Your Manufacturing Revenue Is Growing, but Margins Aren’t Keeping Up

TL;DR

• A manufacturer can have an ERP, a CRM, and growing revenue while still relying on manual processes that erode margins.
• A report from Brazilian business newspaper Valor Econômico highlights a manufacturer with BRL 50 million in annual revenue and 20 employees dedicated to manual back-office tasks.
• AI agents can reduce this workload, but they need clear rules defining which decisions they can make and which they must escalate.
• Automating before governing pricing, terms, approvals, and exceptions only makes errors happen faster.

Why Back-Office Automation May Not Restore Your Margins

A recent report published by Brazilian business newspaper Valor Econômico highlighted a growing tension: manufacturers and distributors are generating more revenue but earning less profit.

The article cites Beyond the Bytes and describes a typical middle-market manufacturer with BRL 50 million in annual revenue and 20 employees dedicated to manual back-office tasks. It also reports a projected average 15% increase in client profit margins, with an estimated three-month timeline.

For a CEO, the relevant point is not just the size of the team involved. It is the operational contradiction: the company already has business systems, yet it still depends on people to interpret information, transfer data, and decide how each situation should be handled.

The underlying problem, therefore, is not necessarily a lack of technology. It is the layer of manual processes sitting on top of systems such as ERP and CRM platforms.

This scenario often creates a false choice. On one side, the company can keep adding people to absorb operational growth. On the other, it can quickly hand the workflow over to AI agents. The first option preserves the cost structure. The second, when adopted without governance, can reproduce the same problem at a larger scale.

A Manual Process May Be Hiding an Unstructured Decision

Not every repetitive task is purely about execution.

When processing an order, applying commercial terms, or reviewing credit, the back office may be making small decisions that have never been formally defined:

• Are the terms within policy?
• Can the order proceed without additional approval?
• What discrepancy can be tolerated?
• When must the situation be routed to a person?
• What qualifies as an exception?
• What data supports the decision?

As long as these answers remain embedded in individual employees’ experience, the operation runs on interpretation. ERP and CRM systems store information, but the rule connecting that information to an action may remain scattered across people, messages, and habits.

This is precisely where seemingly simple automation becomes a risk. The agent can execute the task, but it cannot find an explicit boundary between when it should decide and when it should escalate.

The Critical Boundary: When to Act and When to Escalate

An AI agent handling billing, credit, or orders needs more than access to data. It must operate within a defined decision boundary.

That means establishing in advance:

• which situations can be handled autonomously;
• which commercial limits cannot be exceeded;
• which exceptions require approval;
• which information must be recorded;
• when the agent must stop execution and escalate the case.

The sequence matters. First, the company structures the decision. Then, it delegates execution.

This principle is relevant in other contexts as well. According to PCWorld, Claude completed a FreshDirect purchase with support from 1Password, authenticating the user, selecting items, and completing the order without direct browser interaction.

The autonomous execution is technically significant. Even so, the economic questions remain: what price is acceptable, when should an item be substituted, and when should approval be requested?

In B2B commerce, these questions carry more weight because each transaction may include a negotiation history, commercial policies, and account-specific terms. An agent does not eliminate that complexity. It needs to receive it in a governed form.

The same principle appears in a case reported by Finextra involving CaixaBank and Visa. The companies completed a transaction initiated by an AI agent within authorization and control models defined by the user. The key was not simply allowing the agent to make a payment, but establishing the boundaries within which it could operate.

The Goal Is Not Maximum Autonomy

Evaluating AI agents solely by the number of tasks they automate can lead to the wrong conclusion.

The economic objective should not be to maximize autonomy. It should be to expand autonomous execution only where the underlying decision is sufficiently clear, while preserving human escalation for cases involving risk, ambiguity, or significant financial impact.

That changes the question being asked in the boardroom. Instead of asking, “How many people or steps will the agent replace?” leadership should consider:

• which decisions have already been formalized;
• where exceptions consume the most capacity;
• which errors have the greatest impact on margins;
• which situations can be handled consistently;
• which decisions still depend on unstructured context.

This approach also prevents the projected average 15% margin increase cited by Valor Econômico from being treated as an automatic result of adopting AI. Any improvement depends on the structure on which the agent is deployed.

The Cost of Inaction

Doing nothing leaves the company trapped in the paradox described in the report: more revenue, more operational effort, and margins that fail to keep pace with growth.

But acting in the wrong order also has a cost.

When a chaotic process is automated, the company may reduce execution time without improving decision consistency. Errors then move faster, affect more transactions, and require additional layers of human review.

The cost of inaction, therefore, is not just failing to adopt AI. It is also delaying the formalization of commercial knowledge that is currently distributed across individual employees.

That knowledge includes pricing criteria, terms, approvals, limits, and exceptions. As long as it remains implicit, every new automation initiative will need to reconstruct it. Once it becomes governed policy, it becomes a reusable operational asset.

Principles for Introducing Agents Without Automating Chaos

• Identify the decisions hidden inside manual tasks.
• Separate repetitive execution from commercial judgment.
• Define autonomy limits before choosing what to automate.
• Formalize pricing, terms, approvals, and exceptions as policies.
• Establish objective criteria for escalating cases to people.
• Preserve decision records and auditability.
• Start with workflows whose logic can already be described clearly.
• Measure outcomes by consistency and financial impact, not just speed.
• Treat negotiation history as company knowledge.
• Expand autonomy only after validating the decision boundary.

FAQ

Should Every Manual Process Be Handed Over to an AI Agent?

No. Manual processes can combine simple execution with contextual decision-making. The company must separate these two dimensions before automating them.

Don’t ERP and CRM Systems Already Provide This Governance?

These systems are part of the infrastructure, but their existence does not mean that commercial criteria, exceptions, and escalation limits have been formalized. The Valor Econômico report highlights exactly this issue: manual work continues to exist on top of the systems.

When Should an Agent Escalate a Decision?

When the transaction exceeds defined limits, presents an exception, lacks sufficient data, or involves a decision the company has not yet converted into an explicit policy.

What Customers Are Already Experiencing

“The native AI agents can answer technical questions about the catalog and assist with negotiations using actual business data.”

Carlos M., Head of Sales, in a review published on Capterra.

A Case That Illustrates the Point

A JOKR case published by Retail Tech Innovation Hub reports that the quick-commerce company reached EBITDA break-even after a five-year rebuild centered on AI and automation.

The experience reinforces the central point: automation amplifies the existing structure. The outcome depends on well-designed rules, criteria, and processes established before implementation.

Read the case at Retail Tech Innovation Hub.

About This Publication

O Custo da Venda is a CWS Platform publication focused on the economics and governance of B2B commercial operations.

CWS Platform is a B2B Commerce Platform for Governed Negotiation. In this architecture, commercial decisions are structured before execution.