Is AI being implemented well in Peruvian businesses?

Real progress and governance gaps coexist

The answer requires examining each process. Some Peruvian companies document employee assistants and AI-generated content. There are also gaps in training and oversight. The reviewed sources identify these practices and difficulties; they do not establish whether implementation across the entire Peruvian economy is good or bad.

For this article, implementing AI well means addressing an identified need with authorized data, an accountable person and comparable outcomes. Using an assistant to draft text is one stage; embedding it in a process involving customers, ERP or CRM requires additional controls. Ask which task improves, who reviews exceptions and what it costs to deliver an accepted result.

What the Peruvian evidence says, and its limits

EY Peru’s AI & Boards 2026 collected an online survey of 250 board members, shareholders and senior executives between January 12 and March 5, 2026, associated with seven sectors. These are individual responses; they should not become a count of 250 distinct companies or a representative sample of micro and small enterprises. [1]

In its statement dated June 15, 2026, EY reports that only 10% of the organizations in the study had approved formal AI policies and that 68% of executives had received no training on its use and impact. Some 78% believed it improved their decisions. That last figure measures a perception, not demonstrated productivity or integration with operational systems. [2]

What is actually being implemented

On August 10, 2026, Interbank reported that its first phase deployed Claude Enterprise to a group of knowledge and technology employees, with training and support. The statement proposes expanding generative and agentic AI; it does not demonstrate that all its processes operate autonomously or publish a measured financial return. [3]

Outside banking, Alicorp explained on February 12, 2026, that an AlaCena campaign used AI-developed content produced in one week. This is a concrete marketing application, without an experimental comparison of sales or profitability. [4] In a María Almenara case dated November 13, 2025, Microsoft describes adoption of Microsoft 365 Copilot; introducing agents appears as a future plan rather than a completed deployment. [5]

These institutional and commercial sources document practices reported by their participants. Their scope does not establish how other businesses perform or replace an independent evaluation.

Documented shortcomings and gaps we should evaluate

EY identifies difficulties involving knowledge, resources, governance and resistance to change. [1] A response should include training for concrete tasks, authorized tools, a process owner and a budget for review and support. Buying licenses without explaining when to use the assistant, what information it may receive and how to correct errors leaves important decisions to individual employees.

Each project should also evaluate these potential technical gaps. They are risks we recommend checking, rather than demonstrated failings of the companies cited:

• Disconnected data: CRM, ERP and files may define customers or products differently. Assign owners and rules for reconciling identifiers.
• Outdated information: define acceptable age for prices, stock and documents, and how to stop an answer when evidence is insufficient.
• Broad permissions: authorize queries and actions by user and purpose. A persuasive answer does not grant permission to change a sale.
• Superficial evaluation: test errors, ambiguous requests and human handoffs alongside routine cases.
• Incomplete costs: count integration, licenses, consumption, monitoring, review, support and corrections; include team preparation.

An example: from a sales assistant to a reliable operation

Consider a distributor preparing quotations. This example is illustrative. First integrate CRM customer records with ERP prices and stock, including currency, tax and validity rules. Preserve the source and update time. The assistant then prepares a draft using only data the salesperson is authorized to access.

If a product code is ambiguous or stock information is stale, request review. A person confirms prices, availability and terms before issuing the quotation. If recording the transaction is authorized later, use a unique identifier to prevent duplicate writes on retries and reconcile what was requested with what was actually recorded. Measure the complete journey, including correction work, and collect feedback from sales and operations.

How to address the gaps: an adaptable 90-day plan

This sequence is a working proposal, not a guaranteed timeline. Choose a bounded process with observable errors, a business owner and users willing to test it. Adapt the schedule to integration needs, budget and risk.

The metrics that should justify investment

• Time to an accepted result, including review and rework.
• Material errors per case and the proportion of results requiring correction.
• Human handoffs and time to resolve an exception.
• Total cost per resolved case, with implementation costs allocated over an explicit period.
• Access incidents, outdated data and duplicate actions.
• Effective team use and changes incorporated through feedback.

Compare tasks of similar difficulty with the initial process. Freed time needs a defined use to become additional capacity; it does not automatically become accounting savings. If review effort grows or quality falls, narrow the scope and correct the problem before expanding.

Privacy and governance in Peru

Law No. 31814 has a regulation approved through Supreme Decree No. 115-2025-PCM, published on September 9, 2025. It addresses risk levels and phased implementation. Law No. 29733 and its regulation approved through Supreme Decree No. 016-2024-JUS also require consideration. [6] [7] Before uploading customer data, identify purpose, access, retention and provider terms, including transfers where applicable. Review applicable requirements with specialist advice; this article does not interpret particular deadlines.

Where Cloudata and Dova can help

Cloudata can help define the process and prepare its operational data foundation. Dova organizes transformations in Mappings, execution context in Sessions and coordination in Workflows. Its documentation distinguishes drafts, publication and execution, and provides guidance on connections and workers. [8] Monitor supplies execution tracking; outcomes must be checked against the business rule. [9]

This work can support an AI project with prepared data and a verifiable workflow. It does not assume Dova includes an autonomous agent or independently solves model evaluation, privacy or organizational change. A specific proposal should identify available connectors, responsibilities and acceptance criteria for your use case.