's, you know, it does sort of lend anyone who's sort of curious into sort of like giving it a go almost because it is far more accessible than, say, like another big, you know, technological thing in the last few years.

Obviously when you know, everyone started talking crypto, everyone started talking distributed ledgers but the thing is, you know, how does the everyday person start going about doing a distributed Ledger? It just didn't really have the same traction. Whereas now it's just so available, that like even if you're mildly curious, you can start exploring.

Sarah Cooper   
Right. I think we'll go into the second poll.
What areas of Tax Ops in your institution would benefit most from AI in AI automation?

• Due Diligence - 30%
• Tax reporting & filing - 30%
• Withholding tax optimization - 20%
• Compliance & risk management - 20%

I'm going to say for me with having 15 years of tax OPS experience covering all of these, I would jump at the chance of having it across, definitely across the board there, but I think if I was to pick one, first and foremost my go to straight away would be on the due diligence part of the process, yeah.

Rich Kent   
I mean, in what way?
What would you?

Sarah Cooper 
Obviously you know OCR is, is great.
You know, kind of the best one I've seen is obviously here and that's not because of working actually really is the best one I've seen. But I wonder what else out there, if people think that they can be more than OCR in the due diligence space.
Yeah, that would that would be my one, first and foremost, I guess just from an operational perspective and then obviously I would jump straight to reporting.

Rich Kent 
OK, OK. And what area like data quality, that sort of thing or?

Sarah Cooper 
Yeah, the data quality piece massively, especially the linking from the data side to the tax form side.

Obviously I think the calculation pieces in the middle. I think they would naturally happen, but the the two end pieces for that for me would be the biggest bits that I'd love to see AI use as much as possible. And if it could be done, that's the thing. Can it be done. How much of that process can be utilized with AI or Gen. AI to make it better?

Rich Kent  
I thought those two might be a little bit higher in the percentage, but I thought like when you think about like perceptions would be that the actual compliance rules maybe wouldn't necessarily lend themselves to AI.
So they might be a little bit lower, but you know reporting and the due diligence data quality and you know there are so many, OCR least mentioned there's so many ways that AI could be used in those areas that I thought the split would be a little bit leaning towards those two.

James Sweetman 
I think the leaning of onto due diligence and tax reporting makes sense. I agree with Rich. I think the leaning would be more and I think the leaning on to those areas may also make sense from a point of view.  

If you think about, you know, really AI and ML has been around for so long. The real, you know, advancements which were happening is the fact that AI is getting so much better at dealing with unstructured data.

So those are the two like pillars. I would think there's the least structured data. You know the other two are very much more like codified. You know you get this type of withholding, you get this type of payment and that kind of stuff, so I feel that that makes sense.
I think what's kind of interesting, I think the leaning is correct. I think leaning I would have expected more, but I think the I think what's quite telling is the how the even it is across the board is. To Rich's point. People are still people are still grasping at how to productize this. There isn't a clear sense in my head like, well, we've got lots of ideas at TAINA and what we're going to do with us, our stuff. But I think in terms of institutions, it's a bit like we know this is a thing. We're just not entirely sure, and that spread is quite, quite reflective of that.

Sarah Cooper   
Well, I agree. I also think that you know Rich you mentioned the point that people have AI designate budgets now and but I just think in the tax operation space specifically it's one, it's really hard to obtain a budget for something that is still so I guess new to that, that, that part.

You know, if you're talking in in the IT team or the tech team straight away, they'll be all over it. But from a tax perspective, given some of this, processes are still quite manual. Jumping to AI is huge.

And it's the fear of the unknown. And then also because tax operations deal with a lot of reg compliance, maybe there's a fear of are we being compliant if we use you know is there any risks involved? Is there any bias that's involved if we're using AI?
You know, I guess just operationally, I think it's maybe it needs to catch up, maybe it doesn't, but I think there's a fear of the unknown around AI in that space.

James Sweetman 
Sensitive to risk areas adopting innovation.
Yeah, it can feel a bit wild. Sorry, Sean, I think I've interrupted you probably on that.

Shaun Boughey   
No, no, it's a great point. I was just going to swing back to a point. Sarah and James brought up, and Rich. All of you actually really were leaning towards kind of a more kind of paper base beyond structure piece instead of towards the compliance side.
But I  would see there's really strong benefits towards the compliance side.
So where James mentioned it structured, it's kind of more rigorous and laid out.

It's something AI thrives on. It has its rules set. It knows what to do. And with that rule set, you can start looking.
Detection. You can start cross checking things. You can use that to make sure your compliance is even stronger than before, right? It just comes down to again going to go off James words. That's how you productise that.

But it definitely has benefits and I think that's maybe where people are starting to see it. That's why it's more of an even split is each area with a lot more you can dig into as people start exploring.

Rich Kent   
Ai is absolutely brilliant with the unstructured data, but it's also incredibly good at automating tasks. So if you do have structured data, it as you say, Sean, it lends itself to that area as well, so.

James Sweetman   
Yeah, because I think and you're right though Rich cause, it's like it does lend itself very much to those automated tasks is you know when if you start playing with AI.
Very quickly, kick into that moment of like, OK, What data am I training on it? You know how consistent am I being? Am I being narrow enough with its scope to get the same results each time?
And that's the like. You know, I challenge anyone who's listening to this to like you try ChatGPT and just like, try putting it in a couple of times and see the divergent of like response you get. If you're going to productise it, it has to be consistent.

So having you know you're right, I think it feels like a silver bullet for due diligence and tax reporting, but realistically the privatization is going to come from probably the other two, which will be very interesting codified rules, narrowing the scope of agents.

Maybe having AI agents working in hierarchies. which I know sounds A bit daunting and brave new world, but the idea that you have well as you have it, you have an AI Agent checking on other agents. It's literally like mimicking an organizational structure. So you have supervisors essentially.

Shaun Boughey   
Think of it how we all work as humans, right? We get people to go, actually. You're getting the AI to essentially sense check itself. What is one of the big things?
And I'm sure will come onto it later. How do you deal with hallucinations?

 
How do you deal with a piece of its fact checking, but it may not know the answer if it starts to kind of cross check itself as well before it replies to you, you start to limit some of that as well.

Sarah Cooper  
Only level quality assurance isn't.

James Sweetman 
Yeah, I know. It's very cool.

Sarah Cooper 
That was a good topic. OK, right.

What is the biggest challenge your organization faces when adopting AI in tax operations.

• Data quality & integration - 0%
• Budget constraints - 25%
• WLack of AI expertise - 13%
• Security and privacy concerns - 63%

Do you know the biggest thing that stood out for me on this and probably for all of us here was that data quality and integration with 0%?

Sarah Cooper 
I found that one absolutely a little bit mind blowing. I don't know. Maybe I shouldn’t.

Rich Kent 
Why is that?
Why is it just sections? We need to work on?

James Sweetman

I'm gonna make a shot in the dark, and I'm gonna say there's a correlation between exploring in our earlier poll result and a low result there. I think when you explore, it's such a cool thing to demo.Hey, I throw in this thing and I get this result and it's really quick and it look it's written like an entire podcast for me or whatever it is and it can narrate. The podcast is great.
Try and make it write the same podcast again based on your data, and if you have data variance. If you start like making and building it into your system.
 

The variance that Shaun's point about hallucination, the variance of data, can lead to large hallucinations and things you don't expect. So I think it's only when you really get into like trying to get replicable results in a productized way of AI that suddenly you may start encountering that more and more. So maybe it's a sort of, you know, it's a shot in the dark, but maybe it's just a sort of this is an early stage thing. So maybe people haven't tried their AI tapping into their data at the moment.
Maybe they're still to the other. What was the other result was around risk and maybe it's still like hey, let's explore with it in the playground over here.
We don't want to bring it in, let it touch like, you know, maybe.

Rich Kent   
Maybe that shows like a general.
I'm struggling with the word, but people don't really understand how to do it correctly. How to use AI correctly. Anybody as you say, they can download ChatGPT and you can get a demo up and running in minutes. But to productionise that to, you know, have a really reliable product that takes skill and time, right? And maybe that's what we're seeing in that result there maybe.

Shaun Boughey   
Yeah, I've got. I've got a few strong opinions on this topic. Given put in my security, how and I was straight off. I am not surprised. I'm shocked It's not higher for the worry around security and privacy, right?

You're essentially asking to send off your company's data to a service to then run AI over it. And there's horror stories about being used for training and being collected and all sorts, right? Well, again, there's ways the industry are adapting. Not everything's closed source. You can run it on your own hardware. You've got other ways to do it around commercial agreements,

So it's definitely moving in the right direction, kind of as an industry. And I think this comes into the piece works is still relatively new at large scale, but I think even the companies providing it are still learning where we need to land and what customers want.

Rich Kent   
Yeah. Can I ask you a question?
Just so we, you and I, we spend a lot of time talking about security and AI. But do you think people really appreciates that it is an issue?

I know we saw 63%, so I'm not being silly but to your point, I'm surprised it's not higher. Do you think people just go with the flow and they don't really think through the implications of doing it incorrectly?

Shaun Boughey  
I don't think that's truly the case. So obviously we get involved with a lot of different companies and we see a lot of kind of risk based policies and kind of pieces we have to work around. And one of those we see quite common at the moment is talks around the usage of AI what?
Means companies and businesses are thinking about this holistically at a high level and kind of what risk it means. I think the piece. What we still got to develop on as a whole? Is what is at risk, but not just what the risk is, is how you manage that risk.

Cause like everything you can mitigate risk, you can manage it. There's an acceptance level somewhere. Each business will be slightly different, but it's finding ways to manage that. What may just be a case of you have an AI model instead of it being ChatGPT somewhere in the cloud.
It's a model you run locally within your hardware within your data centres, and it still provides the same functionality, but you know it's at least ringfenced to your area.
You're not having to worry about anything else, so that's straight away. A big privacy win, right?

James Sweetman   
Which also brings benefits of having it trained on your data for your for your purposes. But yeah, I think you're right, Shaun. It's interesting.

Sarah Cooper 
Just a quick one on the lack of AI expertise. Was that 13%, do we think that that's across an institution or maybe more specifically just in the tax operations space, the lack of AI expertise and you know obviously in Tax Ops, you still do get your tech?
Support, but I guess it is quite maybe niche. You know sort of specific.

I mean, I don't know. I don't know if I'm putting everyone into a box here and I shouldn't be.

James Sweetman   
No way I think. I think it's an interesting one because if you were to say if you were to ask people like, what's the what you deem as AI expertise like I think sometimes like there is a huge growth of prompt engineering.
I say if you said prompt engineering, you'd be like, what does? What does she mean?
It's this the whole idea that you're either via code or via, you know, just actually just.

Closely written English like you're essentially giving tasks and sectioning out like an architecture or prompts and tasks to AIs as though you were giving it to, you know, the sort of classic comparison is like, you know very just fresh out of school, like intern who's, like incredibly great at accessing the web and getting everything in one place, but has it doesn't know where or how to start.
So you know, prompt engineering is a huge part. I know. Sean, will, you know, probably and Rich will school me on this but it's it's also like the whole idea of, you know, the data and the privacy concerns, you know, building an AI model into your own into your own environments.

And, you know, at a very high level like especially the idea of  extracting your data and putting it in a format that can be essentially read by AI that could be using a OCR technology like TAINA has. Or it could be you know which large language model do you use to cluster and group your ] data and make those associations.

And then you can use different LLMs for either that human interaction or how they analyze and access those clusters of information. If you see what I mean. So it's not just like, it's not just this one shot thing where you just like pull in an LLM and it's like, wonderful and it does everything like ChatGPT has done a great deal of, like, hiding a lot of that complexities and all of the nuts and the bolts, which and the expertise to build that up.

I think sometimes people you know actually wow, that's a ton of expertise like and very specific. I mean, I think Linkedin's got some very high pay roles, if anyone's interested. Like for very specific knowledge about different types of LLMs and how they work.

Shaun Boughey 
I think there's also a piece here where the lack of I'm not. I'm not surprised it's so low in a way, because AI is a term for AI experience. That's really broad, right?

So here at TAINA we all know OCR amazingly well.It's why it works so well. That's why we work with a lot of universities on it, right? That is a part of kind of AI.
It's a machine learning part, but if you were to say, do we have expertise in generative AI? Or deep learning for doing automotive kind of mapping.
No, that's not our business. So it could be a case, but we've left it slightly broader to kinda go on the other side, right, so.

Sarah Cooper 
Right. OK. Last question,

Do you believe AI Gen. AI can help financial institutions achieve better compliance with international tax regulations like factor in CRS?

• Yes, AI can boost compliance - 63%
• No, AI won't have much impact - 0%
• Somewhat, depends on the rules - 38%
• Not sure, need more info - 0%

So we had two with no votes, so no, AI won't have much impact and two not sure need more information.
I don't know if I'm if I if I'm surprised by those two. I think for me the 38% of somewhat depends on the rules. I mean that we all know that FATCA and CRS is it constantly changes.And I think that maybe the general consensus out there with this one specifically is, is there too much frequent change in order for AI to keep up. My answer would be of course it would. But you know should be able to keep up and probably would keep up better than maybe a human can.

If I if I can say that but I wasn't kind of massively surprised that those two were zero and the top votes are where they are. I do think AI can boost compliance massively.
I think it's more to.
I think it was one of our earlier polls and I think, James, you mentioned on it more and Shaun, maybe it's more the compliance piece of it where it actually should be used and utilized more than where people are looking at. Can I help a pain point?
Which is the due diligence and reporting?

Rich Kent 
I thought there would be. I thought the low answers wouldn't be zero because one of the previous poll points said there was like one in six, had no plans to look at AI.So kind of you're thinking that, well, if one in six don't have no plans to look, then, surely there would have been more than 0.

James Sweetman  
I like. I like. I like the person who's saying, you know, one in six. They're they're like, no, it will affect us. I'm not gonna look at it. I like that kind of cavalier.

Shaun Boughey 
So playing on the other side though, that could be going back to the other pool, right?

Maybe this budget constraints?
Maybe they know it will impact it.
They just don't have the budget to be able to actually look at it right.

Rich Kent
Or security concerns. Yeah, yeah.

Shaun Boughey
Because yeah, all security. There could be a multitude of factors.

It could benefit. Maybe we could see it. It doesn't mean it's off the table. We're just not looking.

Rich Kent
Mm hmm, yeah fairpoint.

James Sweetman 
But I think to your point, Sarah, there is, there's something which is about. With these rule sets. So if you're looking at the IRS and you look at their various publications and they're pretty good because they publish everything, but you know, you may get a different interpretation of that based on like, you know, different amendments and you may get different interpretations when you go to different conferences.

You then also get, you know, different interpretations. When people come up against audit. And a lot of those kind of decisions and you know presentations and everything like that, you know they are not always necessarily documented.
They're not always necessarily published, so they're not public domain.
So then you're kind of left in the question of, like, OK, where do you get this from?
 

And then you also have a situation where you know the writing of some of the tax rules and regs can seem. Can be contradictory. You know, we've seen it whenever we've been speaking with Big 4. Sometimes we have to get a whole bunch of guys in one room specialists, and they need to hash it out because they'll say, well, what about this Reg? And it contradicts this one. Then it contradicts this one.
 

Now, if you've got like tax experts and some of the guys we've worked with have even like, you know, the authors of, like, various parts of, like the IRS rule set, like, imagine the capability of hallucinations. Like, imagine what that could occur there. So there is. So there is a sort of a funny thing here which I think is interesting when they say depends on the rules for that 38% where I think it's it's not just depends on the rules. It depends on like how the rules are conveyed. If you know what I mean.

Rich Kent  
I'll tell you something quite interesting. All right? We so we've asked like entities, financial institutions.It would be fascinating to see what tax authorities think if we ask the same questions to tax authorities, I wonder what they'd say.

Sarah Cooper
I think would be really open with how much you know, even auditors, right?
They'd think it'd be a blessing if they ever choose more in the compliance space.
I really do.

And Tax Authorities, they don't just take, they're not just looking at one piece of data they're looking at it's a collective right?
So maybe they can then utilize AI to across the board, but that's another. Another talking point really.

But thank you everybody for joining and I hope everybody enjoyed listening and who watches it.