Achieving finance planning at scale - Planful CTO Sanjay Vyas on how AI can change FPM

Being on the ground in Vegas for Planful Perform 2022 (replays here) was ideal for my goal: get a firsthand update on what matters to today's finance leaders. It was also invaluable for understanding Planful's direction, particularly with their Planful Predict AI solutions. (See my opinionated event roundup, and customer story on Planful Predict: Signals).

But I didn't get everything.

I missed out on a virtual session with Planful CTO Sanjay Vyas  - but conducting the interview after the conference had a big advantage. I was able to pester Vyas with my remaining questions on Planful's AI strategy.

Ivy - Planful's approach to tackling FPM scale

But Vyas had a surprise for me: his big talking point was Planful's new data fabric, code-named Ivy. This advancement didn't get a ton of airtime at the show, though it was featured in the product keynote. Some of Planful's biggest customers are already running on Ivy. So let's start there: why did Vyas start our call by talking about Ivy? As Vyas told me:

Ivy is our new data pathway. It's a massive overhaul of our data layer under the covers. It was driven mainly by larger customers bringing in massive data sets, and processing those data sets. And so it took us a couple of years of R&D. The good news is that we rolled it out already, to our largest customers. And they have seen immense benefits in performance and precision.

When you're a fast-moving cloud vendor, you don't ever want to be inhibited by issues of performance and scale. Vyas believes Ivy solves that for Planful. How does financial precision fit into scale? By the decimal point - Ivy expands the numbers *after* the decimal point you can include in your projects. Vyas adds:

Queue processing, for example, becomes automatic. Once you're got Ivy, you don't have to worry about it; it's much faster; you're getting real-time data. Ivy uses some really cool in-memory techniques to make all this processing faster. By the end of this month, I'm expecting to file our first patent on it. Once that's done, we'll share the details of the patent as well. That's something that I'm really excited about.

And how soon will all of Planful's customers run on Ivy? "By the end of the year," says Vyas. Vyas sees Ivy as integral to Planful's international momentum, supporting companies dealing with inflationary economics, and currency conversion volatility. As we joked on the call (but it's actually quite serious), the difference a few decimal points can make at international finance scale is massive - you aren't going to get accurate projections without that type of precision.

Vyas cited the example of a global travel customer, who tried implementing a competitor's product, and got an unsatisfactory result: 

Then they came to us; we successfully implemented them... They were the first ones to get Ivy, and were very happy with the performance improvement.

Planful Predict - answering my top questions

From what I could tell at Planful Perform, Predict is starting to gain momentum. Though only a minority of customers are live yet, the interest is strong. So I asked Vyas: from his view as a CTO, what would constitute a successful year for Planful Predict? He responded:

The big thing is, we want this technology to be pervasive. You want this to be adopted widely. You want it to be used across the platform. Planful's approach is different - we've taken an approach that this needs to be woven into the fabric of our platform... The data doesn't move to another engine somewhere else.

Vyas referred to competitors that use hyperscalers like Amazon to apply AI features:

We don't have any problems of moving your data from one place to the other. Doing that entails security challenges, but also entails latency when you move data from one place to the other, and that results in data being stale as well.

Whereas in our approach, the data stays where it is. It gets augmented with your Projections; it gets augmented with the Signals data that we find. So for us, in terms of success, I think massive adoption will be success.

Vyas acknowledged that we must also be realistic about the pace of finance adoption. We must show users we can help them achieve their goals. That's why he believes Planful's new "business user experience" will play a key role:

We've invested heavily in the user experience. We know that it's one thing to say,  'Hey, check mark, you got AI/ML.' But it's another thing for users to actually start using it. And that requires an intense focus on usability. And I think that's one key distinction. So that's what we are betting on, is that the usability of the system will drive the adoption.

Planful's engineers told me they were able to accomplish forecasting accuracy with relatively small training data sets. Whether AI can accurately forecast with smaller data sets is an important topic - one that is hotly debated amongst data scientists.

Vyas told me there are nuances to working with financial data:

When I'm looking at clean data from road traffic versus financial data, here's a couple of differences: missing fields. That's just a fact of life. So for example, a company has been tracking travel expenses for sales under a certain account. When the new CFO comes in, they look at the chart of accounts and merge those into a different account. And that account is now slightly different, right? You have the data, and suddenly that account disappears.

The other thing is adjustments. You could have negative balances. Then there's the financial data hierarchy; you can also have rollups to the top value. So, for all these reasons, it was very important for us to be able to account for this, as well as to be able to take a few years' data and do the projections.

My take

Planful's momentum seems to be in a good position to continue, despite the economic headwinds - with Ivy and Predict expanding the possibilities. I'm fascinated by this "smaller data sets" solution, and touched on it in my conference review. Predict: Signals, the first Planful Predict product, is, in my view, less dependent on big data sets (think use cases like anomaly detection). But Predict: Projections is another matter. As I said to Vyas, customers will put Planful's AI projections to the test, against their own manual efforts.

Vyas is confident Predict: Projections will meet this challenge. Putting relevance on more recent financial data is part of the key. Planful also advises customers on how to pull more data from mergers and other events that can factor into projections. It will be interesting to see how close Planful holds its "secret AI sauce" internally, versus sharing winning tactics with the AI community, or filing patents.

As for the adoption of Predict,  the one challenge Planful may face is: some of their competitors include AI-enhanced FPM in the core license. Of course, Planful could still wind up being the better deal - just because you need an additional license doesn't necessarily make the offering more expensive than a competitor's. I just think it will be important to see if a tension emerges between embedding Predict across all of Planful, and charging for the license to use it. Will this hinder Planful's ambitious Predict adoption goals? Vyas told me the way forward is by making the business case:

Take the example of Signals. Even if you found ten errors in your data, and you compute the total sum of those errors in terms of $200,000 - then Signals pays for itself. Some of that will be on us to express the business value directly. We're ready to do that.

We look forward to documenting more case studies as Planful Predict makes its case.