As we reflect on Season One of Pioneers and gear up for Season Two, we found ourselves inundated with tons of provoking questions about AI and how it’s impacting highly regulated industries (finance, banking, healthcare, and insurance).
In this special episode of Pioneers, Ankur Patel sits down with our producer, Jake Hurwitz, to recap and answer the top 8 most common and powerful questions that we’ve received over the last few months from our listeners.
Check out the full episode here:
Q: In general, what are a lot of people getting wrong about AI nowadays?
Ankur Patel: The major pitfall with AI is believing it can magically solve every business problem. Visionaries paint a compelling but unrealistic picture of its omnipotent potential. Yet pragmatic leaders need impact today - not years down the line. So reset discussions around AI's current, real-world capabilities instead.
For example, machine learning currently excels at narrow, rules-based automation versus broad human judgement. So target repetitive tasks rather than creative work. Conversational AI can streamline customer service inquiries but still falls short for complex complaint resolution. So deploy chatbots for common questions and route the rest to human reps. Computer vision has mastered categorization but not contextual understanding. So it can tag images or documents but won’t interpret their subtleties.
This dichotomy between hype and reality sets up failed experiments which breed skepticism. But the issue lies in poor scoping, not shortcomings of AI itself when applied appropriately.
The advice then is evaluate your problem set, honestly assess limitations of today’s AI models, carefully scope pilots to those guardrails, and demonstrate incremental gains as proof for more ambitious expansion. This pragmatic, incremental integration counters disillusionment from trying to leap before you can crawl. Be a realist despite the surrounding unrealistic optimism.
Q: At the enterprise level, should companies buy new AI technologies or build them in-house?
Ankur Patel: There's confusion around whether companies should buy new AI or build it in-house. There's no perfect answer for every org - it depends. Companies like Google and Amazon build everything in-house. ML is a core competency woven into their DNA for decades. Their existing expertise enables them to build effectively.
But many Fortune 500s lack that ML competency despite strengths in domains like healthcare or construction. For them, there's a choice:
- Invest to build ML in-house as a new core capability, knowing it may take years to do properly with enough resources.
- Or, buy from an AI specialist for faster ROI, benefiting from their existing ML expertise to enable more immediate impact.
These aren't mutually exclusive long-term decisions though. Companies can buy first for quick returns, then build in-house later either in parallel or sequentially as they mature.
Many companies get stuck deciding when they should just commit to an option, and then refine as they go. There's no need to overanalyze when the payoff comes from taking action, not from perfect theoretical planning.
So in summary - companies like Google with existing ML strengths should build in-house because it already aligns with their competencies. But companies new to ML can realize quicker benefits buying first from specialists, then optionally developing custom capabilities later as comfort with AI grows.
Q: What's different when it comes to AI in the enterprise world versus the consumer world?
Ankur Patel: Consumer AI can be playful - generating creative images or audio just for fun without real stakes if it fails. But enterprise AI must drive proven business value and be highly reliable. AI directly supporting mission-critical business decisions like mortgage approvals can't afford to sometimes work and sometimes not.
So for browsing headshot filters, synthetic media quality doesn't matter much. The consumer just moves on if it’s imperfect. But if an AI mortgage model is inaccurate, it creates massive financial risk.
Enterprise AI adoption requires:
- Reliability - it must perform consistently without failure or degradation
- Measurable value - clear, tangible impacts to metrics like revenue or ops efficiency
- Trust - models must be transparent and expectations set properly on where humans still need to be involved
So while consumers enjoy AI as futuristic enhancements, enterprises must evaluate AI against grounded, operational, and dollars-and-cents metrics before deploying to protect against very real downside risks. The threshold for roll-out is much higher in business contexts.
Q: How should I think about where to actually apply AI within my organization?
Ankur Patel: The most promising areas to apply AI in organizations today are:
1. Automating high-volume middle and back office tasks:
AI exceeds at repetitive data entry, verification, and processing - freeing up human workers to handle higher judgement roles. Target manual workflows with 5-10 people - enough scale for automation to reduce headcount up to 80%.
Ensure you can document processes formally first. If humans need detailed protocols to reliably complete a task, AI does too. Unstructured workflows with many exceptions and special cases challenge automation.
2. Conversational self-service:
Chatbots accessing data and documents enable employees to self-serve information much faster than waiting on analysts and research teams. Providing quick access to knowledge helps workers make decisions independently.
3. Coding assistance:
AI coding tools boost engineering velocity - helping developers write, review, test, and release code more efficiently. This accelerates innovation and product improvement.
When assessing where to start, consider both potential value creation from improving the work itself as well as the trickle down benefits from freeing up human talent to redirect their energy towards more complex and strategic efforts. AI should enhance the capability of an organization holistically - it’s not just about the specific function being automated in isolation.
Q: Should I use AI to reduce headcount and increase margins, or to enhance human productivity of my existing workforce? What is the right approach?
Ankur Patel: AI should primarily empower your existing workforce through augmentation versus outright replacement. The goal is to amplify human productivity on high-judgement tasks by automating lower-value repetitive work. This benefits approximately 70% of employees by freeing up their capacity to contribute more strategic value.
For example, underwriters and claims professionals remain extremely difficult roles to hire and train for. So rather than reducing headcount, AI alleviates talent shortages allowing you to handle growing volume without expanding teams.
Workers with deep expertise thrive when leveraging AI assistance whereas new hires struggle meeting quality bars quickly. So AI preserves institutional knowledge while lessening recruitment and onboarding burdens.
Now, the remaining 30% in more routine functions may see their roles eventually phased out or reduced through automation. But if net productivity rises from both refocusing your top talent and automating the long tail, your organization’s capabilities grow holistically.
Just like the calculator augmented human math skills, industrial age roles will transform but overall output should dramatically expand thanks to this hybrid human-AI approach.
The change management required still warrants sensitivity and support for impacted teams. But reframing AI as a talent multiplier rather than just an expensive redundancy mechanism realizes the greatest benefits.
The goal should be elevated organizational potential. AI lowers costs by needing fewer low-skill hires over time. But the larger win comes from unleashing Constraints on your most strategic people assets to push performance frontiers.
Q: Healthcare involves many manual processes like pre-authorizations and claims filings to determine procedure coverage. How can AI help automate and enhance these workflows? Where should we start and what benefits could we expect?
Ankur Patel: Healthcare workflows like pre-authorizations and claims processing currently involve high friction - requiring countless calls and manual reviews to determine coverage eligibility. AI conversational interfaces can slash the legwork needed to surface answers.
For example, instead of reading a complex 50-page insurance policy to assess what procedures or medications apply, patients and staff could simply ask a natural language AI bot powered by that documentation. The bot would instantly parse the policy details and respond with relevant coverage specifics - is this procedure covered, what’s the out-of-pocket cost, etc. This self-service model distributes information faster without playing “phone tag middleman”.
Further use cases include AI claims assessment to accelerate reimbursements. Today, overseas teams manually evaluate forms to issue payments - taking weeks. But AI can instantly process documents and validate billing for instant reimbursement.
Healthcare is rapidly pursuing these workflow innovations because they unlock tremendous efficiency gains:
- Patient/staff frustration drops dramatically thanks to improved self-service
- Call volume decreases as policies become queryable on demand
- Claims lag time compresses from weeks to days or hours
- Billing costs may decrease given less staff overhead needed
So in summary - start applying AI conversational interfaces on top of complex backend systems that currently require heavy human invocation. Whether insurance documents or billing protocols or drug guidelines, surface the logic through AI bots.
This unblocks huge administrative bottlenecks, accelerating revenue cycle management. The model pays dividends across patients, practices, providers, and insurance alike by slashing wasted time.
Q: How is AI impacting key insurance areas like underwriting and claims processing currently? How does this differ from general finance applications?
Ankur Patel: Insurance is highly fragmented with countless carriers, policy types, brokers, and backend processors. Much of this administration relies on offshore business process outsourcing to manage repetitive tasks like underwriting and claims.
While inexpensive, these manual workflows are hugely inefficient - slow, inaccurate, and challenging to scale. AI promises tremendous cost and quality improvements by automating the drudgery. For instance, AI can extract and validate info from applications and forms as accurately as an experienced human. This cuts overhead and boosts lagging processing times plagued by labor shortages abroad.
Beyond cost savings, AI also unlocks revenue opportunities around customer experience and retention. Applicants today struggle to comprehend policy details to determine optimal coverage, deductibles, renewals, etc buried in dense documents. But conversational AI assistants can surface answers directly from those materials in seconds. So instead of reading a 50-page policy, you simply ask the AI bot questions in plain language to get tailored recommendations.
Delivering this level of quick yet reliable support attracts and retains more customers. If GEICO handles your claim seamlessly via AI, you’ll likely remain loyal versus competitors with archaic experiences.
So in summary, AI drives two central benefits:
- Middle and back-office automation to slash administrative costs and speed throughput. This requires applying AI to repetitive, rules-based workflows.
- Customer-facing self-service to provide personalized policy guidance. This requires natural language AI interfaces analyzing complex documents to address common questions.
Combined, carriers gain operating leverage via automation and top-line growth through sticky premium experiences. AI unlocks a potent flywheel - one funding the other’s expansion.
And this dual productivity and satisfaction upside should outweigh modest near term implementation costs. Over time, AI will likely become table stakes to remain competitive across evolving consumer expectations.
Q: What one or two key pieces of advice would you have for executives in manual industries primed for disruption by AI? How should they approach evaluating and adopting AI to really see benefits? What should they be aware of?
My top advice to manual industry leaders pursuing AI is to set realistic expectations and move incrementally based on tangible impact. First, rigorously identify your highest value AI opportunities through current state analysis and partner input. Rank order use cases and build an evaluation roadmap. Then pilot short 2-3 month projects in your top priority domains, focused on productivity and efficiency gains over cutting-edge innovation. Lock down ROI rapidly even if the initial scope is narrow.
For example, inject AI into a sub-process like data extraction from applications rather than end-to-end claims or underwriting transformation. Quick wins, even if small, build confidence and justification for further investment. But isolated impact still proves the model for wider rollout.
Conversely, don’t boil the ocean pursuing multi-year, ultra-futuristic AI without intermediate waypoints. Over-engineering flashy capabilities that lack grounded utility today wastes time and money. And if stakeholders grow impatient from prolonged development without visible returns, entire initiatives risk abandonment regardless of theoretical potential.
So in summary:
- Rigorously identify and rank your opportunity landscape
- Pilot high-potential areas rapidly with a tight scope
- Realize quick ROI as proof for further buy-in
Following this crawl, walk, run approach ensures you balance both practical results and ambition based on your organization’s appetite and aptitude for change.