
Companies spend millions on systems designed to manage everything from pricing and sales approvals to contracts and billing. Yet even with sophisticated technology in place, seemingly minor discrepancies between these systems can quietly cost businesses significant revenue. As artificial intelligence becomes more deeply embedded in enterprise operations, the challenge is figuring out where it can deliver measurable financial results.
Eshaan Jain has spent nearly 15 years working at the intersection of enterprise technology, product management, and business operations. His career includes experience at Amazon, PwC, and Accenture, where he has worked on complex enterprise systems, commercial processes, and AI-driven solutions. His expertise spans configure-price-quote (CPQ) platforms, contract lifecycle management (CLM), machine learning, and AI governance. Among his projects is work on a machine learning system designed to extract critical commercial clauses from transportation contracts representing billions of dollars in business.
In this interview with IntelligentHQ, Jain discusses why revenue leakage remains a persistent problem for large enterprises, how AI can identify costly inconsistencies across commercial systems, and where human judgment remains essential. He also shares his perspective on the economics of enterprise AI, the limitations of autonomous decision-making, and why companies should measure AI investments by the revenue they protect rather than the number of tasks they automate.
Revenue operations sits at the intersection of sales, finance, legal, pricing, and product. Where do you see AI creating the most value when those functions need to make decisions together rather than operating in separate systems?
The highest value I see is at the handoff points between sales, finance, and legal, where a pricing exception one team approved runs into a contract term another team changed later. In the CPQ and CLM programs I’ve worked on, that boundary is where almost all the real friction lives. It’s rarely a bad discount rule or a slow contract review. It’s that the pricing exception a finance approver signed off on last quarter sits in one system, the legal redline that changed the payment terms sits in another, and nobody reconciles the two until an auditor or a customer does.
AI’s best use in revenue operations right now is reading across those systems to catch discrepancies among the approved price, the contracted term, and the billed amount before they turn into a dispute or a write-off. In practice, that means matching quote, contract, and billing records on the same account and product IDs and flagging any field that doesn’t reconcile, a discount, a payment term, or a product configuration, before month-end close, not after. That’s the job that actually reduces revenue leakage. It’s also a lot less exciting than most AI-in-revenue-ops pitches make it sound.
Across major enterprise environments, I’ve seen custom clause-extraction systems read negotiated contract terms across multi-billion-dollar portfolios with high accuracy, so finance and legal aren’t relying on someone’s memory of what a contract said. Revenue operations has the same gap today, just spread across CRM, CPQ, CLM, and billing instead of one contract repository. The pattern holds regardless of scale: the system of record and what someone actually negotiated, disagree, and nobody notices until the money is gone.
Many enterprises have accumulated years of pricing rules, contract policies, approval thresholds, and commercial exceptions. How can organizations turn that institutional knowledge into something AI systems can actually use without oversimplifying the business?
The mistake I see most often is treating the exception as noise to clean up. It usually isn’t. A five-year-old approval threshold that applies only to one product line during a specific renewal window exists because someone negotiated it to solve a real problem. Flatten it into a generic rule to make the model simpler, and you’ve reintroduced the exact problem it was created to prevent.
What works is treating pricing rules, approval thresholds, and contractual exceptions as structured, versioned data with provenance: who approved it, when, for which product or customer segment, and why. The CPQ implementations that get this right give every discount and approval exception that history, rather than leaving it in a static rulebook someone wrote once and never revisited.
The organizations that get this wrong usually try to solve it with a single master rules engine meant to cover every case. It never does. The exceptions that matter most, the ones protecting a strategic account or a regulatory requirement, are exactly the ones that get simplified away first, because they’re the hardest to encode. Keep them explicit and versioned instead of collapsing them into an average.
That traceability is what separates an AI system that can explain why a deal needs finance sign-off from one that just says no. I don’t think explainability here is a compliance checkbox. It’s what keeps a revenue operations team trusting the system enough to use it instead of quietly routing around it, which is what happens to every tool people don’t trust.
Revenue leakage is often caused by small inconsistencies rather than one major failure, an incorrect discount, a missed renewal, an outdated price, or a contractual obligation that was never enforced. How can AI help companies identify those patterns earlier?
Leakage almost never comes from one big failure. It’s a discount approved for a pilot that never expired. It’s a renewal date that lived in a spreadsheet instead of the CLM. It’s a price list that was updated in the CPQ but not in billing. Each one is small. Add them up across a large base of accounts, and they become real revenue that never gets collected. It rarely shows up as a single line item anyone can point to.
A classic failure in complex quote-to-cash setups is a bug in mid-contract amendments, in co-termination and asset-based ordering. A customer adds lines partway through their term. Because of an edge case in how the system prorates and recalculates co-term dates, a discount that should have expired carries over to the new term. On its own, that’s a rounding-level bug. Across a large base of active contracts, it’s a slow, steady drain that nobody sees until someone compares what was billed to what was actually contracted.
The earliest and most reliable signal is reconciliation, not prediction. Compare what was quoted, what was contracted, and what was billed, and flag the gaps as they open instead of finding them at quarter close. I’ve been working on an architecture built around that idea: detect and explain revenue leakage at the transaction level, instead of discovering it in an audit six months later.
Honestly, the hard part is getting quote, contract, and billing data into a form where you can compare them at all. Most enterprises keep each in a different system, with a different customer ID and a different deal ID for what is, on paper, the same transaction.
How do you think AI will change the role of the traditional deal desk? Could systems eventually evaluate pricing, terms, risk, and approval requirements in real time while a salesperson is constructing a deal?
Some of that is already happening. It just isn’t called AI. Today’s CPQ guided-selling tools use rules and automated integrations to check pricing, discount eligibility, and approval routing in real time while a rep builds a quote. The deal is checked against current rules before it ever reaches a deal desk analyst. That part scales well, because pricing and threshold checks are predictable once the rules behind them are set up properly.
Where I’d push back is on the idea that the deal desk becomes fully automated. Think of a non-standard deal, a strategic account asking for something unusual, or an exception that technically breaks policy but makes business sense. None of that can be scored in real time. It’s a negotiation, and the context doesn’t live in any system.
I’d rather see AI take over the rule-based checks that deal desk analysts do by hand today, slowly, and leave the judgment calls to the people who will own the relationship if the deal goes wrong. If you sell the rest as full automation, you end up with a chatbot standing in for a risk function. I’ve watched that pitch fail in real rooms.
The deals that work best still have a named owner for the judgment call. It could be a solution architect or a RevOps lead who can explain why a specific exception was approved. That accountability doesn’t automate well, even when the checks around it do.
Companies often invest heavily in CRM, CPQ, CLM, ERP, and billing platforms but still struggle with fragmented commercial data. Before adding AI, what does an enterprise need to do to create a reliable revenue-data foundation across those systems?
Before adding any AI, you need one definition of a customer, one definition of a product, and one definition of a contract that every system agrees on. That sounds basic. It’s also the part almost nobody wants to pay for, because it isn’t a project you can demo. It means data migration, matching records to the same entity, and cleanup across systems that were bought at different times by different teams for different reasons.
In CPQ and CLM implementations, most of the weak points I’ve seen aren’t in the AI or the automation logic. The product catalog in CPQ doesn’t line up with what the ERP calls the same SKU. Or a contract amendment exists in the CLM but never made it into billing. Any AI model built on top inherits that mismatch and gives confident, wrong answers, which is worse than no answer at all.
This is where most AI transformation programs I see actually fail. They skip the messy-data problem: contract clauses, product catalog cleanup, and the quality of quote-to-cash data. They add AI on top anyway. The result is agentic AI as theater: a demo that works on hand-picked examples and falls apart on the account with three amendments and a manually typed exception.
The fix is the same everywhere: clean IDs, one source of truth for pricing and product data, and a real audit trail for contract changes. That is what makes the AI layer trustworthy later. Skip it, and you’re automating the fragmentation instead of fixing it.
When should an enterprise build its own AI capability around revenue operations rather than relying on functionality embedded in commercial software platforms? What factors should leaders consider when making that decision?
My rule of thumb: build when your data is your edge, buy when the workflow is a commodity. Approval routing, notification logic, and basic guided selling are mature features that most CPQ and CLM platforms already do well. There’s no reason to rebuild them.
Some things are different: interpreting your own contract language, spotting discount-exception patterns unique to your business, and detecting leakage in how your revenue actually breaks.
Custom clause extraction for specialized agreements is a good example. A generic, off-the-shelf tool rarely reaches the accuracy needed on large, complex sets of contracts. The value came from how we modeled our specific clause types and exceptions. The underlying model mattered less than the taxonomy we built on top of it.
There’s a third factor: what it costs to keep running whatever you build. A custom model or pipeline isn’t a one-time build. It’s an ongoing commitment to inference costs, retraining, and monitoring. I spend a fair amount of my own research time on token economics and cost per inference, because I think cost, not model capability, will decide which AI programs survive the next couple of years. A team that builds a differentiated model but hasn’t priced out what it costs to run at scale hasn’t made a build decision. They’ve taken on a maintenance commitment they haven’t budgeted for yet.
That’s the real test in a build-versus-buy decision. It’s considering whether you own something proprietary worth defending and can afford to keep running it. Everything else is a vendor’s job, not yours.
If a CFO or chief revenue officer wanted to know whether an AI-enabled revenue operations program was actually working, what metrics would you recommend tracking beyond traditional measures such as time saved or number of tasks automated?
Time saved and tasks automated measure activity, not results. A CFO or chief revenue officer should ask what changed in the money. How much revenue was protected from leakage that would otherwise have gone unnoticed? What is the dollar value of pricing or contract exceptions caught before they became write-offs? How much did unbilled or underbilled revenue drop, where it was tied to expired terms or missed renewals?
I’d also want the AI’s error rate measured against a human baseline, not just how fast the system runs. That accuracy number needs a real benchmark: what a manual review catches versus what the automated system catches. On the CPQ and CLM side, a big cut in contract creation time only matters if deal accuracy stays steady. So cycle time should be paired with a first-time-right rate: the share of contracts that go through without a later correction, not just how quickly they were created.
One more thing I’d track is an audit trail of every exception the AI flags: what was flagged, who reviewed it, and what they decided. I’ve been developing a framework around this question: are an autonomous system’s decisions explainable enough that a human can meaningfully sign off on them? A model being right most of the time isn’t the same as someone being able to answer for it when it’s wrong. Without that audit trail, you can’t tell whether the system is catching real problems or just generating noise that someone rubber-stamps to clear their queue.
If a program can only show hours saved and can’t show what it protected or where it reduced risk, I’d say it hasn’t proven its case yet.
Disclaimer: The views expressed here are solely my own and are shared in a personal capacity. They do not represent the official position of any employer or client, past or present.

Ricardo Tomaz is VP of Finance & Operations at Ztudium, the 4IR technology company behind Businessabc, Citiesabc, and IntelligentHQ. He owns the group's finance and operations function, spanning financial strategy, budgeting, reporting, and multi-entity operations, and works directly with leadership to align commercial goals with data-driven insight.
Working at the core of a technology group operating across AI and blockchain, Ricardo writes on the practical intersection of finance and emerging technology: how automation and digital transformation are reshaping business operations, what data-driven decision-making looks like in practice, and the questions of governance and ethics that come with AI-led systems.
