Teams keep investing in AI, but a chief revenue officer (CRO) still waits weeks to learn why win rates slipped, what the best closers do differently, or where the quarter will land. The tools are in place, yet the answers arrive in pieces that a person has to assemble. This guide is for revenue operators and leaders who have already bought some sales AI and want complete answers instead. It lays out a decision order for using AI strategically: fix the operating model, decide which judgments stay with people, automate the reviews that only recompile known data, give AI the whole revenue picture, and connect answers to action.
Revenue data lives in fragments. The CRM (customer relationship management system), email, call recordings, and the data warehouse each require a different language to access, and no single AI sees all of them. Most tools extract a partial insight from one fragment and then leave the user to put it to work.
The problem shows up in questions every CRO asks. Questions like: Why are win rates dropping in EMEA (Europe, the Middle East, and Africa), and what needs to change? What are our closers doing differently, and how do we scale it? Why do we keep losing to a particular competitor? Where will we land this quarter, and what are the headwinds?
Answering any of them by hand means pulling data across several systems, which takes weeks and specialized expertise. When an answer does arrive, it often lands in a dashboard, and nobody changes a call script, a coaching plan, or a deal review because of it. When no one acts on an insight, the answer fails. Deciding in order prevents that.
The first decision is not about a tool. Quoc Tran, Vice President of Revenue Operations at Replit, was one of the first hires under the CRO and built a go-to-market motion from scratch, with no processes, rules of engagement, or systems in place. On Revenue Mavericks he described what happens when an AI layer goes on without processes.
When you're layering on an AI layer without processes, you get a bunch of garbage. You get a bunch of noise. Just like fuel on the fire. Building and formalizing process is just the first step.
Lesley Young, a go-to-market leader who has worked at companies including Box, Informatica, and Twilio, made the same argument from the leadership side.
If those insights are built off of an inconsistent operating model, they're not going to enable you to take advantage of what AI has to offer to go to market, at least that's my very strong opinion.
Her fix is an operating cadence. She compares it to a college syllabus for a semester, and she ties it to a 13-week frame. People know what the team reviews and when, so they can plan their work against it. That consistency, in her telling, includes the forecast, the QBRs, and how the team talks about a forecast.
She also took advice from David Pidwell, an Informatica board member, in a leadership program: boards care about predictability, growth, and efficiency. That is a theme that connects to building a forecast your CFO will trust.
In practice, this means writing down the forecasting process before you automate it. If the stages, definitions, and review rhythm are inconsistent, an AI layer is likely to carry that inconsistency forward, which is the risk both guests describe.
The second decision is which judgments remain with people. Quoc Tran gave a concrete example. One of his field engineers received a cold outbound email that was not personalized, and the team debated how to bring the human touch back.
You can tell that it was AI slop.
He forwarded the email to his marketing and sales teams as an example of what not to do. He described AI as a useful research tool and a way to pull bullet points into an email.
You can't replace the human touch of, you know, reviewing it.
Coaching is the other place where judgment matters. Lesley Young described leading with data in deal reviews and skills coaching, because emotion can distort what a manager sees. How do you take the emotion out, but provide people with feedback that they can activate?
Good AI sales coaching starts from evidence a rep can verify, and conversation intelligence can supply that evidence from real calls.
She also argued that AI should take over the non-essential work that keeps people from using their reasoning. That frees sellers to spend more time with customers, understanding the problem and the consequences of leaving it unsolved. The human part of the job gets larger when the administrative part shrinks.
For a CRO, the practical step is a short written list. List the judgments that stay with people (outreach tone, deal strategy, coaching) and the work AI takes.
Once the process is stable and the human judgments are named, the third decision is which reviews to automate. Quoc Tran's test is whether a review only recompiles data the organization already has. The ability to do a forecast that is a lot more math than art.
If you look at opportunity management and board reviews and QBRs and MBRs, all of that should be automated.
Here QBRs are quarterly business reviews and MBRs are monthly business reviews.
Analysts often assemble that work by hand. They pull numbers from the CRM, rebuild slides, and reconcile fields before anyone discusses what the numbers mean. Machine forecasting and automation can reduce the time spent on that assembly.
Automation changes what RevOps is for: less compiling, more questioning. Asking the right questions at the right time.
In his view that includes understanding where sellers' motivations lie, since some will try to get around the systems to make a buck. It also includes helping executives anticipate issues, such as an international expansion.
Use this frame when you define RevOps responsibilities for the next few years. The team spends less time producing the review and more time deciding what the review should ask and what the data does not yet explain.
Automating reviews only helps if the data behind them is complete. A model that sees the CRM but not the emails and call transcripts can report stage and amount but cannot explain why the deal slipped. The reverse is also true, because a tool that reads calls but not CRM history cannot tell you whether a pattern holds across a quarter.
Structure the data as carefully as you choose the model. Structured fields, such as stage, amount, and close date, have to sit alongside unstructured material, such as email threads and call recordings, in one connected view. A question about win rates in one region should be answered from every system that touches those deals, not the one that was easiest to query.
One response is to assemble this yourself. A data lake can hold all of the records, but it takes major time and money to build, per Terret's positioning, and it still does not connect answers to execution. The question of whether to build or buy deserves a clear look at what your team will have to maintain after launch.
When you evaluate any revenue AI, ask what data it reasons over. If it reasons over only a fragment, expect only a partial answer.
The last decision is how AI work gets divided. Complete answers only matter if something turns them into changed behavior, and that takes two different jobs. One is designing the go-to-market system, which covers the sales process, the playbooks that describe how top closers win, and the approach to competitive losses. The other is executing inside that system by coaching reps, scoring deals, and generating forecasts. Agents can be pointed at the execution side first, and the use cases for AI revenue agents show where that work tends to pay off.
We put design first, following Quoc Tran's process-first point: execution depends on the process underneath it. Skipping design and deploying agents directly risks automating whatever habits already exist, which echoes his view that building and formalizing process is the first step.
The link between design and execution is its own discipline. Revenue orchestration connects signals, decisions, and actions across systems so that each answer ends in a next step. Once several agents touch the same accounts, coordinating AI agents so they do not collide can become a task of its own, and it is worth planning before you scale.
We built Terret Nexus around an answer-to-action approach that follows this decision order. It works in four steps: ask your question, get answers, automatically operationalize them, and activate them at critical moments.
A leader asks a question, such as why win rates are dropping in EMEA. Terret Nexus reasons across what we call the Revenue Graph, a connected view of structured and unstructured data from every revenue system. Our AI Architects analyze your revenue data across that complete picture and design the go-to-market system around it, including sales processes, closer playbooks, and competitive strategies. Our AI Agents then execute what the Architects design by deploying workflows, coaching reps, scoring deals, and generating forecasts. Those actions reach people at critical moments, so the rep sees the next move in the flow of work.
Your sales team stays the driver throughout. The Architects and Agents work to give your people the best performance system possible, and they do not replace the judgment this guide argues should stay with them. Each deal adds signal, so the Architects and Agents improve, and your team can win more.
If you want to test the approach on your own data, we can talk through what that would look like for your team.
Start with the operating model. That means agreed stage definitions, a consistent review rhythm, and a written forecasting process. Lesley Young and Quoc Tran both argue that AI built on top of an inconsistent model gives you noise instead of something a team can act on.
Keep the judgments where trust and context matter, such as the tone of outreach, deal strategy, and coaching conversations. AI can support research and drafting and surface evidence, but a person should review what goes to a customer and decide how to coach a rep. A short written list of what stays human and what AI takes makes this easy to enforce.
Automate the reviews that only recompile data the organization already has, such as opportunity reports, board packets, and monthly or quarterly business reviews. Deciding what the numbers mean should stay with your team. If you want to see where automation can reach into scoring, deal scoring and forecasting agents are a reasonable place to look.
A dashboard reports an insight. Changing a call script, coaching plan, or deal review takes a person acting on it. Someone still has to decide what to do and then carry it out. The idea behind revenue action orchestration is to connect the answer to an action in the flow of work.
Not necessarily. In our positioning, a data lake can store the records, but it takes major time and money and still does not connect answers to execution. List the pipelines and schemas your team would maintain after launch.
Choose your measures before rollout, and tie them to the questions you could not answer quickly before. Look at how long it takes to get a trustworthy answer on win rates or the quarter, and whether coaching and deal reviews change. Sales productivity is one baseline you can compare results against.
No. RevOps spends its time on different work. Once the compiling is automated, RevOps spends more time asking the right questions at the right time, understanding how sellers use the systems, and helping executives plan ahead. The team stops producing the review and decides what it should ask.
Test this decision order on your own deals. We can show how Terret Nexus answers a real question across your systems, how our AI Agents turn that answer into action, and what it would hand to your sales team in the flow of work.
If you would like to see that on your data, Request a demo.