Key takeaways

  • AI-written outbound fails when no person owns the final line. Buyers respond to messages that show real understanding of their business and role, and a human review step is what supplies that.
  • AI layered on a team with no defined process produces noise. A stated objective and a documented process have to exist before any model can sort signal from garbage.
  • Insights built on an inconsistent operating rhythm tend not to hold up. A steady weekly and quarterly operating rhythm, plus complete data behind it, comes before AI.
  • Match the tool to the job. MEDDIC qualifies deals and does not forecast. Start coaching from data, not opinion.
  • The order is objective, process, operating rhythm, and complete data, then AI, with the sales team in the driver's seat.

AI tools are already in your stack. Outbound drafts flow out of your sequencer, dashboards multiply across your customer relationship management (CRM) system and call platform, and the results are mixed. Some reps send more email without booking more meetings, and some managers have more charts without a clearer view of the forecast. Usually the problem is what the tool was placed on top of, not the tool. This article covers where AI breaks in a sales org, why it breaks, and what to put in place first so your team stays in the driver's seat.

Failure one: AI-written outbound with no human line

The first failure is the easiest to see because it lands in everyone's inbox. Using AI in sales works well for research and first drafts, and some teams respond by letting it write and send the whole message.

Quoc Tran, Vice President of Revenue Operations at Replit, described the result on Revenue Mavericks. One of Replit's field engineers had shared a cold outbound email he received that was not personalized.

You can tell that it was AI slop.

Quoc Tran, Revenue Mavericks

He forwarded an AI-generated email to his marketing and sales team as an example of what not to do.

The team should use it for research, and he described how one person pulled four bullet points about a prospect and turned them into an email quickly. The line he draws is at review.

You can't replace the human touch of reviewing it.

Quoc Tran, Revenue Mavericks

A person still has to decide whether a joke lands or reads as cheesy.

Tran also receives a lot of cold outbound himself, and he said he responds to messages from people who understand where he is coming from, his business, and his role. The message does not have to be perfect, but you can tell when someone took the time to understand the reader.

When AI drafts and sends without a human line, your buyers tune out the same openers and filler. The fix is a defined review step in which a rep owns what goes out under their name.

Failure two: layering AI on a team with no process

The second failure: a team adds AI to a process nobody wrote down, and the AI inherits the confusion.

When he joined Replit, the go-to-market (GTM) team had no processes, no rules of engagement, and no systems, so the team built them from scratch.

When you're layering on an AI layer without processes, you get a bunch of garbage. You get a bunch of noise. You have to kind of sift through the signals.

Quoc Tran, Revenue Mavericks

The cost includes more than bad output. Without a process, you are constantly fighting over whether the data is right and which fields are missing. Our reading of why is that a model has little to go on when stages are undefined, handoffs have no owner, and no one has said which fields reps are expected to fill in.

Once processes are in place and AI is built on top. That's just like fuel on the fire.

Quoc Tran, Revenue Mavericks

His team can pound through inbound, and they built internal tools in hours once the process was in place. In our view, a team with a defined process also has a clearer picture of what a good result looks like, which makes AI output easier to judge.

This is work that revenue operations (RevOps) owns: defining stages, rules of engagement, and handoffs, and that definition is the input AI needs. If those pieces live in people's heads, write them down before you buy another tool.

Failure three: insights built on an inconsistent operating rhythm

The third failure sits one level up. A team can have a process on paper and still run it differently every week, and then the insights built on top of it mean little.

Lesley Young, a sales leader who has been doing a lot of fractional work recently, said she treats the insights themselves as table stakes. The harder part is the foundation under them. If those insights are built off of an inconsistent operating model, they will not let a team take advantage of what AI has to offer in go-to-market.

Lesley Young, Revenue Mavericks

She called that her very strong opinion.

Young compares an operating rhythm to a syllabus, which tells you on day one what the next 13 weeks hold. In week one the team looks at slipping deals, and by mid-month two the forecast is locked. She was firm that the operating rhythm, including quarterly business reviews (QBRs), has to hold steady.

It needs to be consistent... that includes the forecast, that includes the QBRs, that includes how we talk about a forecast.

Lesley Young, Revenue Mavericks

A consistent operating rhythm is also what makes a sales forecasting process trustworthy. When every manager defines commit differently, or forecast calls change format from month to month, an AI model has little to work with beyond the inconsistency.

Data completeness sits under the operating rhythm. Revenue information is scattered across the CRM, email, call recordings, and a data warehouse, and each source holds part of the story. An AI that sees only one of them can give a confident answer that is still partial. You have to supply consistent habits and complete data yourself.

Failure four: using the wrong tool for the job

The fourth failure is a mismatch between a method or model and the job it is asked to do. It shows up most in forecasting and coaching.

Young gave an example from sales methodology. MEDDIC is a deal-qualification framework, commonly expanded as Metrics, Economic buyer, Decision criteria, Decision process, Identify pain, and Champion.

MEDDIC is not a forecast tool. It is not a methodology either. It's how we test a deal.

Lesley Young, Revenue Mavericks

It tells you whether a deal is real. It says nothing about the quarter's total.

A similar mismatch can appear when a team expects a general-purpose AI assistant to produce a forecast from whatever was typed into the CRM. We draw this analogy ourselves; Young did not. Improving forecast accuracy depends on consistent inputs, because inconsistent stages and stale close dates will produce an inconsistent forecast.

A lot more math than art. As AI orchestration matures, opportunity reviews and business reviews should be automated.

Asking the right questions at the right time. Get around the systems to make a buck.

Quoc Tran, Revenue Mavericks

RevOps has to understand where the motivations are. Automation and human incentives pull in different directions, and RevOps has to hold both.

Coaching has the same issue, and Young's answer is to lead with data.

When you go into a conversation to coach, it's like, what is the data we're using.

Lesley Young, Revenue Mavericks

She once showed a rep who believed she was a poor salesperson, and who said she could not handle the stress, a record Young had tracked of where the rep stood in month one, what she would do, and where she landed in month three. The rep was within about 10K of her number, and the data gave them something concrete to talk about.

Young used a spreadsheet for this. Our own view is that the same principle carries over to an AI sales coach that starts from evidence. We recommend running coaching and deal reviews on conversation intelligence data from real calls.

What to fix before you turn AI on

The sequence is objective, process, operating rhythm, and data. After that, turn on AI and keep the sales team in charge of the result.

Start by deciding what the AI is supposed to change, such as response rates on cold outbound, time to a forecast call, or coverage on coaching. A tool with no target produces activity only.

Write down the process and assign it to an owner, then set the operating rhythm by deciding which meetings happen when, what each one reviews, and which definitions stay fixed for the quarter. Capture a baseline before rollout so you can tell whether AI moved anything; learning to measure sales productivity gives you that starting point.

Before you pick a tool, list the systems where revenue information lives and ask whether it can see all of them or only one. Only after that does it make sense to let AI agents execute a defined process, because an agent can only carry out a process someone has already designed. Throughout, keep a person on every output that carries your name, as Tran's team does with outbound.

How we approach AI in the sales org

One more problem sits underneath the four failures above, and it is the one we built for. Revenue data sits in the CRM, email, calls, and the data warehouse, and each needs its own query language. No single AI sees the whole picture, so none can give a complete answer, and a partial answer is hard to act on. We built the Revenue Graph to provide the complete revenue picture across structured and unstructured data in your revenue systems, so the questions a chief revenue officer (CRO) asks, such as why win rates are dropping in a region or where the quarter will land, can be answered from everything the team has rather than from one source.

Terret Nexus is our answer-to-action engine, and it is built on two layers. Our AI Architects deliver the answers: they analyze the complete revenue picture and design the GTM system. Agents handle execution: they deploy workflows, coach reps, score deals, and generate forecasts based on what the Architects design. The flow is to ask the question, get the answers, operationalize them automatically, and activate them at critical moments. We built Terret Nexus to connect the two layers. Your sales team stays the driver, and the objective and process you set remain the starting point.

FAQs

Should AI write and send our outbound without a person reviewing it?

No. AI is useful for research and first drafts, but a rep should own the final version of anything that goes out under their name. Buyers can tell when a message shows no understanding of their business or role, and they tend to ignore it.

What do we need in place before we turn on AI?

Start with a stated objective, then a written process, then a steady operating rhythm, then complete data. Each layer gives the model something to work from. Without them, AI tends to produce noise that your team has to sort through by hand.

How do we know whether AI is making a difference?

Capture a baseline before rollout and tie it to the objective you set, such as response rates on cold outbound or the time it takes to reach a forecast call. Without a starting point, you can see more activity but you cannot tell whether anything improved.

Can AI fix a forecast our leadership does not trust?

Not by itself. A model built on inconsistent stages, stale close dates, and changing definitions of commit will return an inconsistent forecast. Fixing the inputs and the operating rhythm comes first, and that is also the foundation for a forecast the CFO will trust.

How can managers use AI to coach without joining every call?

Managers can start from the data in recorded conversations instead of their own memory of the calls they attended. That makes it possible to coach reps without sitting in on every call, as long as the coaching conversation begins with evidence the rep can see too.

Should we build our own revenue AI or buy one?

It depends on how much engineering time you can commit and how many data sources the system has to read. Revenue information is spread across the CRM, email, calls, and a data warehouse, so the connections and upkeep add up. Teams weighing whether to build your own revenue AI should work through those costs before committing to either path.

Where do AI agents fit in the sales process?

Agents fit after the objective, process, operating rhythm, and data are defined. They carry out a process that someone has already designed, such as deploying a workflow or scoring a deal. AI revenue agents work best when the team knows what a good result looks like and keeps a person responsible for the outcome.

Get AI working inside your sales process

If AI has not delivered what you expected, pick one tool already in your stack and write down the outcome it was meant to change. Then check whether a written process and an owner sit behind that outcome.

We can walk through how this applies to your process, including where your data sits today and which steps are worth settling first. Request a demo and we will start from your questions.