By Jon Reed

April 17, 2024

(Grant Halloran talks AI for finance)

Most vendors have an upbeat LinkedIn presence, but Planful has been especially perky, peppering my newsfeed with feel-good updates (e.g. Planful Maintains Global Expansion Momentum in 2023).

But despite the intriguing intersections between generative AI and finance, CFOs are dealing with a boatload of economic stress factors. So how is Planful helping finance leaders in the fray?

Time for a catchup with Planful CEO Grant Halloran, who told me that the economy is proving more resilient than many expected:

We just, had a record Q4 in the company's history, and we just had a massive Q1  - year over year bookings growth were up 40%...  I'm curious to see what you think about the economy, but it feels to us that it's stronger than people expected, given this inflation news and everything else. Companies are buying; companies are investing; the labor force is strong. Obviously, we'd all prefer the rates to be lower. But it appears that we may be in for this kind of level for a while.

Investing in enterprise AI - we're in a turbulent phase

Is AI part of this economic resilience? I would say yes, but only to a modest degree: investment in anticipation of future AI results appears stronger than any type of surging project ROI at this point. What does Halloran think?

Certainly in our case, it appears that companies are really investing. There's a lot of passion around how AI can be adopted, especially now that we're starting to figure out ways to make our knowledge worker labor force more productive and smarter.

So what is Halloran's advice for CFOs feeling the AI heat? Many finance leaders are feeling a nice warm bunsen burner, amped up by the CEO/board: "What are we doing about AI? How do we get to ROI? How do we protect against the risks?"

Halloran agrees we are in a 'gold rush' phase of AI, but he sees an upside: huge investments in compute by 'big AI' players.

We're in a fairly turbulent kind of phase of adoption of this new technology. You used the gold rush analogy, with Nvidia and others that are obviously providing all the tooling and infrastructure to enable this stuff. The great news behind that is: there are gigantic investments going into the infrastructure - that gives us a very good chance of making this real.

The case for AI experimentation - top-down approaches to AI fall short

But then Halloran surprised me. Instead of the usual AI talking points I hear from vendors, he argued for AI experimentation:

What we've seen is that there's a lot of experimentation happening inside businesses today. So I think that's the right approach in this phase, how long will it last - probably another year or two. So you're starting to see a mixture of uncoordinated and coordinated activities happening in companies.

Halloran believes top-down AI adoption will dampen momentum: 

If you try to have a centralized, top-down approach to this, you're going to miss out on a lot of opportunities. In my opinion, this is the type of technology that you want to try to give out into the hands of different teams. Marketing has got different use cases, sales and service has different use cases, engineering and whatnot. We've seen a fairly rapid explosion of different AI tools for those different functions. Companies are letting their employees code, and kind of figure it out for themselves.

But what about the problem of shadow AI, and employees going 'rogue' with external tools that might pose a risk to corporate IP? Halloran thinks that's a controllable problem:

When ChatGPT - just to pick one of the prominent players - first started getting commercially available tools in the hands of humans,  there was a lot of wild wild west stuff happening. Employees were doing some of those things and credit to IT teams, that they were able to figure that out and lock that down pretty quickly.

Hmm.. I'd like to debate that one a bit more, but I agree with Halloran's central point: done properly, there is no inherent conflict between AI security and experimentation.

Give it out to the folks under controls and security, with IT still involved in that stuff, then just see what naturally or organically arises from that. Folks inside a development environment - they will figure things out way faster than a centralized team of so-called experts, which, frankly, none exist in this space at this this moment.

A strong statement - but is it wrong? Yes, gen AI experts with enterprise bona fides are out there, but they are certainly a rare breed. Companies who have those types certainly aren't letting them go without a substantial retention effort. So what's a company to do? Wait for some miraculous AI skills and resource convergence, or press on? Halloran says we should do the latter. Halloran isn't just advising other companies to take the plunge: his teams are doing the same at Planful. Experiments put AI learning into motion:

The best thing to do then is to observe and gather data, information, anecdotes, etc. from those teams. That's what we're seeing the smartest clients of ours do. That's what we're doing at Planful. We have a whole program where we've been monitoring this stuff. Now we're starting to move to a next phase where it's a little bit more organized. We've studied the different tools; we've studied the use cases.

With those proof points in hand, a broader rollout is next:

Now we're putting it out more in a more formal way into the hands of a bunch more employees and about a third of our employee base, and giving them these tools. Then we have a system and method by which we can gather the feedback, gather the metrics that are helping us understand whether they are actually adding value from that for the employees, and therefore the company itself.

Why AI for productivity is a limited conversation

Then Halloran went after something that's been bugging me: the AI productivity obsession. I understand why productivity apps are a logical/conservative starting point for gen AI, but it starts to feel insular pretty quickly. Halloran advises companies to shift their thinking: into AI for growth.

There's been a lot of emphasis on employee productivity in this space. But I agree with something that Aaron Levie, the CEO of Box said recently said. I think the starting point is to think about growth. How can these tools help our company grow? How can they help us invent new solutions to problems for our customers. Productivity, in my opinion, is a secondary goal that companies should be focused on.

Now let's put this in action - how does all of this apply to finance teams specifically?

As it pertains to CFOs, there's tons of automation already available to finance teams and accounting teams. It's very easy to test whether the task was executed correctly or not. If it is executed continuously/correctly, then you use it.

Halloran raises this caution: higher-level AI finance projects will require a longer timeframe.

Where I think there is strong interest, but patience is required, is in the higher cognition activity. Now we're talking about analysis; we're talking about financial planning, updating forecasts in the company. Now we're talking about things that are closer to the mission critical activities of the business.

One: accuracy is paramount. In the spread, we can't get the numbers wrong. Two: these activities require a lot more judgment and historical context.

Cognitive AI for finance is about processes, not tasks

But that's not all: this type of cognitive finance is about processes, not tasks.

Very importantly, from a technology perspective, these are much more about processes. So a lot of the [AI] fixation is on tasks right now. I tend to think about it - and my company tends to think about it - more in terms of how teams get things done across collaborative processes.

For Halloran, processes across tasks is the real potential - and the big challenge.

This is where the big promise of AI is. However, this is also presenting the big challenge for enterprise software companies like us. Where the real value comes is when these technologies help reduce the friction and improve the accuracy of getting a result done... It's easy for me to [generate] code that says, 'I want to add a user to the system.'

But if you think about a collaborative process that might involve dozens of people over weeks - you're doing a task. That task relates to some business outcome. How can the AI seek to understand and then get smarter about what it is you're trying to achieve, not just help you get a task done? That's the higher cognition - that's where I think the grand value in generative AI and its supplementary technologies is going to really come for companies.

My take

Halloran's case for AI experimentation is wildly different than the typical talk of packaged AI solutions I hear these days. This reminds me of the (productive) AI arguments we had at Constellation's Connected Enterprise event this fall, where we talked about failures of corporate AI imagination versus what will lead to a changed/better workplace.

That's not to say Planful doesn't have ambitious AI product plans. I've written about Planful's Predict AI solutions in the past; we can expect a major update in less than a month's time at Planful Perform in San Diego (I'll be on site). I expect to hear much more about Planful's gen AI ambitions this year, but for those who want a flavor, this comment from Halloran struck me as particularly revealing:

The big light bulb for us is processes versus tasks.

We will be unveiling our initial release of what we're putting in the hands of our customers, our vision, and our method for how we're going to continuously improve this over time, and get our customers to buy into it.

I asked Halloran for an example of a compelling internal process 'experiment.' He talked about bearing down on forecasting. Planful has already helped companies regularize forecasting, moving it from a cumbersome annual exercise. But in the push to continuous planning, Halloran sees a major role for both gen AI and other forms of AI, pulling in multiple stakeholders, instantly revising numbers across the forecast. When you consider the thousands of business users who manage budgets pertaining to those forecasts, Halloran isn't exaggerating:

That's a mega process; it can go way too long; it can take weeks to forecasts.

He believe AI can dramatically impact this type of process:

Frankly, I do believe that eventually we'll be able to take a process like that down to a day.

If Planful can pull that off, it won't matter whether Planful calls it AI or 'speed forecasting' or 'automagical finance.' - finance leaders will be intensely interested.

I've been in plenty of debates over how fast AI will evolve and how accurate gen AI can become without a significant scientific breakthrough. I get the sense Halloran and I could have a fun back-and-forth about that (he is more bullish on the speed of self-driving breakthroughs than I am for example). But here's the thing about 'future of AI debates': either side can be wrong, and that's of no use to CFOs who need to hit numbers this year. At Planful Perform, I plan to dig into the metrics of gen AI value with Halloran's team, to further my grasp of what Planful has learned on that aspect.

Put the debates aside for now; where enterprise AI gets interesting is pulling different technologies together to fill gaps. In the forecasting scenario Halloran described, gen AI doesn't exactly excel at mathematical calculations, but other forms of AI/compute can handle the math and predictive/planning elements.

Meanwhile, gen AI can pull processes and people into an engaging flow. Well-designed AI is not about waiting for another breakthrough, but making the most of what it available now, while curbing the downsides and ensuring customer trust. Planful looks set to push a different kind of AI message for finance at this year's Perform; let's see if they deliver.