How Orum is Enabling Real-Time Payments (part one)

Founded in 2019, Orum is creating financial infrastructure allowing for real-time money movement in the U.S. In its own words, “Orum’s API-based products will give enterprise customers unique funds-availability modeling, empowering them with the information required to shift to real-time payments, and enable access to real-time payment rails without necessitating bank integration.”

Orum’s debut product, Foresight, provides predictive intelligence for ACH transactions, using machine learning to predict funds availability and flag potential fraud. Before a transfer is initiated, clients who have integrated Foresight’s two APIs, receive information about the transfer’s probability of return, risk percentile, and other outputs such as maximum transfer amount and recommended hold time.  This helps partners identify return risk before it occurs, including NSF (non-sufficient funds) transactions, a huge source of friction. Foresight’s ML models are trained to the unique properties and risks inherent to each enterprise customer and continually updated.

Orum’s second product, Momentum, selects payment rails for any given transaction based on each enterprise’s preferences for speed, cost, and risk, intelligently directing some to ACH, some to the RTP Network, or some to wire transfers, for example. Orum will also integrate crypto and eventually, FedNow. Momentum’s predictive intelligence is powered by Foresight and both Foresight and Momentum are API-based.

What really animates the Orum team seems to be helping consumers. Eliminating friction in the financial system has the potential to improve lives. The financial difficulties caused by the COVID-19 crisis underscored the importance of giving people faster access to their money. Making funds available sooner could theoretically help consumers avoid payday lenders. Moving money in near real time will cut down on overdraft fees.

Founded by fintech veteran Stephany Kirkpatrick, Orum is a fully remote and geographically dispersed team and currently has 45 employees. In early April, the firm announced a $21 million in Series A funding led by Bain Capital Ventures. Inspired Capital, Homebrew, Acrew Capital, Primary, Clocktower, and Box Group also participated in the round. Stephany, formerly GM of LearnVest and LearnVest@Work, is CEO.

Stephany Kirkpatrick

Q.    Stephany, what percentage of ACH transactions can clients settle same day using Foresight?

A.    It’s such an interesting question because ACH has many variants. There is the traditional path, which is typically settled the next day, and there is same day. So, technically you could settle on a same-day basis though you pay a higher fee for that transaction.

But the way we think about it at Orum is to ask if we can identify, using machine learning and a private, proprietary data network that we’ve built, the settlement risk. Can we provide that intelligence at the transaction level well before settlement occurs, well before the enterprise moving the money creates the file and processes it? 

Then, how can we go a step further and look at the T+60 return risk, which is where the tail-risk fraudulent activity lives? Because ACH isn’t just a settlement system reconciling transactions over the next three days. Unlike wires and other payment rails which have a “good funds” model, ACH has 60 days after a transaction settles in which a consumer can come back and say, “It wasn’t me.” That’s where we see a lot of fraud occur. It can cost companies a lot of money and is ultimately what causes banks and financial institutions to slow down consumer access to funds during that settlement window.

So the answer to your question, Jim, is that the percentage of transactions that are likely to be low risk is quite high. That varies of course with the customer base and the platform. What our product, Foresight, is designed to do is identify which transactions are low settlement risk and low return risk. We find right now that on average 85% have eligibility for same day provisional credit. While the transaction may not have settled because it’s still pending on the traditional rails, it is possible to identify the risk parameters and feel confident letting the end user have access.

Again, I would call it provisional credit, but it’s a really interesting way to essentially create synthetic real-time money movement by just adding an API and not changing anything else in your system as it pertains to ACH — in particular, no core changes. So it’s a very easy way to get to a place where money feels instant a good majority of the time.

Q.    Can you quantify how much money is lost to the sorts of friction in payments that Orum aims to eliminate?

A.    ACH is literally what makes the U.S. financial system go ‘round. Yes, we’ve all done a wire, sure we write checks and of course we sometimes pay with cash or cards, but the vast majority of transactions — whether it’s mortgages being paid, rent being paid, moving money from my Bank of America account to my Fidelity account to invest, putting money into Venmo so I can pay you, Jim, back for brunch — all of those things are powered by ACH.

And now, e-commerce is actually looking to do more. There are lots of companies thinking about how you checkout and pay with your bank. ACH is a huge part of our system — nearly $62 trillion worth of money movement last year alone, and over 26 billion transactions, according to Nacha. That volume has increased by more than 1 billion every year for the last six years, and the value has increased by more than $1 trillion every year for the last eight years; so it is a large chunk of the payments space, and is only getting bigger.

When an ACH transaction follows a “happy path,” the receiver of funds is paid and all is good. However, ACH transactions can be returned for a variety of reasons – if an account is closed, if there is a stop payment, or if a name doesn’t match, for example. 

One of the most common reasons for an ACH return that accounts for by far the largest volume of overall returns is because an account does not have the funds available, or is NSF (“non-sufficient funds”). This results in both the company not getting paid as they agreed upon, and the consumer getting charged a hefty fee by their financial institution – creating a vicious cycle that helps no one. How could we turn that around? How could we say that it doesn’t take three to five days for my money to get from point A to point B, and more importantly, ensure I won’t be penalized because when the money is drawn down it will not only settle faster but there will be intelligence helping determine availability of funds so it can be withdrawn from my account at the right time?

That’s not the next Friday or the next payday, because we know that payday isn’t any longer an every-two-week thing. We have lots of companies pioneering early wage access. We’re moving to daily wage availability. The question becomes, when should the transaction happen? That’s not always going to be tied to the obvious scenario we’re used to using. There’s a lot of friction that we can unwind and many costs that can be saved, leading to banks and financial institutions offering far better products. 

It’s conceivable that banks partnering with us would never charge you an overdraft or an NSF fee. They would know when you’re going to drop below zero based on Foresight, and they would offer you a $100 credit to use instantly at a really low interest rate, so you never go below zero. This does away with a lot of predatory lending. Getting rid of these problems is the beginning of an incredible next chapter for how our financial system could operate. And it super-serves the underserved. 

Furthermore, we hypothesize that because of faster funding, existing consumers will want to move money more often, which incentivizes them to save more. Our analysis of one of our partner’s data has showed 2x more transfers when they received faster funding. 

Q.    But do banks really want to do that? They don’t want to walk away from what they are making in annual overdraft fees, do they? I could see banks not liking this idea very much.

A.    I’ve been in the financial services industry since I graduated college. You used to be able to make a killing on up-front commissions and A Class shares, etc. But where are we today? We’re at zero commissions and zero trading fees. That isn’t because E*Trade, Fidelity, Robinhood or anybody else wanted that to happen. It’s because consumer demand has required that the way big financial institutions make money be done according to what best suits the customer. We have real suitability requirements to think about. 

Banks don’t love the idea of transitioning away from what is essentially guaranteed income, but I do think that banks recognize and understand that there are now different ways to build financial products that produce alternative streams of revenue and that actually retain customers and that protect and watch out for those customers. 

When we talk to bank partners and say that we can cut fraud by 50% and reduce NSF by 40%, and that because every time you have an NSF, you incur operating costs before you can charge the customer, the costs of guessing wrong versus the value of seeing it right with Foresight are pretty obvious. When you get into conversations about how you can increase net promoter scores, and how you can decrease the number of calls to customer support (which is a cost center), there are many reasons why we have a lot of traction with banks and fintechs who want to move into this next generation. 

Q.    How many transactions is Foresight currently processing each month? 

A.    I can share a little bit. Just remember that we’re in private beta so the numbers are still held pretty close to the vest. We are scoring upwards of 1M transactions a month, and this number is growing 90% week over week. We scored over $150M+ in transactions in the most recent quarter.

We work with different fintechs, challenger banks and investment platforms, and they serve a super-wide variety of customers. It’s very early days for us, and the reason that our volumes are to us high, and probably to the rest of the world low, is because when you are building a product that moves money and is designed to do real-time risk management, you approach each customer very thoughtfully, you hand pick the best partners in the market, and you move very methodically through the process of confirming that everything is working from end-to-end before you try to drive maximum volume. But we’re really coming through this private beta period seeing exceptional results. 

Our partners are thrilled. We had a bank integrate with us that was bumping up against its Nacha limit, and sometimes going over that 15% overall permissible return threshold. In less than two weeks, we had cut that in half. So we’re providing real, meaningful value. Depending on the problem you’re trying to solve, it could be faster funding, it could be fraud reduction, it could be NSF reduction. There are lots of different value props, but ultimately as we continue to expand access through the back half of this year, the volume that we’re processing will multiply 10-20X very quickly.

Q.    How accurate is Foresight at flagging fraudulent activity? Are you flagging fraudulent activity that partners were previously missing, and are you able to avoid false positives?

A.    Every system has its imperfections and no fraud solution; no real-time intelligence platform will catch 100% of risk. We’re very clear about that. We’re not here to replace every other mechanism you have for catching fraud.

But when you look at ACH and you look at the transaction-level information about the probability of a return and then what type of a return — cash flow based or fraudulent — our results are very, very effective – the AUC is .85-.95 for our customers. In terms of the false-positive rate, that’s one of the reasons why we have our partners still manually review and decide what they want to do with the risk information. We don’t dictate, we consult with you to design your implementation and you could be different from the next partners in terms of risk tolerance. If perfectly good transactions get caught due to your risk-threshold, manual reviews can help protect the customer. But ultimately our job is to identify things that look anomalous.

You, Jim, might have a history of small dollar transactions and then do one for $20,000. It’s relevant to flag that as outside the norm. (Size of transaction might not be the key variable here but it’s a good example.) We would be remiss in saying that because we’ve known you as a customer for six months, anything you do is good. Because we don’t even know if it’s you. We don’t know who you are. We have no personally identifiable information (PII). We don’t know names; we don’t know credit worthiness. What we look at is about 55 parameters including which device you’re logging in from, what type of transaction you’re doing, what time of day it is — there are many things that drive the machine learning here, not just these simple things — that suggest whether a transaction should be reviewed. Our system is very effective at catching fraud. We’re seeing a 50% reduction in fraud across the board. Holding money for three days (which has been the standard for risk management) is not an obvious way to identify what is going to happen 60 days later. 

We’re matching historical patterns via trained datasets that go back into supervised models as one mechanism, and we’re able to draw up that information into alert systems to let all of our customers know. While Platform A might be experiencing a hot attack, we can alert everybody else to the parameters of these bad transactions so they can block them before they arrive. It’s a really cool way to think about real-time risk management, which is fundamentally what we need to get to real-time money movement.

Q.    Are you bringing any third-party data into the platform to aid with fraud detection, or is your determination purely based on the data that surrounds the individual transaction?

A.    It’s a combination of licensed data and enterprise consumer data. We’ve built up a data network where we can identify bank accounts across multiple financial platforms, and we’ve seen from our data that consumers use up to 5 financial platforms from a single bank account. 

Everybody benefits from contributing information that on a de-identified, aggregated basis can allow us to pattern-match what is happening with how an account and routing number is being used here, for this transaction, versus how it’s been used to pay a Verizon bill or for a mortgage or for an investment.

Look at the way we built our data network, which today reflects about 20,000 different platforms who all have some type of ACH. Putting money into Venmo is an ACH that is as interesting to us as a Quicken loan that is repaid automatically every month because we see all the behavioral parameters. We look at data that helps us identify anomalous patterns indicating outcomes that ultimately lead to loss for the platform. 

We use only proprietary data, and the model is different in that we are not just waiting every two weeks to update the score or the risk and again we’re not looking at it from the creditworthiness perspective, or even looking at who the person is, but we are bringing in data on a real-time basis so our partners will generally hook up to our streaming APIs, event-based systems that will pull in datapoints as parameters change. We also support batch-based file exchange, so these partners will send data every 30 minutes, meaning you’re still seeing a pattern unfolding in real-time. It matters way less what happened six months ago than what’s going to happen in the next six minutes. That’s really how we think about it. 

Q.    How do you price your two services?

A. Our Foresight product, which is focused on ACH intelligence, is designed to be priced in a way, based on tons of research, lots on conversations, that suits the needs of our partners. And what the partners need is risk management they can rely on whenever they feel they need to check. Unlike a unit-based, per-API-call product, which you will see in the market, and which certainly has validity for some use cases, we believe ACH intelligence is something you should be able to check more than once a day. The way a batch-based system works is to say that you might have a transaction that you request at 9:00 a.m., but the enterprise moving the money isn’t going to send it out until just before the last cutoff window closes at 5:00 p.m. What’s been happening in the interim?  What other activity have we seen that suggest that maybe that was a high probability of return transaction? 

If we charged on a unit basis, people would be dis-incented to check again because of the cost, even though there’s a clear ROI. So we decided the best path is to make it a subscription, so you pay a licensing fee tiered against our emerging, our growth, and our enterprise customers, and it’s unlimited use. As a result, we find that partners build retries into the platform, so that intra-day they can re-look at the level of risk so that they have the opportunity, right before the ACH goes out for processing, to evaluate one last time and make a final decision. I think that’s a super-effective way of using the product in the way it was designed. 

What excites us about this is that it also helps partners who are growing rapidly. Many of our customers are in high-growth mode which is why they want to be able to move money faster and with lower risk and cost. They don’t want to be penalized for that growth. If you have to pay on a per-transaction basis and your transactions are growing two or three X each month, then you feel penalized for growing and you don’t have a clear line of sight to your costs. We thought it was really important to meet growth partners where they are.

Q.    Pricing is one of the hardest things for B2B fintechs to get right, particularly when they are just starting out. It’s a difficult but interesting challenge.

A.    Our pricing is something that will continuously evolve. What we’ve heard, and I think it’s true, is that old-school SaaS pricing is out. The idea that you would charge an implementation fee when it’s just an API requiring a few lines of code. Or that there are up-front costs? That’s out the door. 

Certainly, unit-based pricing makes sense for some models. On our Momentum product — which actually does move money and takes care of every aspect of the transaction — we have to use unit-based pricing because it’s for a given transaction.

Q.    Are you working with processor banks behind the scenes?

A.    We’re working with a couple of banks for processing. It was important to us that it not be just one bank, whereas I think you’ll find many other companies build products using a single solution. Before we’d even written a line of code, we knew we wanted to launch with two big, amazing partner banks and a third to fast follow. And obviously all types of rails — ACH in all flavors, wires, and the card rails. And real-time payments. We’re on track to integrate with FedNow. There’s also a quickly emerging thesis around our interests as a country and the movement toward digital currencies with settlement on the blockchain.

The way to think about Momentum is to say that we’re future-proofing the way you move money. You’re never going to want to integrate with a single set of rails or with each individual solution. 

Integrate with Orum, and we do the same thing Amazon does with same-day delivery. We optimize for speed, cost, risk, day of the week, based on each company’s preferences. With Amazon, you don’t care if the package came or UPS or FedEx or DHL. You just care that it came instantly. The seller doesn’t care either. They leave it to Amazon to determine the best, fastest, smartest route. We do the same thing. Our system is designed to think about the optimal way to pull funds, and the optimal way to push funds in a way that consistently achieves 24/7, 365 money movement.

# # #

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.