Editor’s note: This is AI Impact, Newsweek’s weekly newsletter where each week, we will explore how business leaders are unlocking real value through artificial intelligence.
Read More on AI
Tap or click here to get this newsletter delivered to your inbox.
Signal Capture
Signals from the frontlines of AI adoption
WorkFusion CEO Says AI Agents Have Moved Past Pilot Mode in Banking
By Adam Mills
The case for AI agents in banking is being tested in one of the industry’s most heavily scrutinized corners: financial crime compliance. Adam Famularo, CEO of WorkFusion, says banks are using the technology because staffing alone cannot keep up.
WorkFusion, a UiPath company, builds AI agents for that work. Famularo said the pressure comes from a simple imbalance: Banks are being asked to review more activity than their staffing models can handle.
“The sheer volume of transactions that are coming through are 10x what they used to be, and it’s growing every single day, every single year,” Famularo told Newsweek.
Banks have long responded to financial crime risk by hiring compliance analysts, adding systems and answering regulators. Famularo said that approach has limits when the number of alerts keeps rising.
“The amount of alerts that were being created were more than any human can manage,” he said.
At one top 10 bank, Famularo said, meeting regulatory expectations would have meant hiring about 1,800 workers, then training and retaining them in a role that is difficult to staff. Instead, the bank brought in Evan, WorkFusion’s adverse media screening agent. Famularo pointed to a recent run as evidence of the scale involved.
“Eighty million entities screened in one day,” he said.
That kind of use case has made financial crime compliance an early proof point for AI agents—high-volume work, existing analyst processes and decisions that banks can compare against what humans already do. Famularo said WorkFusion’s agents are no longer limited to trial deployments.
“We have over 10 of the top 20 banks that are live, in live production with our technology today in one of these areas,” he said. “So we have live use cases everywhere.”
The clearest starting point has been screening, including checks against sanctions lists, customer names and adverse media reports. Those use cases, Famularo said, have been the fastest for customers to adopt.
“When we focus on screening alerts, it’s all about volumes,” Famularo said.
Other areas require a different role for the technology. Famularo pointed to fraud, investigations and know-your-customer reviews (KYC), the checks banks use to verify customers and assess risk, as areas where agents are used to help analysts move through more involved cases.
“What we’re trying to do [in those cases] is remove a lot of the complexity so that it can make it easier for those fraud people to be able to do their work faster,” he said.
False positives are one reason banks can put numbers against the work. Compliance teams spend large amounts of time reviewing alerts that ultimately do not require action.
“Anywhere from 50 to 70 percent time savings,” Famularo said. “We can fully adjudicate those false positives literally 70 percent of the time.”
The agent model, he said, can play a larger role in the review process.
“We actually know, based upon our models, where this has been approved or denied before,” Famularo said. “We then, based upon those approvals or denials, can then write out our logic flow of why we blocked or let somebody go through.”
Famularo said implementation is where confidence builds.
“At the end of the day, the banks need to see it running in their system,” he said. “But once they can see it run, and then we can do some training of the model to be specific to the bank, then they can build and develop that confidence.”
One of WorkFusion’s less conventional bets was cultural rather than technical. The company gave its agents names, faces and identities, an approach Famularo said helped customers treat the technology less like another software tool.
“The most successful companies were the ones that treated our agents as part of their workforce,” he said.
That sort of framing also shapes how Famularo talks about jobs. He pushed back on the idea that banks are mainly using the technology to remove people, saying the bigger change has been capacity.
“I haven’t seen many people go away,” he said.
For level-one screeners, Famularo said fewer people are needed in the most basic review roles, while more have moved into level-two screening, higher-end work or other parts of the bank.
Famularo said the next phase is likely to create more reasons for banks to lean on the technology, especially as AI and other tools increase the amount of activity moving through financial systems.
“You’re just going to have more transactions, more work to be done, more bad actors trying to leverage systems to do bad things,” he said. “I think it just opens us up to find more ways to use technology to stop those bad actors as time progresses.”
Core Intelligence
Genpact’s BK Kalra Says AI Agents Need to Understand the Work
By Adam Mills
Agentic AI looks different when software is expected to act inside invoice processing, claims handling, procurement approvals, compliance checks and other enterprise processes that have been patched for years.
Genpact is trying to turn that operating layer into an advantage. The company built its business in finance, procurement, supply chain, insurance and risk operations, first inside General Electric and later as an independent business-process services company. CEO Balkrishan “BK” Kalra argues that history gives Genpact the context necessary to make AI agents useful inside real workflows. “That distinctive domain is shining more,” Kalra told Newsweek. “We are now seizing that moment.”
The reason, Kalra said, is that the agentic AI market has become crowded quickly.
“You hear a lot of noise in the market,” Kalra said. “Everybody is agentic.”
Genpact’s argument is that the real difference shows up in the last mile of enterprise operations, where exceptions, handoffs and judgment calls do not fit neatly inside standard software or public training data.
The stakes are higher because many companies are trying to scale AI across work environments that are already fragile. A new HFS Research report, produced in partnership with Genpact, surveyed 2,002 enterprise executives and found that technology, data, process and talent “debt” are limiting AI value. The report found 85 percent of leaders believe those debts limit AI value realization, while only 6 percent were classified as proven debt remediators.
Old technology is only part of the problem. Large companies have spent years working around fragmented data, manual steps and employees who know how to fix problems because they have seen them before. Agents can expose those weaknesses quickly because they do not merely produce information. They can start moving work across systems.
Vijay Vijayasankar, Genpact’s global agentic AI officer, said models need the right structure around them before they can be trusted in production workflows.
“Those models are super important, but the model by itself very rarely helps in an enterprise context,” Vijayasankar told Newsweek.
The risk is that automation reaches the work before the work is ready. If agents are added to broken workflows, they can carry incomplete data and poorly designed handoffs deeper into the business. Elena Christopher, vice president of strategic programs in Genpact’s chief growth office, described the danger plainly.
“They’re doing perhaps the wrong steps faster with nobody watching,” Christopher told Newsweek.
Enterprise AI is becoming a test of process knowledge. Companies need to know which tasks can be handled by large models, which require smaller models or rules-based software and which steps need human review because the consequences are too sensitive to leave to probability.
Genpact’s broader bet is that agentic AI will reward companies that understand the work deeply enough to rebuild it. Agents layered onto old systems, bad data and unprepared teams can make the engine room more brittle instead of more dependable.
You can read more about Genpact’s pivot and the research in the full article here: Genpact’s Agentic AI Bet Starts in the Engine Room.
Upcoming Webinars
Is India on the Right Side of the AI Trade?
India’s role in the global technology economy has long been shaped by scale: a massive IT services sector, fast-growing capital markets, digital public infrastructure and a young workforce. AI could raise the stakes even further by changing where value is created, who captures it and how countries compete.
In an upcoming “AI Impact Forum” session, Dr. Ranjit Tinaikar speaks with Shri Ashishkumar Chauhan, CEO of the National Stock Exchange of India, about whether India is positioned to benefit from the AI revolution—and what the technology could mean for the country’s economy, financial markets, market infrastructure and next stage of growth.
Join the live discussion on Thursday, July 23, at 10 a.m. Eastern. Register for free.
Prompt Injection
What’s one recent insight you’ve learned about AI?
“AI innovation is often diluted through consensus, risking the loss of meaningful progress. If your AI deployment isn’t causing reactions internally and with your customers, you’re not being opinionated enough.
The market is flooded with ‘AI-powered’ everything. But there’s a vast difference between bolting AI on and rebuilding with it as a foundation. Bolting on adds convenience without disrupting the underlying structure. Nobody’s upset. Nobody’s confused. That’s the tell. True foundational change shifts how people interact with your product and where they find value.
Being opinionated isn’t about surface-level differentiation or small experimental additions. It’s about deliberate decisions in the core product that not everyone will agree with right away. As more and more AI capabilities come to market, access to models is no longer the differentiator. What matters is the prioritizing choices you make and how strongly you embed your point of view into the product experience.
But here’s what cuts through all of it. Get to customer value. If you’re delivering better outcomes for your customers, that’s unarguable. Nobody can debate whether you should be using AI when the value is impossible to ignore. That’s where focus belongs.”
Have your own lesson to share? Email us at: a.mills@newsweek.com
Run Log
AI use case of the week
By Adam Mills
To a shopper, blow dry cream and hair serum belong at different moments in a hair routine. To an AI shopping model, the product pages can look almost identical.
Kimberly Shenk, co-founder and chief executive officer of Novi, a company that helps CPG and retail brands improve how AI shopping systems understand and recommend products, has been studying why products can be routed to the wrong category in AI-driven shopping experiences.
Novi used AI to test how shopping models interpreted product labels and attributes. In one analysis, the company focused on blow dry cream because the category shares language with hair serum. Both can be described with terms such as moisturizing, frizz control, heat protection, lightweight, silicone-free and plant-derived. At every label weight Novi tested, hair serum remained the most common misclassification for blow dry cream products, showing that a stronger category label alone did not fix the confusion.
A second category behaved differently. Plant-based milk also contains language that can overlap with other products, but the word “milk” gives the model a clearer signal. Under the highest category-label weight Novi tested, every plant-based milk product was routed correctly, and none landed in nut butters.
The analysis showed where product data can fail. A label helps when the category name gives AI a clear clue about where the item belongs. It is less useful when the surrounding attributes sound like those of a neighboring category. In those cases, the model may need stronger signals about how the product is used, when it is used and what alternatives it should be compared with.
“AI does not understand product context the same way a shopper does,” Shenk said.
Brands cannot assume product descriptions written for human search will work the same way in AI shopping. A shopper knows blow dry cream is used on damp hair before heat styling and serum is often used after styling. A model may only see overlapping words unless the product data makes that distinction clear. Before AI can recommend the product, it has to understand the shelf.
Have an AI use case to share with us? Email us at: a.mills@newsweek.com
Context Window
■ AI leaders met with G7 officials as Europe and other U.S. allies weigh tech sovereignty concerns, after U.S. model-access restrictions underscored the risks of relying on American AI providers. [ABC News]
■ HSBC plans to add more than 200 AI use cases through an expanded Google Cloud partnership, with early priorities in customer personalization, financial-crime risk and staff productivity. [Computer Weekly]
■ BCG says AI agents could remake corporate finance within two to five years through real-time close, automated reporting, dynamic forecasting and a potential 50 percent reduction in staff needs for today’s workflows. [BCG]
■ Nvidia CEO Jensen Huang said wider AI adoption will require new social norms, clearer national security rules and more U.S. energy capacity as public concern grows over jobs, data centers and regulation. [AP]
■ A new policy proposal urges federal agencies to improve how they track AI’s workforce effects, including through new survey questions on AI use, employer-provided tools and the relationship between adoption, wages, hiring and job displacement. [FAI]
Transfer Protocol
Tracking executive moves across the AI landscape
Katie Gross, previously president of Suzy and a senior leader at Cint, Toluna and Mintel, has been appointed president of Panoplai, where she will drive commercial strategy and growth for the company’s digital twins, synthetic data and AI-powered insights platform.
Jane Barrett, who has served as Reuters’ head of AI strategy since 2024, has been named head of product at Reuters, bringing newsroom, AI and product experience as the organization folds AI more directly into its product and engineering structure.
Barak Turovsky, previously chief AI officer at General Motors and vice president of AI at Cisco, has joined Paramount as executive vice president and head of consumer AI, where he will help develop AI-powered personalization, content discovery, engagement and platform intelligence across Paramount+ and Pluto TV.
Travis Nixon, previously senior vice president of artificial intelligence optimization at Dollar General, has moved into the role of senior vice president and chief data and artificial intelligence officer, overseeing AI strategy, delivery operations, data engineering, enablement and business process management.
Nadia Carlsten, previously chief executive officer of DCAI and vice president of product at SandboxAQ, has been named president and chief executive officer at Smartbird, where she will lead the AI infrastructure provider as it builds dedicated AI infrastructure as a managed service.
Know someone on the move in AI? Send job change info to a.mills@newsweek.com
Magic Moment
What’s the most fun or unexpected way you’ve used AI lately?
“I split my time between Sydney and New York, roughly six weeks in each city. Keeping plants alive on the other side of the world turns out to be very difficult. For a long time, my approach was basically watching them slowly die over my Amazon video monitor.
Last time I was back in Sydney, I decided to fix that properly. I planted new seeds, installed a soil monitor that tracks temperature and hydration, and set up two Ring cameras pointed at the plants. Then I built an agent. It pulls in the camera feeds, the soil data, local weather, and sun exposure, analyzes everything, and uses that to calculate how much water the plants need, which is delivered by controlling a Wi-Fi-connected tap I integrated into the system. It sends me an alert when something needs human attention, like fertilizer.
Three weeks ago, I went back to Sydney and discovered the plants had completely outgrown their pots and were hanging down to the floor. I like to think no one can take care of my plants better than I can, but the agent is a reasonable substitute.”
Experience some AI magic? Tell us about it at a.mills@newsweek.com
Tap or click here to get this newsletter delivered to your inbox.