# Before investing more in AI in 2027, examine your revenue operation

Fabio Munhoz · 2026-11-20

Canonical: https://fabiomunhoz.com/blog/revenue-operations-ai-diagnostic-2027/

Language: en

A practical diagnostic for deciding where agents can help and what needs to change before you expand automation.

## In brief

A useful diagnostic follows a real commercial journey and locates where work loses information or decision-making capacity. It should identify a priority problem, an owner, evidence to collect, and a review date. The solution may involve process, data, integration, or AI.

An AI budget can start with a list of tools. I would rather start by following a sales opportunity.

Where did it come from? Why did the company choose to pursue it? What information reached the salesperson? What happened after the proposal? If someone corrected the process, was that correction available to the next person?

These questions locate the work a new tool is supposed to improve. They also help a team avoid adding an agent to a process whose underlying problem remains unclear.

Earlier in this series, I described a [Revenue Harness](https://fabiomunhoz.com/blog/what-is-a-revenue-harness/) as a structure connecting context, execution, controls, and learning. Before building that structure, examine the handoffs where your operation loses information or the ability to make a decision.

## Choose a journey you can observe

An assessment of the entire operation can become an inventory that is difficult to use. Start with a bounded journey: from the arrival of an opportunity to the first proposal, for example.

Bring together the people involved and select records with different outcomes. Looking only at won deals leaves out situations in which the process did not work. This first review does not need to produce a statistical estimate. Its purpose is to identify questions and gather evidence for investigating them.

Consider an illustrative scenario. A company sells implementation services. Marketing captures interest, sales identifies the need, and a technical team estimates the work. Proposals take too long. The initial response is to ask an agent to write them faster.

Following the journey reveals missing information about integrations. The proposal is waiting because nobody can estimate the scope. Faster writing will not resolve that dependency. The first intervention may be to improve how information is collected and checked before the technical handoff.

## A framework for the conversation

I would use the questions below to guide the diagnostic. They are not a certification or a maturity index. Each answer should lead to a verifiable record, an accountable person, and a specific gap.

| Dimension | Question | Evidence to look for |
| --- | --- | --- |
| Objective | Which commercial outcome do we want to improve? | A defined outcome and a description of current performance |
| Context | Does the next person understand why this opportunity deserves attention? | Origin, need, constraints, and account history |
| Data | Are the fields used for decisions clearly defined and current? | Records with a source, date, and owner |
| Handoff | Does the next stage receive enough information to proceed? | Entry criteria and reasons for returning work |
| Control | Who can authorize, stop, and correct an action? | Applicable responsibilities and permissions |
| Learning | Does a correction change the next cycle? | A recorded decision and an observed process change |

Mark what has been demonstrated, what is missing, and what remains unchecked. An unknown answer should not receive a reassuring score simply because a CRM field exists.

## Separate the problem from the instrument

After the diagnostic, identify the kind of intervention required. It might be a business decision, a data correction, an integration, a process rule, or a task that benefits from AI.

That distinction changes the investment. An ambiguous field requires agreement across teams. An outdated copy requires a fix to the flow of information. Research across varied sources may justify model assistance, provided the result can be checked.

Anthropic recommends starting with the simplest solution and adding complexity when it improves outcomes. Its guidance distinguishes predictable workflows from agents that decide how to carry out their work. This is a useful technical reference for avoiding the assumption that every commercial need requires autonomy. [Source: Building effective agents](https://www.anthropic.com/engineering/building-effective-agents).

## Who should participate in the revenue diagnostic

Choose participants according to the journey. Sales can explain how an opportunity is qualified; the technical team can identify what it needs to estimate delivery; RevOps can show where the information is recorded and how it moves between systems.

Ask each person to bring an example rather than a general assessment. If someone says the brief always arrives incomplete, open a record and identify exactly what was missing. There may be a question nobody asks, an answer nobody saves, or information that exists but is inaccessible to the person who needs it.

These situations call for different corrections. The exercise should end with a map of the observed journey and a short list of dependencies. For each dependency, identify who can resolve it and how the team will verify that it has been resolved.

Also record what stays outside the first test. In the implementation scenario, preparing a summary does not authorize an agent to estimate a price or promise a delivery date. Defining the scope allows the team to compare the work without introducing a commercial change it has not yet evaluated.

## Define the first test and the limits of its conclusions

In the implementation example, the company could test an AI-prepared scope summary. Its information would be traceable to authorized records, and a technical reviewer would check it before a proposal was prepared.

The test needs a definition of success, such as fewer handoffs returned for missing information. It should also track review effort and newly introduced errors. Time saved in preparation is not an operational gain if the work reappears during correction.

Define the period, eligible cases, and stopping conditions before starting. An observed improvement may justify continuing the test. Attributing it to the intervention requires examining other changes during the same period.

The assessment should end with an actionable choice: which problem to address, who owns it, what evidence to collect, and when to revisit the decision.

**Before approving another tool for 2027, ask your team to show the handoff it intends to improve.** If nobody can show where work gets stuck, the most useful next investment may be making that journey visible.

## Questions and answers

### Where should a revenue operations diagnostic begin?

Choose a bounded journey, such as opportunity intake through the first proposal. Bring together its owners and examine records with different outcomes to locate gaps and dependencies.

### How do you know whether the problem needs an AI agent?

First establish whether the difficulty comes from a decision, data, an integration, or a process. Consider AI when there is a suitable task and its output can be checked.

### What should the first test measure?

Define the desired outcome and also track errors, review effort, and rework. Record eligible cases, the test period, and stopping conditions before starting.
