Automating Financial Workflows and the Role of Human Judgment with Albert Lee
In this episode of FP&A Unlocked, host Paul Barnhurst sits down with Albert Lee, FP&A leader, AI finance coach, and corporate trainer, to explore how artificial intelligence is transforming the finance profession. Albert shares practical insights from implementing AI in corporate finance, coaching CFOs, and training finance teams on responsible AI adoption.
Albert Lee is an FP&A leader, AI finance coach, and corporate trainer with over 13 years of experience helping finance teams embrace AI and automation. He has led FP&A functions across multinational organizations, including Merlin Entertainments and The Cookware Company, driving forecasting, profitability modeling, and process improvements across APAC. Today, Albert coaches CFOs and finance leaders through the AI Finance Club while delivering AI training for global organizations and professional finance communities.
Expect to Learn:
Why AI should be your finance copilot
How to build trusted AI workflows
When to use AI vs. Power Query
Best practices for secure AI adoption
Key skills for the future of FP&A
Here are a few relevant quotes from the episode:
"A strong analyst using AI becomes much stronger. A weak analyst using AI simply makes mistakes faster." - Albert Lee
"The best FP&A professionals don't just explain the numbers, they explain what they mean and what should happen next." – Albert Lee
Albert Lee shares practical insights on using AI to enhance FP&A without sacrificing governance, accuracy, or human judgment. He reminds finance professionals that while AI can dramatically improve productivity, true business value comes from combining the right technology with strong data practices, critical thinking, and trusted business partnerships.
Follow Albert:
Website - https://axiomfpa.com/
LinkedIn - https://www.linkedin.com/in/albert-lee-fcpa-869478118/
Earn Your CPE Credit For CPE credit, please go to earmarkcpe.com, listen to the episode, download the app, answer a few questions, and earn your CPE certification. To earn education credits for the FPAC Certificate, take the quiz on earmark and contact Paul Barnhurst for further details.
In Today’s Episode:
[00:00] – Trailer
[02:14] – Meet Albert Lee
[03:34] – Why Copilot?
[05:27] – AI Automation Journey
[09:47] – AI Training & Coaching
[19:51] – AI Challenges
[23:50] – AI Workflows
[28:18] – AI Governance
[31:23] – AI's Strengths & Limits
[39:40] – Career Advice & Wrap-up
Full Show Transcript:
Host: Paul Barnhurst (00:29):
Are you tired of being seen as just a spreadsheet person while others get a seat at the table? Well then welcome to FP&A Unlocked, where finance meets strategy. I'm your host, Paul Barnhurst, akaThe FP&A guy. And each week I'll bring you conversations and practical advice from thought leaders, industry experts, and practitioners who are reshaping the role of FP&A in today's business world. Together, we'll uncover the strategies and experiences to separate good FP&A from great FP&A, and we'll help you elevate your career and drive strategic impact. Today's guest is someone who's earned that seat at the table and I'm thrilled to have been on the show. Albert Lee, welcome to the show.
Guest: Albert Lee (01:14):
Thank you, Paul. Thanks for inviting me here.
Host: Paul Barnhurst (01:16):
Yeah, excited to have you. So Albert's someone I've known for a few years. I've had the pleasure of chatting with him and am really excited to bring some of his expertise and experience to the show. So a little bit about Albert. Albert Lee is an FP&A leader, AI finance coach, and corporate trainer with over 13 years of experience spanning Big 4 audit, global consumer brands, and Global 500 organizations. He has held FP&A Manager roles at Merlin Entertainment Group and The Cookware Company, leading regional forecasting, profitability modeling, and process automation across APAC. Albert currently serves as a coach at the AI Finance Club, guiding CFOs and Finance Directors on integrating AI tools into finance workflows. He is also a corporate trainer for a Global 500 asset manager and has delivered sessions in partnership with CFA Society. Albert is also an MBA Candidate from the University of Chicago Booth School of Business, a BBA alumnus from HKUST, and is a certified Fellow CPA (HKICPA).
(02:23):
Again, welcome. I love the background.
Guest: Albert Lee (02:25):
Thanks for having me here. And I just kind of met you in my home, but thanks for having me here. I love to be here because Paul, you know me for several years, happy to be here and my honour to be a guest speaker. Yeah.
Host: Paul Barnhurst (02:35):
Well, excited to have you. So we're going to start, this is a question we ask every guest. So curious to see what your answer is here. From your perspective, what does great FP&A look like? How would you define it?
Guest: Albert Lee (02:47):
Actually, this is one of the questions I answered for CFI Corporate Finance Institute. I think my answer is the same, like high impact FP&A always connects the numbers to decisions. There's a free perspective. First, you got to know, understand the business, not just the profit and loss, but also the operations, marketing dynamics, and what keeps the CEO at nice. And then secondly is that you need to generate insights that move the needle. And the final thing I think is everyone, most of the people for a party stake, you need to build a relationship and trust with your business partner. This is my opinion.
Host: Paul Barnhurst (03:21):
Yeah, you definitely have to have that relationship of trust and be able to connect things. So appreciate that answer. Now we're going to spend pretty much the rest of the time digging into mostly AI. Not that anyone's ever talked about AI lately. I've heard it's a big subject.
Guest: Albert Lee (03:34):
You got a lot of AI posts in the LinkedIn. You are too humble.
Host: Paul Barnhurst (03:37):
I've had some fun lately with my AI tips posts kind of making fun of all the AI stuff we see, but I totally get it. I use AI all the time. It's a great tool. And where I'd like to start is I know your training, you focused a lot on Copilot. Why Copilot? What led to Copilot training?
Guest: Albert Lee (03:56):
Actually, I used Anthropic One and also ChatGPT more because I do most of them like the deeper work for these two models. But I use Copilot for training because when it comes to enterprise training, I think Copilot has its own place. It sits inside the Microsoft econ system and also the security, the data governance is best of the best. And also the licencing economics just makes a lot of sense at scale. I mean, I do not have any preference on any tools, but if you come to enterprise training, you got to be Copilot. Just what's I think?
Host: Paul Barnhurst (04:28):
Yeah, there's definitely a lot of your large enterprise companies, no question, have used Copilot. So totally understand that. That makes a lot of sense. You mentioned you use Claude, ChatGPT a lot in your personal life. So ignore training, ignore all that. Do you have a kind of favourite that you find yourself that you use the most when it comes to these tools right now?
Guest: Albert Lee (04:47):
Yeah, I have. I majorly use Claude because Claude is something you know every models have hallucination. When I think that it has that hallucination, I will find ChatGPT to challenge Claude and find Claude to challenge back ChatGPT. That's what I call as cross-model web validation. Yeah.
Host: Paul Barnhurst (05:03):
I think it's a good idea. I mean, I definitely use both and I use Copilot, ChatGPT, Claude, all three of those almost daily. I'd say Claude, I use the most, but I definitely use all of them. And yeah, the cross validation is always a good idea. So how did your journey go about? How did you start spending so much time investing in AI in the work you do? What led to becoming a trainer?
Guest: Albert Lee (05:27):
First of all, I want to tell a story when I was still incorporated. I worked in corporate company, a Belgian company, and the finance function back then has been vacant for several months already. The handover, the documentation is very minimal. And then there are lots of reports that needs a lot of manual adjustment. I have nowhere to go. No one teach me and nobody's there. The finance department is empty. I mean, I got to find a way. Then I turn to AI. I find GPT. I was of desperation and very quickly it's become a mandate. It's walked me through the manual workflows that even teach me how to use a tool called C Data Python D365 Connector. That's a tool I never heard of and never used before. And then I just pull all the data directly from the ERP and auto-populate the Excel. That's an aha moment for me.
(06:20):
I just never think of any tools can do that, but AI helped me a big favour.
Host: Paul Barnhurst (06:25):
Yeah. How much time did that save you?
Guest: Albert Lee (06:28):
Guess maybe 80 to 90% because it's all automation to cover the many work.
Host: Paul Barnhurst (06:35):
I know a lot of people that are finding great use cases, like you mentioned, to figure out how to automate the pulling of data. As one person said it, look, if you're still manually pulling it, you should be using AI more.
Guest: Albert Lee (06:48):
Sure. But there's security issues for every model. So you got to figure it out anyway.
Host: Paul Barnhurst (06:54):
There are some situations where you may not be able to security. You'll find this interesting. It was Chris Riley, he shared on LinkedIn the other week that one of the contracts he signed, basically right in the contract, he had to agree for some modelling he was doing that none of their data could touch AI regardless of what security you had. That was just their policy. You could not use AI on the model. You could use AI to help you with something, but you couldn't give it any of the data. You couldn't even use Excel ad-ins because they'd get the data. It was 100% locked down. So there's definitely times for security reasons, whatever those reasons are for different companies.
Guest: Albert Lee (07:35):
The Belgian company I just mentioned, actually the head of IT stopped me from pulling all the data from the ELP and auto population in the Excel. He said it's too risky. You got to pass through the founder and the chairman. I just hold on it and I never try again.
Host: Paul Barnhurst (07:51):
Yeah. No, I get it. There's definitely... People want their data all to stay within one tenant, one system. As soon as you start passing at different places, there's always more risk. So that's still a big concern, but I like what I know you know Glen Hopper.
Guest: Albert Lee (08:09):
Yeah, I know.
Host: Paul Barnhurst (08:09):
I know. He followed him. Sure,
Guest: Albert Lee (08:11):
Sure. I followed
Host: Paul Barnhurst (08:11):
You. He writes a tonne about AI and he says a lot of people use security at the same time as an excuse. He goes, "The reality is AI is as safe as your SaaS products you're using. If you're on the cloud, they're SOC compliant, they're meeting all the different risk compliance." Now obviously that doesn't mean be stupid. There's still things you need to do. But I think sometimes because it's new, we use security as an excuse versus looking at it and weighing that risk against the other tools we have and saying, "Is it really higher?" I
Guest: Albert Lee (08:46):
Think so, yeah.
Host: Paul Barnhurst (08:47):
I thought that was an interesting kind of perspective he gave, but interesting on the Belgian one. His
Guest: Albert Lee (08:54):
Insights is very good. I follow him. Yeah.
Host: Paul Barnhurst (08:57):
I agree. He's really good. So you did that at the Belgian company and you were like, "Wow, saw the benefit." How did you continue to learn AI from there? What were the other kind of aha moments you
Guest: Albert Lee (09:10):
Had? Another aha moment is I know Nicholas is very famous, Nicholas Boche. So he's one of my friend and he has founded AI Finance Club. And I joined AI Finance Club and then I found that lots of the AI finance knowledge, I don't know how it exists and I don't know how to apply. And then I just joined at Annual Fee and then he's incredibly kind to me. He referred a lot of clients to me and Fauci told me when I have no record on it. So a lot of doors for corporate training. He actually opened the doors for me and I owe him a lot and appreciate for Nicholas' help.
Host: Paul Barnhurst (09:47):
Got it. So was Nicholas how you kind of got into doing the training?
Guest: Albert Lee (09:52):
Exactly.
Host: Paul Barnhurst (09:52):
And what is it you like about doing training on AI? What do you enjoy about it?
Guest: Albert Lee (09:58):
I'm an implementator and not just the one who want to speak. I want to measure the ROI. I want to make sure after the training, everyone got benefits, got value. They cut a lot of time and then save a lot of money works. And then every time I've finished the corporate training, I want to measure it. And that gives me a lot of feeling of satisfaction.
Host: Paul Barnhurst (10:19):
It's always nice when you know they use it and you're able to see real results. What's your typical training look like? Are you training people to build agents, to use Excel, how to do analysis, write prompts? Talk a little bit about what kind of training you're doing.
Guest: Albert Lee (10:36):
So I basically do two types of training. One is the AI finance coaching in AI Finance Club. We have accelerator programme and then I coach every individual CFO and finance directors one-on-one for their project, for their AI finance, real analysis and real case study. And another one is like I train the corporations, large and small corporations. I usually tell them how to do the AI finance in a governed way, how to keep all the audit trails, keep all the track records. You can trace back to the different figures. And at the same time, you also can save a lot of manual works and do a lot of automation. And sometimes I even teach them how to use agents, but agent is not my major case. Major case is still the LLM.
Host: Paul Barnhurst (11:26):
Yeah. So you're not doing a lot right now with agents. It's a lot more with the LLM and -
Guest: Albert Lee (11:31):
A bit like the research agent, the analyst agent in Copilot. And sometimes you can say in the Cloud... I use Cloud Cowork sometimes for my personal business. Yeah, something like that. Yeah.
Host: Paul Barnhurst (11:44):
Have you used Copilot's Cowork yet?
Guest: Albert Lee (11:46):
Copilot Coworks, I'm still applying. You got to pass through the frontier programme. I'm still on the procedure. Yeah.
Host: Paul Barnhurst (11:55):
I believe I'm part of the Frontier programme from previous and I've been meaning to look into Copilot cowork. I haven't. I've used Claude Cowork for some things and I need to use it for some more.
Guest: Albert Lee (12:06):
Claude Cowork is just amazing. Can do lots of work, can make judgement , can utilise the tools, can apply the judgement . I mean, amazing. Amazing tools.
Host: Paul Barnhurst (12:14):
Favourite use case so far for Cowork? What's the thing that's kind of blown your mind? You're like, oh wow, I don't have to do that anymore. I can let the tool do it.
Guest: Albert Lee (12:22):
For Cowork, sometimes I will use... Actually for Copilot Cowork, I'm still in a very beginning stage, but I can tell you one function. It's like Copilot in Excel, something like this, or add this in Copilot, something like this. So I can extract the information from the PDFs, thousands of pages of tens of annual reports, and then just make sure every figure is extract, go to specific cells and specific location in an Excel template. So within 10 minutes, 15 minutes, you got a whole financial analysis or even investment analysis reports ready. When back then before AI, you need to do a lot of many tasks, many stuff, like two to three hours. It just cuts a lot of time right now. Yeah.
Host: Paul Barnhurst (13:14):
And funny enough, I don't know if you saw this, there's a guy on LinkedIn he shared, I want to say I think it was $1,000. He gave $1,000 to ChatGPT to Copilot to Claude. He gave it all its preferences and said, "You need to tell me what stocks I should buy and sell." And he's up 180% on Claude.
Guest: Albert Lee (13:36):
Water.
Host: Paul Barnhurst (13:36):
He's up on the others as well. That's amazing,
Guest: Albert Lee (13:38):
Man. Yeah.
Host: Paul Barnhurst (13:39):
Like 40 and 20 or something and he's just 100%. He does the trades himself, so he doesn't have co-work where it's actually processing the trade, but he completely just follows whatever. He figured, okay, it's 3,000 bucks. If I lose it, I lose it. Let's see what happens. And I think a total he's made now, he's doubled his money overall between the three different LLMs or something like that. It may even more. He'd been doing it for quite a while and I was like, it's pretty amazing what they can do.
Guest: Albert Lee (14:05):
Got to learn from him.
Host: Paul Barnhurst (14:07):
I know. I was like, I don't think I'm ready to give $1,000 to my LLM to spend.
Guest: Albert Lee (14:14):
That's a whole lot. That's a whole lot. I cannot pay for that much.
Host: Paul Barnhurst (14:18):
Although it could be fun just to... I'm tempted to just have it pick stocks and just track it for a few months without money type of thing and see how it does. It'd be interesting. Although as soon as people start getting good at that, the opportunity goes away because the market will start adjusting for it. You have to be early at some point it will, I would imagine, although... And you never know because it's probabilistic. Some people might win, some people might lose. It'd be really interesting to do a study. And I know we're off base from FP&A, but I think. No worries.
Guest: Albert Lee (14:51):
No worries.
Host: Paul Barnhurst (14:51):
It's okay.
Guest: Albert Lee (14:52):
It's okay.
Host: Paul Barnhurst (14:53):
Take 50 people with each of those tools. Take some money and track it over two years and just see how it changes over time and what happens and how good the return is. I think it'd be an interesting study because it's going to give different answers to everybody. If they're all using the same prompts, started from the same place, how divergent and different would everybody be at the end of a year or two years? It would be really fascinating to see.
Guest: Albert Lee (15:19):
Think so. Yeah. Kind of
Host: Paul Barnhurst (15:21):
Interesting.
Guest: Albert Lee (15:22):
Yeah.
Host: Paul Barnhurst (15:23):
Yeah. And I imagine we'll see those type of studies. It's just a new world. Matter of time.
Guest: Albert Lee (15:28):
Yeah, matter of time. Yeah.
Host: Paul Barnhurst (15:29):
Yeah, exactly. All of it's a matter of time. So I know you do the finance club. You've also done some enterprise and startup training. Do you have a preference? Do you like the coaching, the one-on-one, the startups, the big enterprise companies, or does it matter? Do have one you prefer? To
Guest: Albert Lee (15:46):
Be honest, that's no preference for me because I can tell you the reason, because I think smaller companies deserve access to the same AI capabilities the big enterprises are deploying because the technology has just go to a point that's already 20 to 30 people business can run workflows that is already at a Fortune 500 budget or level. So I don't want to be the guys that's only taking their big logos. If just some small team want to get you serious on the AI practise, AI finance practise, I'm just willing to have with them and for them as well. Yeah.
Host: Paul Barnhurst (16:21):
That makes a lot of sense. At the end of the day, for the most part, they're all using the same tools. They're all using the same models. Sometimes they may have their own customised AI in some of these companies, but they're still using more than likely either ChatGPT or Anthropics model, maybe an open source like Llama or some of these others out there. But the backend models are all pretty similar. Claude's model, its answers are better in certain areas. I don't think they're better in others versus Gemini or ChatGPT. But right now I think most people feel like it's the most complete and that it's the easiest to use the way they've built skills and the way you got co-work code and Claude just all in the desktop ap. I don't know that the package right now is the best, I think. By the time we release this episode, that could change.
(17:17):
True,
Guest: Albert Lee (17:18):
True. One more
Host: Paul Barnhurst (17:20):
That we play is another model.
Guest: Albert Lee (17:23):
Yeah.
Host: Paul Barnhurst (17:24):
Yeah. I mean, I don't know if you saw that, but they're on pace. So last quarter Anthropic did, from the numbers I seen, they did four billion in revenue.
Guest: Albert Lee (17:33):
Yes.
Host: Paul Barnhurst (17:33):
This quarter they're on track to do 10.
Guest: Albert Lee (17:37):
I believe they already surpass OpenAI, right? I'm not sure, but I read some news saying that they're going to surpass that.
Host: Paul Barnhurst (17:43):
I think in this quarter they'll pass OpenAI. They're going to pass OpenAI. OpenAI is around 30 something. I
Guest: Albert Lee (17:48):
Think so. I think so.
Host: Paul Barnhurst (17:49):
I believe. I think so. So if they did 10 on a run rate basis, I don't think they've passed them on a yearly basis. Yeah, run rate. But if they continue on a run rate, I believe by next quarter they'd be bigger than OpenAI.
Guest: Albert Lee (18:03):
Yeah. Everyone's talking about the cloud right now. Everyone's talking about them.
Host: Paul Barnhurst (18:09):
Yeah. And it's amazing how many companies are switching to Cloud. And a friend of mine made a good point. Many people know David 14 as we talk AI, and I'll get your thoughts on this. But he said, look, the reality is sure Cloud's great and you can switch, but Copilot can probably meet 90% of your needs. And the bigger problem is your people don't know what they're doing. Pick a tool, train them, get benefit, give it some time. Don't just switch to the latest tool because everybody says it's the best right now. Your thoughts?
Guest: Albert Lee (18:40):
I think I'm kind of tool agnostic. I don't have any preference or any tools. Just I want to pick the tools that I like to use and I am comfortable to use. That is the best for me. I don't care its functionalities or functions or the power. I just want to use the one I am most comfortable with.
Host: Paul Barnhurst (19:01):
Yeah, no and that makes sense. And so I think you kind of agree it's a good point that companies need to be careful. Switching for switching sake usually doesn't work well because AI want the latest technology. If you did that with an ERP, it'd be a nightmare. We've all been through an ERP implementation and it's not something you change lightly. Yes, it's a lot easier to switch AI at least right now. Will that be case in two years when you've built it into all your processes? I don't think it will be as easy, but fascinating to watch. So what do you find is the biggest thing people need help with? Is it the prompting? Is it understanding the type of tasks they can do, building workflows? What do you see most people struggling with in finance with
Guest: Albert Lee (19:51):
AI? I think the people who are struggling with, first of all, they do not know which tools to use. The amount of tools is just overwhelming. It's just a lot of to use, a lot of to learn, and then they don't know how to pick which tools. The second thing is that I think the evolution is so fast. Back then, two years ago, we are talking about AIM and now we are talking about agent and AI become a very hard word. And sometimes I don't think a lot of things need to use agentic AI, but sometimes people just want to do the things using the most hyped tools or the hyped words rather than using the most suitable tools. So that's what I think in the current situation.
Host: Paul Barnhurst (20:33):
Yeah. I mean, I think a great example, I was on a call and somebody's like, "I have the standard process where I need to pull some data and do some different things. Can I use AI for that? Why would you? You have Power Query, it's deterministic. Just set up your steps and run it each month." And it was like they wanted to use AI. Well, that's a cool new thing so I could figure out how to use it. And I just said, "I wouldn't in that case. Just you already have a tool that you know works, you know you're going to get the exact output you want and you can automate it. " And I think sometimes people think, "But I got to show I'm using AI." And at the end of the day, any good leader could care less what the tool is. If you get the right answer, you're doing a good job, it's saving time, you're being productive.
(21:19):
If you're doing the things that a good employee does, I could care less if you call it AI or whatever you call it. And I think sometimes people need to remember that because we feel like, "Oh, well, I got to be on the bandwagon." Yeah, you should be using AI, but that doesn't mean it's the right tool for many situations.
Guest: Albert Lee (21:38):
And you know what? Actually, I'm a big fans of Power Query and Power BI. So if I can solve the issues by only using Power Query or Power BI, I would not use AI. Just no need.
Host: Paul Barnhurst (21:50):
I'm 100% with you. I'm a huge Power Query fan. I mean, there's definitely times when you don't need AI or use AI to help you write the code in Power Query. Because at the end of the day, as I've talked a lot about on the show and you know, generative AI is probabilistic.
Guest: Albert Lee (22:08):
Exactly.
Host: Paul Barnhurst (22:09):
You're not going to get the same answer every time. You need those deterministic pieces in there. And Power Query is deterministic. If I give it the same data and the same code, I'm going to get the same output whether I run it one time or a million times. Never going to be the case with generative AI without a deterministic step in there. And so that's something I've been preaching quite a bit, helping people realise you need to think about the amount of variability you could have in the process and that will change how you build it. I imagine you've had some of these conversations with people you're talking to.
Guest: Albert Lee (22:48):
Yep. Actually, we have a programme called AI Finance Accelerator, a base promotion. But for the Nicholas AI Finance Club, in the week four to week six, we have teaching on something about probabilistic. We are deterministic. So for myself, I will use the draught of strategy plan and the planning and also the generation of code. I may use some probability model like LRM, like Cloud and GPT, but for the calculation part, the factual part, the contents part, I need to use Python. I cannot use LRM. So it's like a killing combo for Python plus the ChatGPT or cloud. It's a killing combo. I mean, you cannot wholly rely on each one of them.
Host: Paul Barnhurst (23:36):
100% agree. So yeah, we're on the same page there. So I'm curious, when you are training people on AI, how do you ensure they have sufficient data? The other challenge is you have to have enough data to produce good outputs from AI.
Guest: Albert Lee (23:50):
Actually, we have some custom GPT called a fictitious data generator. So just describe the situation, the data you want, how many rows, how many cells. And then you give it a contest, give it the requirements it will generate for you. But of course, I mean it's not as good as the real world situation because in real world data, you got a lot of messy stuff. You got imports from an export from the GLP, lots of different sources of data. We cannot mitigate that to be honest, but we try our best to provide the fictitious data.
Host: Paul Barnhurst (24:25):
Yeah. So you're often doing the training on fictitious data, but what guidance do you give them about the real world? Because we all know real data is messy. There's a lot of companies that are struggling to get good output from AI because of data challenges. So what would you say to people on that front?
Guest: Albert Lee (24:46):
Actually for real world data, I usually go to the corporate training part because it is not really included in the AI finance club. And we have real data to be trained in AI finance, but sometimes you got to deal with confidentiality and security. So corporate training, when I sign the NDA, the non-disclosure agreements, I will teach people how to do this real data stuff. So it's go back to my framework of the model, how to automate the data using the governance and also audit trail, something like this. But a lot to talk about, so I don't think we have enough time. Sorry about that. Yeah.
Host: Paul Barnhurst (25:23):
Yeah. So how do you think about just a high level? What's your framework that you teach? Just what's the basic idea?
Guest: Albert Lee (25:31):
So for example, I can give you some examples, something like garbage out. Sometimes you need to audit before you automate. For example, when you point to a data set, you've got to describe the data where it comes from, who owns it, how often it's refreshed and what the known issues it have. And then you start to start narrow. The second step is don't try to feed AI order data lake from your ELP. Provide a clean well-understood data set like a single P&L export from the PL and then build a workflow around it. And then expand it a bit by bits by bit in the workflows. And then the last step is to build the verification into the prompt itself. I teach people to ask AI to flag inconsistencies, missing values and anomalies as a part of our outputs. So not just produce the answer, but also do the data quality check by using LRM itself.
Host: Paul Barnhurst (26:29):
DLMs are great for checking things. It's always good when you load some data. First ask it to review the data and just ask questions about it. It's a good way to start. I don't know if you've seen this. Copilot just added in Excel plan mode. Have you seen that yet?
Guest: Albert Lee (26:44):
Yes, I have seen that. But I did not use that to be disclaimer. Yeah, I did not use that yet, but I know that's a function.
Host: Paul Barnhurst (26:50):
Yeah. So I used it. So one of the cases we tested when I did my mod squad series for financial modelling is we had all the tools build a deferred revenue schedule. And then I started having them build it and provide instructions. Well, this time, instead of my big long prompt that was really detailed, I tried to really short prompt and I used plan. And it brought up a number of things that I didn't have in my typical deferred revenue schedule that were really helpful. So I was really impressed. I got a better end product than I did with my detailed prompt that I'd worked on forever because there were just things that I wasn't including. Now I could have added them to my prompt and asked for them, but I hadn't necessarily though of it and what I was doing. And so I was impressed with, I've only used it that one time, but I was impressed with the job it did.
(27:40):
And I think we'll see more and more of that. We've seen a lot of tools do that of I think all the agents will eventually have some kind of plan mode first. Because the reality is if you just build, rarely are you going to get what you want on the first try and you're much better to go through all the questions, go through the planning, really get into detail. You're going to get a lot closer than if you just let it build with one prompt.
Guest: Albert Lee (28:06):
I got to check it out. Yeah.
Host: Paul Barnhurst (28:08):
Yeah. I would love to get your thoughts once you've had. One thing you say, I know when you train AI, you say AI is a copilot, not an autopilot. Talk a little bit about that.
Guest: Albert Lee (28:18):
I think anyone use AI seriously knows it intuitively because in my opinion, AI is like your analytical copilot. It helps you navigate, watch the dashboard and spot the blind spots you miss. But the wheel is still in your hands. The accelerator, the brake are still in your hands. So you will not just let a copilot decide where the plane is going to land. That's my opinion and that's my idea. And the suggestion, the definition and also the accountability. For my opinion, it must stay with the humans.
Host: Paul Barnhurst (28:53):
Got it. Yeah, no, you're an example of the pilot. A lot of the plane is flown by the machine, by the automation, but the pilot is there the whole time and ultimately the pilot makes the call. All right. Another question I have for you. We talked about data, but what's the key with AI to having kind of good verification, a governance layer? We all know AI can hallucinate. We know there are mistakes it can make and ultimately we're responsible for it. So how do we make sure we have a good kind of verification governance layer? Any advice there?
Guest: Albert Lee (29:34):
I would give an advice for three dimensions, verification, explainability and governance. No verification is knowing what to check. You don't need to check those low risk stuff. You just check the high risk of the key financial figures, the citations because AI hallucinates sometimes and time sensitive data because the data keeps changing all the time and then casual claims and scenario assumptions. For explainability, you should be able to understand the data and restate it to the people who don't know. So use simple English to rephrase it. If not, you just shouldn't put your name on it because an unexplainable conclusion is just not useful. The first thing and the final thing I think is the cost governance. I personally have a right G model. R means red. Y means yellow. G means green. So green means that AI can run by itself. Yellow means AI draughts, human refuse.
(30:32):
While for the rest, I think humans can list while AI only advice. These are my three steps approach.
Host: Paul Barnhurst (30:37):
Makes a lot of sense, right? There's certain things where it's like, all right, I'm comfortable letting it run. If there's mistake, it's not a big deal. Or I've run it 50 times, it's been right every time I'm comfortable enough, this is low enough risk versus the other side where this is going in front of the board or this is going public. I better double and triple check all the numbers before I let it out. So it's always good to have a framework. I like that. And red, yellow, green, everybody knows that. That's a simple one to use. So thank you for sharing that. So I'd love your perspective. We see a lot of promises that AI can do everything. We see a lot of marketing hype. We see a lot of vendors probably over-promising. From your point of view, what's realistic to do with AI?
(31:21):
Where does it struggle? How do you think about that?
Guest: Albert Lee (31:23):
I think AI is good at something like information synthesis, like pull the most important real things from several hundred pages of annual reports in just minutes. Second is first draught generation and also multi-angle framework. Sometimes you've got to build from zero to 70 points stuff. It's very difficult back then when there's no AI, but right now it's very easy. So the points for human to be here is to improve from the 70 points to 95 points. This is the good thing that it can go. And also I think it's also accelerate some of the repetitive manual tasks. For the things it struggles, I guess it's some factual accuracy under pressure because if you just have time limits, after AI the wrong thing, it's just going to hallucinate. And then the second thing is that the context sensitivity, the same data points can generate different frames in different macro environments and industry cycles.
(32:22):
And also I think something accountability is also struggling with right now at its current moment. But I think it has its own pros and cons. I think the most honest framing is like AI is a powerful efficiency multiplier. It amplifies whatever the user brings because if a strong analyst using AI, it will be much, much stronger. While the weak analyst using AI, it'll make the mistakes faster. That's what I think. Yeah.
Host: Paul Barnhurst (32:49):
Makes sense. Appreciate that. All right. So we're going to move on from AI to I have some standard questions I ask everybody about FP&A. In your opinion today, what's the number one technical skill FP&A professionals should master?
Guest: Albert Lee (33:05):
I just answered back then. It's like Power QA. I'm a big fan of Power QA. I'm very contrarian. I think most of the people will say Excel or modelling, but I think it matters a lot. But in modern FP&A, Power Q is the skill that turns you from someone who waits for clean data to someone who creates and own the clean data. So you can pull, you can transform, you can refresh the data automatically. Your monthly close gets lots of faster and then your variance analysis gets much faster. And then you can use the time saved to make even better decision. This is the most important thing I think.
Host: Paul Barnhurst (33:43):
Love it. I'm a huge Power Query fan. Interesting, we've started to see lately a lot more answers around data thinking, system thinking with AI, which part of gets back to being able to clean and model data, which Power Query is something that can help a lot with that. So we're seeing more and more answers that I think centre around data design system thinking than just Excel and modelling. It's interesting to watch it all change, but Power Query got me promoted. I'm a huge fan of Power Query, so I can appreciate that answer. What about softer human skill? What would you say is number one?
Guest: Albert Lee (34:19):
I think communication specifically storytelling with data like the what, so what's now what by the framework by a lot of storytelling guru like Sof and Hamid. What, so what, now what's framework? I think most of the analysts will talk as what, like here are the numbers and then what numbers they are. But I think the good analysts will go to so what? So why it matters? While the great analysts will go to now what? What we should do and what's the call to action? What's the recommended action for the situation? So that separates and Fp&A analysts from the true business partner. That's what I think.
Host: Paul Barnhurst (34:58):
Yeah, communication is always huge. That's a very common answer we get. The framework you mentioned is one we use in my training all the time. The what, now what, so what? So what, now what? Great framework. Next one I'm curious to see and I think I know the answer based on your previous answer. If Excel removed one feature tomorrow, which one would cause you the most panic?
Guest: Albert Lee (35:21):
You know what the answer is, man.
Host: Paul Barnhurst (35:24):
Let
Guest: Albert Lee (35:24):
Me guess. PowerQuery.
Host: Paul Barnhurst (35:26):
Power Query.
Guest: Albert Lee (35:28):
Yeah, PowerQuery. Yeah.
Host: Paul Barnhurst (35:29):
I knew I looked at that one and I said, based on the answer a minute ago, I'm like 99% sure I know exactly what you're
Guest: Albert Lee (35:37):
Going to say. Exactly, exactly. How do I? No bringer for me yet.
Host: Paul Barnhurst (35:41):
Okay. So if you had to pick a second one, which would be number two?
Guest: Albert Lee (35:43):
Number two, maybe the V Stack or X lot up. Just my preference. No reason. No reason. Just my preference.
Host: Paul Barnhurst (35:51):
Yeah, no, that's good. We'll go with that one for number 10. I was just curious to see what you'd say after Power Query. All right. So we have a get to know you section. This is where I ask kind of some fun questions of each of our guests to get to know you a little bit better. So if I said you could have dinner with one person alive in the world today, who would you take to dinner and why?
Guest: Albert Lee (36:13):
Let me ask you, what do you think? Who do you think I will have a dinner with? I think you know the answer as well. Yeah. Just give you some tips. You know that answer.
Host: Paul Barnhurst (36:20):
High on my list would be Warren Buffet. I think he'd be fascinating to go to dinner with. If Nelson Mandela was still alive, I'd love to have dinner with him. A lot of people I could list. Davis Smith actually right now, you have no idea who he is. Most people don't. He started the company Codopaxy. It's a corporation B, Benefit Corp. And his goal with the company is to alleviate what they call not just poverty, but I think they call it severe poverty, which is you make less than $1 a day. And so they use areas that are very poor to source all their materials and just an amazing guy. I've always really respected him. So he'd be my dinner.
Guest: Albert Lee (37:07):
I am a big fan of Warren Barface, but I may choose another person. I watched his basketball game since I was very young, since I was a teen. I think I will choose Tim Bencan, a basketball legend in NBA league.
Host: Paul Barnhurst (37:19):
Definite legend. I'm a big Utah jazz fan, so I can remember some good matchups against him and Carl Malone.
Guest: Albert Lee (37:25):
I watched him play when I was a kid. What I admire is really his championship is his emotional stability. Every time he fell, he lost to Lakers. He got back up and won the champion again. I mean, he across three decades of basketball era and three different playing styles and three different rosters. He still win the champion. I mean, the confidence was him, the composure, the leadership, the ability to make whatever and whoever was around him better. So I will just have a dinner with him. I want to.
Host: Paul Barnhurst (37:55):
Yeah, he had three different teams. You had David Robinson with his first title. You had Parker and Genobili with the next three. The last one, the leader of the team was really Kawhi Leonard in that last time. Kawai Leonard.
Guest: Albert Lee (38:06):
Yeah, exactly. Exactly. Yeah.
Host: Paul Barnhurst (38:08):
And then the others were ageing. So there's really kind of been three different... I agree with you. That's an interesting... I hadn't thought about that. All right, so next one here. If you could have any superpower, what superpower would you have?
Guest: Albert Lee (38:20):
That's a pretty challenging one. I'll choose time travelling because I can tell you the reason because I want to go back to my younger self. When I was young, when I was still in corporate FP&A in maybe different companies, different FP&A department, really don't know how to handle people and relationships. I rubbed colleagues in the wrong way and said things that I shouldn't say. And also I didn't appreciate everyone work with me enough because I know they have that bad fight. Sometimes I just say something that is not appropriate. So if I could go back, I want to teach and coach my younger self how to treat people differently with more patience, with more brace, because the technical skills you can pick up anytime. But the people skills, it's gone for a lifetime learning. Yeah.
Host: Paul Barnhurst (39:07):
Last one. Do you have a favourite movie or TV show you like to watch?
Guest: Albert Lee (39:12):
For movie, there's one. For TV show, that's one. So for movie, I will choose the pianist. I forgot who the main character is, but it's the pianist. So touching for me to see the pianist. For the TV show, I think it's Game of Thrones. Kind of hard to share the storytelling ambition, but even though people say the ending is not too good, but I'm still a big fan of Game of Thrones. Amazing.
Host: Paul Barnhurst (39:34):
I will admit I've never seen an episode, so I'll take your word for it.
Guest: Albert Lee (39:38):
Yeah, indeed.
Host: Paul Barnhurst (39:40):
If you could offer one piece of advice to our listeners to be a better FP&A business partner, what advice would you give?
Guest: Albert Lee (39:49):
Got to be one, right? I think it's the understand the operations from the operation partner or the sales partner, the business partner you have in the sales team or operation team. Because sometimes figures got to be very cool, very cold. You need to understand how to build a relationship. You've got to know the real world situation from the business partners, from the operation partners. You've got to know the news of the whole story. You cannot just know about numbers. The whole story is in the operations, it's in the sales business. So you've got to know from the others, not from numbers, I think. Yeah.
Host: Paul Barnhurst (40:23):
I agree with you. I think that's really good advice. So as we wrap up here, last question. If someone wants to get in contact with you, maybe questions about training or just reach out to you, what's the best way for them to do that?
Guest: Albert Lee (40:35):
You can search in the LinkedIn, search my name, Albert Lee, F-C-P-A. I'm the Albert at ASEAM, and then send me a connection request. Just mention you heard my podcast and I will always reply and connect with you.
Host: Paul Barnhurst (40:47):
All right, perfect. Well, thank you so much for joining me, Albert. As always, fun to chat. Appreciate you sharing some of your experience and I'm sure it is late for you, so we'll let you go so you can enjoy your evening. Thank you. Thank you. Thank you so much for coming on, Albert.
Guest: Albert Lee (41:03):
Appreciate, Paul. See you.
Host: Paul Barnhurst (41:04):
That's it for today's episode of FP&A Unlocked. That's it for today's episode of FP&A Unlocked. If you enjoy FP&A Unlocked, please take a moment to leave a five-star rating and review. It's the best way to support the FP&A guy and help more FP&A professionals discover the show. Remember, you can earn CPE credit for this episode by visitingearmarkcpe.com, downloading the app and completing the quiz. If you need continuing education credits for the FPAC certification, complete the quiz and reach out to me directly. Thanks for listening. I'm Paul Barnhurst, the FP&A guy, and I'll see you next time.