Debra Geister

CEO @Section2_Inc. Re-engineering AML for the digital age. Focused on AI-driven compliance, risk modeling & ending financial crime. Industry Speaker & Expert.

[Debra Geister @DebraGeister1]

When it comes to AML, the consequences of freezing an account on the basis of a false-positive can really mess up a good person’s life. You can't be wrong.

In such instances, the cure is worse than the disease.

A friend of mine’s 84-year-old mother-in-law was born in Iran but had lived here in the US for decades. While her husband was on his deathbed, the bank had sent a request for additional identity docs - part of a new AML procedure.

She was preoccupied and missed it. And after her husband passed, she discovered all her accounts were frozen, and her bills were bouncing.

So she goes to the bank, to the bank manager she's known for 40 years and learns what happened. She goes home, sifts through boxes and boxes of paper to get the info they need to unlock the account, and goes back in.

"Thank you. This is what we need, but it will take up to 3 weeks to unblock your accounts given our backlog."

What???

This sweet old lady, who’s still grieving the loss of her husband, was facing the prospect of having all her utilities shut off for a mistake that was not her making.

Luckily my friend knew someone senior enough at the bank who fixed the situation before saying:

“I’m really sorry, we are handcuffed. The punishment that we get for one missed violation forces us to freeze accounts. If not, we face millions in fines.”

It’s almost chilling.

But you know what’s more upsetting? Little old ladies are being caught far more often than drug cartels who know how to avoid the flags.

When I was at a bank using the flag system, my false-positive rate was 94%. That means for every 100 customer accounts alerted, we found "unusual" behavior that warranted further review on 6 times out of those 100 customers. That means an awful lot of good customers were disrupted like our unfortunate 84 year old customer.

That also makes it really expensive.

The better solution is moving away from rules-based systems (I.E., is she from Iran?) to patterns-based systems (I.E., Does the patterns of this customer meet the all the 7 factors or indicators of terrorism finance patterns?). The more specific we are the more we can pinpoint with a higher degree of accuracy.

Implementing pattern recognition at my prior bank, we went from 94% false positives to 18%.

Now, we're building that capability into Socure.

[Debra Geister @DebraGeister1]

One of the most common questions I get from customers is, "What are the other banks doing to solve this problem?"

Key learning for me:

Having the perspective of working across hundreds of customers can be highly valuable.

[Debra Geister @DebraGeister1]

I describe my role as being a translator.

Banks speak to me in compliance jargon. Yes, it's our secret little hobbit language.

I translate that feedback into product language: "Here's what the product needs to do to meet the bank's regulatory requirements."

When the product features get delivered, I translate them back into compliance language.

👉🏼 To show demonstrable controls, create a workflow or decision scorecard for each path that aligns with the policy and procedure of the bank.

👉🏼 Here's how you tackle each control to prove compliance:

Step 1. Print the decision workflow and scorecard for each branch of the decision tree (and include in your controls map and procedures)
Step 2. Show metrics of the decision process outputs.
Step 3. Pull an audit sample to show that the system is following orders and making the right decisions.

By the same token, I reach out to our customers to see how the products are working and what needs to change in the product. Speaking a common language and feedback loops make any system work better. Products can't stay static in this ever changing environment.

[Debra Geister @DebraGeister1]

One of the most common questions my team gets is: What are other customers like us doing?

In that sense, having a broader perspective on how to solve problems across customers has become really valuable.

Working in a silo, a single client might take months to get the learnings on which set of configurations are most effective for combatting First Party Fraud.

As a centralized aggregator for hundreds of clients, we get those learnings a lot faster and can share best practices.

[Debra Geister @DebraGeister1]

🚀 My role has expanded.

In addition to leading our sponsor bank team, I now am building two new domain expert teams in fraud and compliance.

The new group will be called "Strategic Domain Support," and our job is two fold:

1️⃣ Be the faces of the industry and providing through leadership

2️⃣ Be the subject matter experts that enable our sales team to design and configure use-case specific solutions.

It's like buying a house.

The realtor is going to show you all the options, but if you want to dive deep on configuring the plumbing, roofing, or blowing out the kitchen, the SMEs come in to support.

[Debra Geister @DebraGeister1]

A lot of bankers will ask me how we came up with the 2 score model for watchlist screening.

It's a long story.

In the past, watchlist screening had always been done with a one-score model. On a scale of 0-100, 100 being the highest risk, a client would set their risk tolerance threshold at 85 or 90 and anything that breached that gap would go to manual review. These scores have always been based on the name alone.

The problem? A LOT of stuff was going to manual review due to common names, close matches and mismatches. And some were being missed due to a name only focus.

So, a group of us at Socure got together and started with a blank sheet of paper. If we throw everything we know about Watchlist tools out the window, how would we design something that greatly improved accuracy while reducing false positives?

The discussion aligned on a new concept: what if we had two scores?

👉🏼 One that predicted whether the name itself matched a high-risk individual? Said another way, "Is the entity a match candidate?"

👉🏼 Then, a second score that predicted, "how likely the applicant was the actual person on the watch list?" Essentially trying to "disqualify" the match on an automated basis.

There were deep debates internally around whether this would work.

Those against it said, "The industry is accustomed to one score; we are going to confuse the market."

Those for it said, "The fact that everyone does a single score and it's not working is exactly why we need to try two scores."

PSA: These debates are healthy and just need to be done respectfully in search of the truth.

Ultimately, we chose to test the 2-score model with a small pool of users and they LOVED it.

Accuracy was greatly improved and that is the product that went on to win an Innovation Award in the Watchlist category.
Moral of the story: Don't be afraid to throw convention out the window in favor of a better solution.

[Debra Geister @DebraGeister1]

Begin with the end in mind.

Ask the questions:

“What does this product look like when it's done?”

“How do we know when we get there?”

Having that clarity helps you move toward that eventuality because if the target is hazy, you are not going to hit it.

By the end of 2025, I want every bank to describe Socure as follows:

“A single platform to make my decisions, create my flows, monitor my transactions, and respond to alerts.”

To make that happen, all we need to do is show them what we’ve built and have them try it. The expanded suite of interconnected products will speak for themselves.

Like the iPhone, you just try it, and it all makes sense.

[Debra Geister @DebraGeister1]

Two common traits among great leaders that many people hate:

1) They challenge all "conventional wisdom."
2) They are not afraid to push.

Yes, assertive leaders can be hard to deal with, but they also drive you to grow if you learn how to thrive in their environment.

So here are the counterintuitive things most people don't realize.

🔵 Assertive leaders don't want you to curl up and fall in line. They want to engage in debate, dialogue, and discourse in a quest to find the truth.

🔵 They respect and seek out people who view the world differently, especially in realms outside their comfort zone.

🔵 They have no interest in vanilla, boring solutions that are in line with industry norms. They know the only way you get breakthroughs is through a process of respectful, often spicy, intellectual sparring.

Seek out these leaders.
Don't be sensitive.
Earn your seat.
Change industries.

[Debra Geister @DebraGeister1]

How a bank officer demonstrates watchlist controls with our new 2 step process:

Step 1:

Establish a name match score representing that a name matches the name on a watchlist. For example, Jose Martinez and Jospehine Martiness would be scored on the similarities and differences of the names.

Bank sets that match score threshold. If threshold is triggered, go to step 2.

Step 2:

Now that we have a positive name match candidate, how likely is it that the person is the actual person on the watchlist? (This is the same thing bank analysts try to determine in their reviews manually.) This score is based upon factors such as:

- Geography
- Age
- Gender
- Etc...

Again, the bank sets the threshold to auto-approve if probabilities are below their threshold (even if auto-approved, the reasoning gets archived as a demonstrable control).

If the risk limit is triggered, the watchlist AI assistant gives a recommended determination in narrative form with reasoning clearly spelled out. The compliance analyst can review in 1-2 minutes and make a decision to agree or disagree.

Fast forward to the regulator meeting:

◼️ Here is an auditable record of all hits.

◼️ We can spot-check any hit that was passed automatically with data behind the decision.

◼️ We can also spot-check any decisions that went to manual review, including the data provided to to the analyst and their decision process according to our risk policy.

Any questions?

[Debra Geister @DebraGeister1]

https://t.co/57Zq29A1V7

[Debra Geister @DebraGeister1]

@ProjectLincoln Would not pay 99 cents for an NFT.

[Debra Geister @DebraGeister1]

https://t.co/7xUanzseTg

[Debra Geister @DebraGeister1]

@kellyclarkson #VoiceTop8 need to change the title of the show to Voices. Way too much talent for just one!!!

[Debra Geister @DebraGeister1]

Hey friends! Looking for anyone interested in a contract role. Need PMP certification and experience in FS. Please message me if interested. Project managers needed!

[Debra Geister @DebraGeister1]

Everyday beauty....especially now we need to see the beauty all around us. #rimeice #minnesotawinter @ New London, Minnesota https://t.co/rLgEuh2Guq

[Debra Geister @DebraGeister1]

Watching a Great show with @rezaillusionist in Branson. Grew up in Pipestone, MN and husband from Montrose, SD. Thanks for the impressive show.
#branson

[Debra Geister @DebraGeister1]

Check out my latest article: A Different Type of AML Training Event https://t.co/atwozsH8Ro via @LinkedIn

[Debra Geister @DebraGeister1]

@KFANRosen Wow, it amazes me that in a post about a fellow human being’s (politician or not) gesture that people would attack an act of kindness and the person that performed it. Unbelievable.
I am so sorry for the pain this is causing you and your family, Mark. Prayers for healing.

[Debra Geister @DebraGeister1]

@MNtoAZ2016 @KFANRosen @LawyerLambert No it makes him a good human being.

[Debra Geister @DebraGeister1]

Interesting Times https://t.co/F1SNh3JWxS'