Skip to content
Back to blog
artificial-intelligencemarketingcommunicationautomationsmes

Applying AI to Corporate Communication and Measuring Results

A practical guide to testing AI in email, social media and customer service, with human review, quality metrics and a baseline comparison.

ByJavier Chulvi

AI can help draft replies, classify messages and retrieve information for customer support. Results depend on the process, data and team review. Start with a task you can compare against the existing workflow to decide whether the change is worthwhile.

Email: prepare and review before sending

An initial pilot can classify queries and suggest replies based on approved information. Keep human review and measure the complete workflow, including corrections.

For campaigns, compare variants using a recipient allocation that allows the change to be evaluated. Decide in advance what to track: clicks, conversions, unsubscribes, errors and preparation costs. A change in opens does not establish a sales increase by itself.

There is no universal improvement percentage for using AI. Do not import another company's figures into your budget without checking whether the context and measurement are comparable.

Social media: drafts with sources and an owner

One possible workflow gathers relevant topics, generates proposals and reviews each post before scheduling it. Verify facts, material usage rights and suitability for the company's voice.

Illustrative example: a shop prepares answers to common questions from its catalogue. The team checks prices, availability and terms before publishing. The trial also records rejected drafts and correction time.

Customer service: scope and escalation

Define which questions the solution can handle and when it must hand over to a person. RAG can provide document context, but answers still need evaluation: having a source does not guarantee it is interpreted correctly.

Test routine, ambiguous and out-of-scope queries. Include outdated sources, unavailable services and requests for information the user is not authorised to access.

Measure the whole process

  • Time: preparation, waiting, review and corrections per case.
  • Quality: accepted replies, errors and severity, using the same criteria for both options.
  • Cost: usage, tools, support and human effort; initial investment recorded separately.
  • Service: resolved and escalated queries, satisfaction and complaints, where measurement is consistent.

Record the sample, period and configuration changes. Compare equivalent cases and reserve a sample that was not used for tuning. An observed pilot difference does not establish the same result across the company.

A first step for your SME

Choose one channel and task, agree who reviews the work and define what would justify continuing. Start with organising email or drafting replies from internal documents.

Review which data is needed, who can access it and which providers receive it. MCP alone does not guarantee local processing.

Navel can scope a diagnosis covering integrations, costs and pilot criteria. Request a process review.

Editorial review, 17 October 2026: removed adoption, open-rate and conversion percentages without identifiable sources. These examples are illustrative, not Navel client results.

The LLM fine-tuning service can adapt formats and style while retaining editorial review. For real-estate businesses in Málaga, one proposed application is drafting replies to property enquiries using approved information.

Contact

Interested in this topic?

Let's talk about how we can help you implement these systems in your business.

Let’s talk
Tell us what you have in mind.