The Agent Will See You Now: Inside the Quiet Machinery of Revenue Operations

I was researching a piece on agentic AI in revenue operations – the usual mix of vendor claims, prediction decks, and the same three case studies everyone recycles. Somewhere in that research, I connected with Sundeep K. Bhatia, Founder of SSANZ and COO of Rehab Revenue, and asked what I thought was a throwaway question: where does agentic AI actually replace human judgment in revenue operations, and where is it just automation wearing a fancier collar?

The answer didn’t sound like a pitch. It sounded like someone who’d actually had to defend that distinction out loud, to someone skeptical, across a table. So I asked another question. Then another. Before I knew it, what began as background research had quietly assembled itself into this – five questions, five unhedged answers, and considerably more insight into the plumbing of modern sales than I expected to find on a Tuesday.

Here is that conversation, in full.

 I. The Line Between “Agentic” and “Automated Wearing a Costume”

Every gold rush produces its share of fool’s gold, and agentic AI is no exception. Ask ten vendors what makes their product “agentic,” and nine will describe something that, stripped of the marketing varnish, is a glorified if-this-then-that machine – automation in a trench coat, agentic AI drawn on its forehead in marker.

Bhatia draws the line cleanly. 

The transition to genuinely agentic infrastructure, he says, happens only “when the system can interpret a goal, evaluate context, choose among available actions, execute, observe the outcome, and adjust what it does next within defined guardrails.” 

Everything short of that, in his words, is simply “conventional automation being marketed as ‘agentic AI.'”

Where does the real handoff of judgment happen, then? In the high-frequency, rule-bound terrain of revenue operations: “lead qualification, routing, follow-up sequencing, CRM updates, pipeline monitoring, re-engagement, and identifying next-best actions.” Decisions with clean edges, made often, at a pace no human queue was built to match.

What doesn’t get handed over – what Bhatia insists stays resolutely human – is “high-consequence decisions: complex negotiations, strategic pricing exceptions, relationship management, contract terms, and situations where context extends beyond what exists in the system.”

 ❝ Our philosophy at SSANZ is simple: automate the decision volume, not the executive judgment. ❞

The agents get the assembly line. The humans keep the boardroom.

 II. The Foundation Nobody Wants to Pour First

Every builder knows the unglamorous truth: nobody wants to pay for the foundation. They want the penthouse view. But skip the foundation, and the penthouse simply sinks slower.

Ask Bhatia what has to be right before agentic AI creates value instead of noise, and the answer arrives without hesitation: “the revenue data layer.”

Here is the diagnosis, and it stings a little: “Companies often think they have an AI problem when they actually have a data architecture and process problem.” 

Inconsistent CRM stages. Lead sources attributed by guesswork. Duplicate records breeding like rabbits. Sales activity nobody logged. Departments that “differ” on when a lead becomes a customer. Integrations that quietly stop synchronizing.

Feed an AI agent that mess, Bhatia warns, and “an AI agent does not solve those problems. It simply makes decisions faster using unreliable information.”

His own architecture reads almost like a recipe, ingredients listed in strict order:

 ❝ Clean Data → Integrated Systems → Defined Revenue Logic → AI Agents → Measurement & Optimization. ❞

 ❝ Agentic AI amplifies whatever infrastructure you give it. Clean infrastructure creates leverage. Broken infrastructure creates noise at machine speed. ❞

Give an agent a tidy house, and it will keep it tidy. Give it a mess, and it will simply vacuum the mess faster.

 III. When the Machine Makes the Wrong Call, Who Answers For It?

Every new technology eventually asks its operators the same uncomfortable question: when this thing breaks something, whose name goes on the apology?

Bhatia’s answer arrives without a flicker of ambiguity:

 ❝ The organization owns it [the apology] – not the AI. ❞

His solution is a kind of traffic-light constitution for autonomous decision-making. Green – “AI can execute autonomously,” for things like “CRM updates, routine follow-ups, lead enrichment, and basic routing.” Yellow – “AI can recommend, but a human approves,” covering “significant pricing changes, strategic-account prioritization, and unusual concessions.” Red – “Human decision required,” reserved for “contractual commitments, major pricing exceptions, sensitive customer situations, and material financial decisions.”

Every agent, in this framework, answers to somebody: “an executive owner, operating owner, defined permissions, escalation path, audit trail, and measurable performance thresholds.”

The sharper question, Bhatia argues, was never “who do we blame when the AI gets something wrong.” It’s this:

 ❝ Who authorized the agent to make that class of decision, under what constraints, and what controls existed to catch an exception? ❞

Blame is easy and retroactive. Governance is harder, and it comes first.

 IV. The Timeline Nobody Wants to Hear, and the Red Flag That Should Make You Run

If there is a universal law of vendor sales pitches, it is this: someone, somewhere, is promising a 30% revenue lift in 30 days. Bhatia treats that promise the way a structural engineer treats a building with no foundation survey – politely, but with enormous suspicion.

He refuses “a universal number,” because ROI “depends heavily on the maturity of the company’s existing revenue infrastructure.” 

For a focused deployment against reasonably clean CRM data, he says “30–90 days should be enough to establish whether the implementation is producing measurable operational value” – not transformational ROI, but early tells: “reduced response times, greater rep capacity, better CRM compliance, improved lead coverage, fewer dropped opportunities, and movement in conversion or sales velocity.”

Cost, similarly, resists a single sticker price. “A focused implementation might be in the tens of thousands of dollars,” while enterprise deployments – “multiple systems, extensive data remediation, security requirements, governance, and custom integrations” – “can move well beyond that.”

The red flag, when it waves, waves unmistakably:

 ❝ “Deploy our agents and increase revenue 30% in 30 days” without first understanding the company’s data, pipeline, sales cycle, baseline conversion metrics, and technology architecture. ❞

His own sequence runs the opposite direction – proof requested before the receipts, not after:

 ❝ Baseline → Business Case → Deployment → Measurement → Optimization → Scale. ❞

“If the vendor cannot tell the CFO what metric is supposed to move, what the baseline is, and how the improvement will be attributed to the AI infrastructure,” Bhatia says, “then the ROI claim is not sufficiently rigorous.”

 V. Two Years From Now: Fewer Hands, or Different Hands?

The question every automation eventually forces onto the table, dressed up in a dozen euphemisms: does this mean fewer jobs?

Bhatia resists the easy binary:

 ❝ We do not think the primary outcome is simply fewer people. We think it is far greater revenue capacity per person. ❞

In his picture of a rev-ops team two years hence, AI agents absorb the repetitive operational layer – “data entry, enrichment, routing, monitoring, follow-up orchestration, reporting, pipeline surveillance, forecasting support, and identifying anomalies.” Humans, meanwhile, “move upward into revenue architecture, strategy, experimentation, governance, customer intelligence, complex decision-making, and agent orchestration.”

Where ten people once operated a process, Bhatia suggests, “5–7 people” may increasingly design and govern a system doing “the operational work that previously required 10–15 people.” Not fewer jobs so much as fewer hands per unit of output.

And the real scoreboard, he insists, was never headcount to begin with. It’s “revenue per employee, pipeline per seller, speed-to-lead, sales-cycle velocity, cost to acquire revenue, forecast accuracy, and operating leverage.”

His closing thought, and perhaps the tidiest summary of the whole conversation:

Agentic AI is not about replacing the revenue organization. It is about turning the revenue organization into infrastructure that can think, act, learn, and scale – with humans governing the decisions that matter most.