Priya Jaiswal has spent years at the intersection of banking, business, and finance, translating market shifts into concrete strategies. Her vantage point spans portfolio management, cross-border trends, and the messy reality of transformation. In this conversation, she reframes AI not as a math problem to solve through subtraction, but as a vision challenge that asks leaders to step a little into the unknown. Like Earth set against the vast blackness — as one of the four Artemis II astronauts described — the contrast clarifies what makes humans indispensable: judgment, empathy, and enduring relationships.
When leaders cut staff in anticipation of AI, markets often applaud. How have you seen those “bold moves” play out a year later in productivity, culture, and customer outcomes? Can you share metrics or examples where short-term praise conflicted with long-term performance?
The first quarter after a headline-making percentage cut often looks tidy on an earnings call, but the next year tells a different story. I’ve watched “robot juice” plans deliver quick cost optics while cycle times lengthened and exception queues quietly ballooned. Customer complaints climbed as institutional knowledge walked out, and service recovery required rehiring contractors to re-teach playbooks that used to live in the hallways. The emotional texture inside the firm also shifted — managers reported a brittle culture, with people optimizing for survival instead of experimentation. Short-term applause muted the deeper signal: the work wasn’t redesigned, just starved.
Many firms treat the workforce as a cost to optimize rather than a source of judgment and relationships. What practical steps help leaders quantify the value of institutional knowledge? Can you share an instance where retaining a specific team preserved revenue or reduced risk?
Start by identifying “decision hotspots” where outcomes hinge on tacit judgment — complex credit overrides, wealth client triage, fraud escalations. Map those to revenue and loss-avoidance events; if you can tag even a portion of renewals or recoveries to named teams, you turn invisible capital into a trackable asset. I’ve seen a small coverage group retained during a restructure sustain key-client inflows simply because they could decode unstructured signals from long-standing relationships. Keeping them wasn’t charity — it was an insurance policy against churn and reputational damage. When leaders listened to the floor, they discovered that knowledge was already priced into customer loyalty.
When tedious work shifts to AI agents, roles like wealth managers or risk analysts can change dramatically. How should leaders redesign job scopes in the first 90 days? What specific capabilities, metrics, and guardrails help humans and agents coordinate decisions effectively?
In the first 90 days, rewrite roles around outcomes, not tasks: humans own intent, exception judgment, and relationship moments; agents handle retrieval, summarization, and monitoring. Build capabilities in prompt hygiene, chain-of-thought critique, and escalation storytelling — the ability to explain why a human stepped in. Track coordination metrics such as exception capture quality and customer satisfaction at high-emotion touchpoints. Guardrails should include clear decision rights, model input logs, and red-team routines on edge cases. The feel of the work changes: less swivel-chairing, more coaching your agent — like breathing easier after shedding busywork, yet more accountable for the final call.
New roles will emerge at the seam between human decision-making and agentic systems. Which seam roles do you expect first, and how would you define their responsibilities and performance measures? Can you walk through a pilot org chart and a day-in-the-life scenario?
Expect roles such as Agent Orchestrator, who designs workflows across tools; Judgment Wrangler, who curates human overrides into reusable patterns; and Control Room Analyst, who watches model health and customer signals. In a pilot org, place them inside a small cross-functional “mission team” embedded with a business unit and paired to risk and compliance. Success measures include override precision, time-to-insight on anomalies, and the reuse rate of decision playbooks. A day in the life starts with a morning huddle over last night’s escalations, a mid-day simulation of a tricky scenario, and an afternoon co-creation session with frontline staff to refine prompts and guardrails. By evening, the Orchestrator ships an updated workflow, and the Control Room signs off with an audit-ready trail.
Many companies default to headcount targets rather than a north star for what they want to become. What is a strong, testable vision statement leaders can use to guide AI investments? How would you translate it into a three-phase roadmap with milestones and owners?
A testable vision: “We will be the firm where every client interaction benefits from augmented judgment — faster, fairer, and more human.” Phase one proves reliability on a contained journey with clear owners in the business, technology, and risk. Phase two scales to adjacent journeys while codifying decision playbooks and training. Phase three integrates the seams into core planning and compensation, making augmented judgment routine. Each phase commits to decisions to stop or scale based on customer outcomes and model health, not vanity metrics.
Layoffs are fast; transformation is messy. How can leaders manage the uncertainty without stalling? What rituals, communication cadences, and decision rights keep teams aligned when the plan is evolving, and which early warning signals should executives watch?
Establish weekly open demos where teams show work-in-progress, not just finished slides, and invite dissent. Run biweekly decision logs so people can see why calls were made and what evidence tipped them. Give clear rights: product owns scope, risk owns guardrails, frontline owns customer truth, and data stewards own provenance. Watch for early warnings: shadow spreadsheets resurfacing, rising rework on exceptions, and customers repeating themselves across channels. The emotional cadence matters too — leaders must name the uncertainty and commit to learning in public.
In a transition, people need reskilling and psychological safety. What learning paths work best for non-technical teams partnering with AI agents? Can you share timelines, curricula, and assessment methods that proved effective, including examples of what didn’t work and why?
The most effective path blends role-based labs with live customer artifacts — redacted emails, call notes, and risk memos. Over a few months, rotate through modules on prompt design, failure mode spotting, and escalation writing, capped by peer critiques. Assess via portfolio reviews and observed simulations rather than multiple-choice tests; you want to see judgment under time pressure. What failed were “one-and-done” lectures that treated AI as a gadget — people left excited and then froze when reality got messy. Safety grows when leaders reward thoughtful overrides and celebrate well-reasoned “no” decisions.
Consider a risk function where pattern detection is automated. How should model oversight, scenario analysis, and escalation workflows change? What specific thresholds, audit trails, and human-in-the-loop checkpoints prevent blind spots and ensure accountability?
Shift oversight from static reports to continuous control rooms with scenario drills woven into the workweek. Define thresholds that trigger human review based on materiality and novelty, and log every override with rationale and data lineage. Build audit trails that capture inputs, prompts, and intermediate reasoning so you can replay decisions end-to-end. Human-in-the-loop checkpoints should sit at model changes, unusual correlations, and customer segments with higher harm potential. The aim is not to slow decisions, but to make accountability audible.
Some leaders confuse automation with transformation. How do you separate efficiency plays from genuine capability building? Can you give a step-by-step example where automation freed capacity that was then reinvested into new revenue or customer outcomes, with before-and-after metrics?
Start by asking: after automation, what can humans now do that they couldn’t before? In one case, automating data prep in wealth operations cut the “paper-shuffle” without touching client advice; leaders then reassigned time to proactive outreach and complex planning. The before state prized volume; the after state measured retained relationships and resolved edge cases. Capability shows up as new decisions made well — not just the same decisions made cheaply. If you can’t name the new muscles, you didn’t transform.
Think of the image of Earth standing out against vast blackness. How can leaders use contrast to spotlight uniquely human contributions like empathy and critical thinking? What practices make these traits visible in performance reviews, customer journeys, and product roadmaps?
Use contrast deliberately: pair agent-generated options with a human “why” that surfaces values and trade-offs. In reviews, score how teammates improved outcomes via empathy and critical thinking, anchored to real customer moments. In journeys, mark “human moments that matter” where judgment and care change the arc — a loan denial explained with dignity, a portfolio shift during a life event. In roadmaps, add acceptance criteria for human experience alongside performance. The blackness doesn’t erase the planet; it reveals its glow.
One path is a press release; the other is a plan. What are the non-negotiable elements of a credible AI transformation plan? Please outline the first 100 days, including governance, data readiness, pilot selection, KPIs, and how you’ll decide to scale or stop.
Non-negotiables include a clear north star, decision rights, data provenance, and customer-centered KPIs. In the first 100 days, stand up governance with business, risk, and technology peers; inventory data with lineage; select one pilot journey with real stakes; and define stop/scale criteria tied to customer and control outcomes. Run weekly demos, log every decision, and test failure modes before success stories. Decide to scale only when customers feel the improvement and risk signs off; stop when the plan drifts into subtraction without vision. One path is noise; the other is accountable learning.
You’ve worked across fintech and innovation circles. What partnership models between incumbents, startups, and regulators accelerate responsible AI adoption? Can you share a concrete example with incentives, risk-sharing terms, and measurable outcomes?
The most durable model is a shared-sandbox partnership with pre-agreed risk tiers and outcome-based incentives. Incumbents contribute data access and domain experts; startups bring speed and tooling; regulators observe design choices early, not late. Risk is shared via staged exposure: limited cohorts, reversible switches, and transparent audit artifacts. Outcomes are measured on customer benefit and control integrity, with each party earning more scope as those outcomes are proven. Everyone learns faster when the rules of the game are visible.
What is your forecast for AI and the future of work?
My forecast is that the next chapter looks less like a spreadsheet and more like that Artemis II window — Earth luminous against the void, its meaning sharpened by contrast. By 2025, firms that chose vision over subtraction will have cultures fluent in human-machine teaming, while those that chased percentages will be relearning what they gave away. The work will feel different: quieter busywork, louder moments of judgment, and more time where empathy is the differentiator. If leaders honor what only people can do, the future of work won’t be smaller; it will be more human, and more valuable, than we imagined.
