While excited by the potential of generative AI to change the trajectory of healthcare forever, Swedish Health Services CEO Dr. Elizabeth Wako is cautious about potential potholes on that path. “It’s the biggest breakthrough in many industries and certainly my industry, where AI is critical to the future success of healthcare, but the pain it will take to get from here to there is not going to be small,” she says.
The largest nonprofit health system serving the Seattle metropolitan area, Swedish Health Services encompasses five hospital campuses, more than 1,500 licensed patient beds and 48,000 annual inpatient admissions, bringing in $3.4 billion in revenue. Like other CEOs, Dr. Wako, whose credentials include both an MD and an MBA, worries about the risk of incorrect or biased AI inputs generating dangerously wrong outputs.
While studies from researchers like Harvard and McKinsey suggest that AI could save up to $360 billion in annual healthcare expenses, this promising future is obstructed by a pronounced lack of trust. “As it stands today, gen AI is a tool that fabricates things,” says Wako. “It’s full-on hallucinatory, a ‘black box’ in terms of where the information came from and who owns it. To make a mistake in my industry is to change the trajectory of a person’s life.”
These substantive concerns have not thwarted the CEO’s early-stage AI investments in gen AI at Swedish, an affiliate of the large Providence Health & Services system. Wako has approved capital investments in a range of gen AI tools to reduce the burdens of manual work for clinicians and improve the patient-doctor experience, in addition to the typical gen AI tools used by employees in different business functions for content creation, research assistance, software code generation and writing emails. But her hope for gen AI is to power an unprecedented rise in patient care and clinical team outcomes.
“I’m most excited about a future where all the data is pure and accurate, so patients with similar labs, vital signs, demographics and even DNA are provided medications and a standard of care based on this information,” she explains. “We’re not there yet because healthcare in the past was inadequate or historically biased, resulting in clinical decisions that weren’t always the best. All AI knows is what was done in the past, increasing the risk of propagating the same errors.”
Significant Blind Spots
Similar hesitancy and hope are writ large across all industry sectors. While 65 percent of companies are using gen AI tools in customer service, sales and marketing, software development, product development, R&D, supply chain and other functions, only 3 percent have scaled a gen AI use case in an operations-related domain, according to a set of two 2024 surveys by McKinsey.

—Dr. Elizabeth Wako, CEO, Swedish Health Services
Such advanced-use cases entail the sourcing and integration of data from across the entire value chain to obtain a more comprehensive view of business operations. While a laudable goal, it’s a vastly more complicated and riskier gambit than present-day gen AI tools. The reason, as Wako pointed out, is the threat of suboptimal data. If data is low-quality, insufficiently trained or corrupted by biases, among other failings, and algorithms are poorly constructed, incorrect assumptions that pop out of the box will promote bad business and customer decisions.
“Many CEOs are held back by the fear of AI getting it wrong,” says Javier Saade, founder and managing partner at Impact Master Holdings, an advisory firm to high-growth businesses in technology, financial services, media, consumer goods and services and other sectors. “The potential is boundless, but the risks are infinite.”
Since gen AI-generated outcomes can be wildly inaccurate, misleading and even dangerous, the risk-reward ratio is asymmetric, Saade says. “An example is a super-intelligent computer that decides a particular molecule should be put into a compound to fight a disease, but instead the molecule causes extremely adverse health issues for certain groups of people,” he explains.
That’s just one example of what can go dreadfully wrong. Saade advises checking out the AI Risk Repository created by researchers at the Massachusetts Institute of Technology, cataloging more than 700 AI risks, from biased opinions to machine language hallucinations. “The risk literature should give us pause,” MIT’s researchers warn. “We are potentially in a situation where many may believe they’ve grasped the full picture after consulting one or two sources, when in reality they’re navigating AI with significant blind spots.”
As the blind spots become identifiable and AI model outputs become more precise, the fear of getting it wrong will transform into the fear of missing out. “Sometimes you have to slow down to speed up,” says Saade. “We need some common truths as to how these systems work and where they can be most useful across a business.” The four CEOs we interviewed are taking this deliberate approach to advanced gen AI use cases, assembling internal councils, data quality initiatives and ethics committees to study and learn from what can go wrong to make sure they get things right. Meanwhile, they’re vigorously investing in cloud, automation and gen AI tools, building them internally, purchasing them off-the-shelf or some combination thereof. As Saade, a member of several boards serving publicly traded and private companies investing in AI, puts it, “There’s too much good to do nothing.”
Greater Productivity, Better Care
This is certainly the case at Swedish Health Services, which has a chief data officer leading a work group developing AI guardrails, a chief information security officer leading its information protection committee and a chief ethicist leading its data ethics council. “These three teams feed into our gen AI leadership council, which helps make decisions ensuring data is safe and secure, doesn’t propagate biases and is equitable and ethical,” says Wako.

—Sandra Campos, CEO, PetMeds
The leadership council has given the go-ahead to a handful of gen AI tools, including one designed for clinicians called MedPearl. The tool addresses the challenges primary care advisers confront in giving healthcare advice to patients. “Protocols for particular diseases are constantly changing. MedPearl reduces the cognitive burden of carrying all that clinical data in your head, improving primary care clinician productivity and the quality of referrals sent to specialists,” Wako says.
Another gen AI tool approved by the leadership council is DAX Copilot, an AI-powered ambient listening solution that records and writes up the symptoms a patient expresses to a clinician. “The tool picks out the specific words documenting the symptoms and the physician’s responses and recommendations, freeing them from taking notes. Patients love it because they receive one-on-one attention,” Dr. Wako says.
Also sanctioned by the council is ProvARIA, a gen AI tool using natural language processing (NLP) to organize a physician’s inbox messages based on their content and urgency. For example, a message from a patient reporting severe symptoms goes right to the top of the inbox for immediate attention, whereas a question about a clinic’s hours falls to the bottom. All three tools were designed and developed internally with assistance from third-party partners like Microsoft, OpenAI and Oracle.
Catching Up Quickly
CEO Sandra Campos also is doing a lot more than nothing at publicly traded PetMeds (net sales of $281 million in FY 2024). Campos is on a mission to automate every function and department at the online provider of pet food, supplies and medications to more than 100,000 veterinarians nationwide. “My goal is to create a seamless journey for the customer, giving them the easiest, fastest and best experience,” says Campos, former CEO at luxury global lifestyle brand DVF (Diane von Furstenberg).
She is in the early phases of this transformation, having just become CEO in April. The 28-year-old company has not made significant investments in technology in recent years, a “strategic plus,” Campos says, given the opportunity to make sizable strides in performance.
A case in point is PetMeds’s customer call center, which accounts for 10 percent of sales volume. When a veterinarian contacts the call center with a problem, a quick solution is a viable way to cross-sell and upsell products. “We’re implementing a gen AI tool that listens to the customer’s problem and is inputted with prompts simplifying the work for call center personnel to better serve customer needs, thus freeing them up to suggest other products the customer might be interested in,” Campos says.

—Matt Oppenheimer, CEO, Remitly
She plans to plug the gen AI tool into PetMeds’s customer relationship management (CRM) system, giving call center personnel instant access to information about each veterinarian and the individual pets served by the business, such as their names, ages, prescription needs and whether the veterinarian receives automatic shipments, which is a way to generate recurring revenue for the company.
Other gen AI tools are slated for implementation. In contract management, for example, Campos is looking for a tool that can consolidate the enormous volume of contracts in its system, not just its 100,000- plus customers but also employees, vendors and other third-party organizations. “By having all this data in one place to read and synthesize, we can find correlations and dissimilarities in legal terminology, helping us pursue a more synergistic approach in managing contracts in the future,” she says.
In purchasing and merchandising, Campos plans to revamp the functions with a gen AI tool that improves demand forecasting, ensuring appropriate inventory volumes based on a combination of expected sales and inventory turns. In accounting and finance, routine accounts receivables and other tasks like matching each account’s revenues and expenses are being automated. She is now hunting for a gen AI solution that can collapse all the data into a single view for decision-making purposes.
Once the business functions are duly equipped with gen AI tools, the next step in the AI journey is an advanced gen AI use case for an “orchestration layer,” Campos says, “a connector across all the different siloed tools, bringing them together from an operations and financial standpoint.”
Moving Forward
In contrast with PedMeds’s early-stage AI investments, PagerDuty, a global leader in digital operations management, has used gen AI models for more than a decade. The reason is the critical service provided by the publicly traded company with 1,200 employees and $474 million in annual recurring revenue. “If a customer’s server or an application is down, we pinpoint and triangulate what happened to get them back up quickly,” says CEO Jennifer Tejada.
More than 29,000 organizations use PagerDuty’s IT operations management platform for this purpose, including half the Fortune 500 and 70 percent of the Fortune 100. By extracting information on other incidents in which similar servers or apps stopped functioning properly, the gen AI tool compresses the time it takes to quickly restore client operations while also adding the event to its list of incidents to assist other customers. “Time is money when mission-critical systems are down,” Tejada says.
She points to the faulty software update issued by cybersecurity provider CrowdStrike on July 19. The cascading disaster, which crashed 8.5 million systems at banks, hotels, airlines, retail stores and thousands of other businesses using Microsoft Windows, produced more than $5 billion in estimated financial losses, much of it uninsured.
PagerDuty also designs and develops gen AI models for its own purposes. One tool, called CatchMeUp, is intended to help internal incident response personnel get up to speed quickly on a major incident, an event likely to consume several days to resolve and require multiple responders. A prompt in CatchMeUp summarizes the up-to-the-minute work performed by a previous responder to help the next responder “hit the ground running,” Tejada says.

—Jennifer Tejada, CEO, PagerDuty
CEO Matt Oppenheimer at Remitly is bullish on the use of gen AI tools to generate more profitable business at the publicly traded company, which tallies 2,700 employees worldwide and $944.3 million in revenues. Remitly provides online money transfer and payments at low fees to millions of immigrants and their families in 170 countries. “Where many companies get it wrong about gen AI is they buy a tool because everyone else is buying it, without considering the customer problem to be solved,” Oppenheimer says.
A particular customer problem that Oppenheimer looked to solve was a slowdown in a money transfer, not surprising since the company supports complex international transactions in more than 100 currencies. In 2023, Remitly processed a send volume of $39.5 billion, although it does not publicly state the exact number of money transfers. “Oftentimes, 95 percent of money transfers don’t need customer support, but 5 percent do,” Oppenheimer says. In such cases, the customer typically reaches out to a specialized customer support agent who speaks the person’s language. To reduce the lengthy time involved in many calls, Remitly developed a gen AI virtual assistant tool that resolves customer issues four times faster than live agents, liberating them to assist customers with particularly challenging problems, such as a compliance issue in a specific country in the company’s disbursement network.
Another gen AI tool entails the use of a large language model (LLM) to analyze, interpret and virtually respond to customer texts and emails in 18 different languages. Other tools include a gen AI model that examines money transfer and payment data to ferret out possible fraud, as well as a tool used to increase marketing effectiveness. “It’s been incredibly helpful to our SEO efforts, ensuring we market images across the globe that resonate in each country,” Oppenheimer explains.
Down the Line
Although the gen AI budgets of these four organizations are currently geared toward productivity-enhancing tools, each CEO is intrigued by the potential of advanced gen AI use cases. Oppenheimer, for example, says he is focused on tools that will help the company become more efficient for customers and work that is less tedious and time-consuming for employees.
“I’m not opposed to the use of integrated gen AI solutions to achieve specific goals or solve business problems, but we need clear boundaries to ensure security, privacy and the risk of hallucinations,” he says. “For now, I’m not interested in use cases that don’t still have a human in the loop as a guardrail.”
Tejada’s hope is for a gen AI use case that takes 70 percent of her work off her plate, liberating the CEO “to do important strategic work and enhance the quality of my decisions and communications,” she says. “The challenge for us and most companies has been a significant lack of data scientists and PhD AI experts. There aren’t enough to go around to experiment with large language models and safe gen AI solutions.”
Campos plans to connect all the different tools and data repositories emerging at PetMeds into an integrated system guiding more insightful decisions on business goals and challenges. “It will help us grow and scale the business, reduce expenses and provide ways for people to be better at their jobs,” she says, adding that the AI transformation will require investments in change management and training. “At the end of the day, there may be fewer jobs, but they will be better jobs.”
Wako has a similar vision. “We know that gen AI will change how we deliver healthcare in the future, helping people live longer and healthier lives,” she says. “More data, assuming it is secure and accurate, means even greater opportunities for AI to make a profound difference. That’s the endgame, but we’re not there yet.”