Karmina Review Agent

5. What do you do, specifically?

01. Contrast the data before building the account

Check periods, variations, signs, totals, labels and coherence between tables, highlights and conclusions. When two sources do not match, it does not decide which one is correct by intuition: it points out the discrepancy, shows where it comes from and asks for review.

02. Review the evolution of the strategy

It relates the results to the targets, audiences, content territories and the agreed roadmap. It does not value a month in isolation if the plan requires a cumulative evolution or depends on the time of the course, the season or the campaign.

03. Check the calendar before sending it.

It reviews the balance of topics, audiences, formats, and channels; identifies gaps, repetitions, and deviations; and points out text, adaptation, or technical preparation problems that need to be solved before calendar sharing.

04. Separate incidents from recommendations

An objective error is not the same as an editorial preference. The agent differentiates what should be corrected, what should be valued and what only needs to be observed. This prevents a long list of comments from hiding important decisions.

05. Prepare useful content for reporting

It transforms validated findings into headlines, conclusions and recommendations that the team can adapt to the PPT or report. The result does not replace professional interpretation, but reduces the time spent ordering and drafting a first version.

06. Filter opportunities for the account

It can start from a wider radar of platform changes, formats, habits, referents or trends and filter it according to the strategy, channels, audiences, time and resources of the client. The goal is not to convey fifty news, but to identify the few that deserve a decision.

07. Keep a trace of the review

It records what source you have consulted, what you have detected, what criteria you have applied and what action you propose. This traceability makes it easy to validate the find, correct the system, and learn from the team’s decisions.

6. Two Moments, Two Different Deliveries

We do not mix the close of the month with the review of the following calendar

The Karmina Review Agent works on two related but independent circuits. Each circuit has its own sources, questions and results.

Circuit A. Month closure and support for the report

When activated When the period is over and the necessary data is available.

What you get

• Strategy, objectives and current roadmap.
• Data from authorized sources.
• Report or template period.
• History is needed to interpret evolution.
• Dictionary of customer metrics and criteria.

What to check

• Consistency of data, periods and variations.
• Correspondence between tables, highlights and conclusions.
• Cumulative evolution with respect to the objectives.
• roadmap's milestones, pending or deviated.
• Incidences that require human verification.

What does it deliver

• Executive semaphor of the period.
• Short list of incidents with source and priority.
• Texts prepared to incorporate or adapt to the report.
• Recommendations identified as a RECOMMENDATION FOR THE REPORT.

Circuit B. Review of the calendar before the client

When activated When the calendar is complete enough to review it and there is still time to apply changes.

What you get

• Editorial calendar and state of the pieces.
• Strategy, roadmap, audiences and content lines.
• Voice, language, channels and formatting criteria.
• Results or recent learnings that affect the period.
• Radar of opportunities available.

What to check

• Technical preparation and visible incidents.
• Editorial quality and brand consistency.
• Strategic balance of topics, audiences, formats and channels.
• Application of learning from the previous period.
• Opportunities worth applying, trying, preparing or observing.

What does it deliver

• The light of the calendar.
• Prioritized corrections before shipment.
• Gaps or strategic imbalances.
• A short selection of applicable opportunities.
• Recommendations identified as a RECOMMENDATION FOR THE CALENDAR.

Featured message Each recommendation indicates at what time it belongs. Thus, the team knows what modifies the current report and what needs to enter the next calendar.

7. How it works from start to finish

1. The agreed revision is activated

The agent works according to a defined cadence: when the data is closed, when the calendar reaches the agreed state or when a responsible person requests the review. It does not interpret an incomplete document as if it were a final version.

2. Check that you have the minimum tickets.

Validates that the sources are accessible, that the periods coincide and that the necessary documentation is in force. If a critical entry is missing, it indicates this before proceeding.

3. Order and normalize information

It separates what is planned from what is published, differentiates unique pieces of adaptations per channel and identifies reprogramming, cancellations, exceptions or even provisional data.

4. Apply the criteria of the account

Contrast the information with the objectives, the roadmap, the editorial lines, the style book, the metric dictionary and the agreed review rules.

5. Classify the findings

Each finding is marked as error, discrepancy to check, strategic deviation, opportunity, recommendation or simple observation. It also has a priority and a destination.

6. Prepare a short and actionable result

The first layer is executive: headlines, traffic light and decisions. Details and sources are available for those who need to check the find.

7. A person reviews and decides

The person in charge of the account validates the data, accepts or discards the recommendations and provides context when an exception was not documented.

8. The system learns from correction

Useful decisions are incorporated into the agent's criteria, examples, or evidence. Thus, future revisions depend less on improvised instructions and more on governed knowledge.

8. Example of result

A calendar can be complete and yet not ready to send

Simulated situation The team has prepared the next month's schedule. Every week has content and the main pieces already have copy. At first glance, the calendar looks complete.

What the agent detects

• The priority content line of the quarter only appears once.
• The same audience concentrates most of the pieces, while a second strategic audience is not represented.
• Two publications use a data that is not supported by the indicated source.
• An adaptation keeps the text of Instagram on a channel that needs another approach.
• A recommendation from the previous report has not yet been implemented or discarded.
• A platform novelty is relevant, but it is convenient to prepare a limited test instead of incorporating it into the entire calendar.

How to present it

• Red: Two unsourced claims; stop and check.
• Amber: public imbalance and absence of the priority line.
• Green: frequency, temporary coverage and other revised pieces.
• RECOMMENDATION FOR THE CALENDAR: replace one of the redundant pieces with a content oriented to the absent audience.
• OPPORTUNITY DECISION: Try the new format in one piece and review the result before scaling it.

Result for the team The person responsible does not receive a generic audit or an endless list. It receives the points that can change a decision, with enough context to validate them.

Publication Note This example is an explanatory simulation. Any real case, customer data or testimonial should only be used with authorization.

9. The filter of opportunity

From industry noise to a useful decision for the customer

A novelty does not automatically become a recommendation. The Karmina Review Agent can apply a funnel so that only opportunities that fit with the account reach the calendar.

Radar Ampli Platform changes, emerging formats, benchmarks, behaviours, trends and movements in the sector.

Account Filters

• Active channels.
• objectives and public.
• Editorial territories.
• Time of the course, season or campaign.
• Real resources of production.
• Risk, maturity and available evidence.

Short selection Only opportunities that could affect a content decision.

Decision

• Apply: He is strong and mature enough to take it in.
• Test: It deserves a limited test with measurement criteria.
• Prepare: It does not apply yet, but the account must be left ready.
• Observe: It does not justify any immediate change.

Featured message Radar does not exist to show that we know a lot of news. It exists to decide what few deserve time, budget or space on the calendar.

10. What do we need before we launch it?

We don’t need everything to be perfect. We need to know which source and which person has authority.

The quality of the agent depends on the knowledge and criteria it can apply. Before the pilot, we identify what already exists, what is outdated, what is contradicted and what has yet to be defined.

Essential to get started

• A person responsible for the process and validation.
• A description of the current calendar flow and reporting.
• Strategy, objectives and current roadmap.
• Channels and platforms that actually go into the review.
• Authorized access to the necessary data or exports.
• Current calendar and report template.
• Definition of Priority Metrics.
• Approval circuit and time limit to apply changes.

Highly recommended

• Voice, brand and style guide.
• Publics, editorial territories and criteria by channel.
• Examples of well-resolved reports and calendars.
• History of decisions, incidents and exceptions.
• Data protocol, permissions and confidential information.
• Prioritize recommendations.
• List of actions that the agent cannot execute.

If something does not exist

Let us not replace it with a hidden assumption. We turn it into a project question. We can help document the minimum criteria needed, assign responsibility to it and test it during the pilot.

11. You don’t have to have a perfectly documented company.

Many organizations already have the necessary knowledge, but they have it distributed among presentations, folders, platforms, emails and people. The first phase does not consist of creating dozens of new documents or connecting everything with the AI.

We start with the essential knowledge to review this process: what the account wants to achieve, how it is measured, what source prevails, who validates and what exceptions must be respected. This base can then grow and serve other agents.

Featured message We don't connect all the folders. We select the sources that the organization recognizes as current, necessary and authorized.

12. How we implant it

From the current review to a proven agent within your process

1. Diagnosis of the process

We map how you prepare the calendar and the report today: tickets, tools, managers, dates, revisions, recurring errors and expected result.

2. Knowledge Foundation

We order the sources, definitions, permissions, examples, limits and criteria that the agent must apply. We do not take documents indiscriminately.

3. Design of the circuit

We define when each review is activated, what you can consult, how an incident classifies, what format it delivers and in which cases it should be stopped or scaled.

4. Pilot with real controlled cases

We provide the agent with historical materials or with a delimited account. We compare the result with the team review and document errors, omissions and false positives.

5. Calibration and testing

We adjust criteria, priorities and formats. We create a test bench so that an improvement does not break another part of the process and so that the system can be revalidated when the sources change.

6. Onboarding Team

We explain how to activate it, interpret the results, provide context, correct it and detect when it should not be used. The agent enters a shared process, not a black box.

7. Follow-up and evolution

We measure the use, review time, quality and incidences. When they change strategy, a platform or a responsibility, we update the corresponding source and retest the system.

13. Human supervision and boundaries

Automating review does not mean automating responsibility.

The Karmina Review Agent is designed so that important decisions continue to have a responsible person. The level of autonomy depends on the risk and is defined before the pilot.

The agent can

• Read the authoritative sources needed for review.
• Compare data, texts, goals and schedules.
• Detect inconsistencies and point out the source.
• Classify findings and order them by priority.
• Prepare drafts, headlines and recommendations.
• Record doubts, exceptions and points that need validation.

Agent can't

• Publish or schedule content.
• Modify a campaign, calendar, or final report without approval.
• Give for good a data when the sources do not match.
• Inventing customer context that is not documented.
• Replace the strategic interpretation of the account manager.
• Share information outside of agreed access and purposes.

When it stops and scales

• A necessary source is missing or not updated.
• Two official sources contradict each other.
• It detects personal, confidential or unauthorized data.
• The recommendation could have a relevant public or reputational impact.
• The case does not fit with the criteria that have been tested.

14. What tools and sources can it work with?

The agent adapts to the existing system as long as access is viable, authorized and proportional to its mission. Depending on the project, sources may include:

• Planning tools and editorial calendars.
• Platforms of metrics and social analytics.
• Strategy, branding and roadmap documents.
• reporting spreadsheets, presentations and templates.
• Document managers and project folders.
• Radars or internal knowledge bases.
• History of content, decisions and learning.

We do not assume that any integration exists or that all data can be consulted in real time. During the diagnosis we check the actual access, quality, update frequency and permissions. When a direct connection is not appropriate, the process can work with a controlled export.

Trust note Tool logos should only be displayed on the landing page when integration or compatible stream has been validated. They should not be used as a generic promise of connection.

15. How do we measure if it adds value?

The success is not that the agent generates more comments. It makes the review better and more useful.

Before the pilot we set a baseline of the current process. Then compare indicators such as:

• Time to prepare and review the report.
• Time of calendar review before shipment.
• Targeted incidents detected before reaching the customer.
• Percentage of findings accepted, discarded or corrected.
• Number of recommendations that end up generating a decision.
• Repeated errors that had already been identified.
• Monitoring time needed to validate the result.
• Perception of clarity and usefulness by the team.

Saving time is not an improvement if subsequent human review doubles. That’s why we measure together speed, quality, consistency and supervision load.

16. For whom does it make sense?

It can fit if...

• Manage multiple accounts, channels, or recurring calendars.
• Prepare periodic reports with data from more than one source.
• The strategy exists, but it costs to check the application month by month.
• The reviews depend on one or two people with very tacit knowledge.
• You want to reduce errors before sending materials to the customer.
• You need to turn a wide radar into few actionable opportunities.
• You are willing to assign a responsible person to the pilot.

This is not the first step if...

• There is no minimally defined strategy or goals.
• No one can validate whether a recommendation is correct.
• Necessary data is not accessible or legally usable.
• The process changes completely each time and has no recognizable sequence.
• Look for a tool that publishes autonomously without human review.
• The goal is to replace a professional responsibility, not to improve a process.

17. Why Karmina AI Studio?

We build agents from the real knowledge of the work

Karmina is a digital marketing agency. This means that we know the process that this agent has to review: strategy, schedule, production, channels, reporting, customers, deadlines, exceptions and editorial responsibility.

Karmina AI Studio turns this experience into specialized systems that work with the sources, criteria and limits of each organization. Let’s not start with the demonstration of a tool. We start with the process, the knowledge that sustains it and the person who will continue to respond to the result.

When the project requires it, we incorporate Utopiq’s focus on artificial intelligence, ethics and sustainability to work on governance, impact, risks and responsible adoption.

Recommended links

• Meet Karmina AI Studio → /ca/karmina-agents
• View our → method /ca/metode
• Read the manifest of AI responsible → /ca/ia-responsable

Internal note You need to confirm the URLs and decide if Utopiq appears on all landing pages, only on projects with actual participation or within a general contributor block.

18. Knowledge Foundation

An agent does not start with a prompt. Start with what your company knows.

The strategy, vocabulary, goals, sources, examples, exceptions, and validation criteria do not live within a magical instruction. They are part of the organization’s knowledge.

We call it Knowledge Foundation the process with which we select, order and govern this knowledge so that the agent can apply it. We identify which source is official, who is responsible, when it should be updated, who can consult it and what should happen when the information is insufficient or contradictory.

The base created for the Karmina Review Agent may be reusable. A voice guide, a KPI dictionary or an audience map can also feed future content agents, reporting or planning, always with their own permissions and rules.

Recommended link /ca/knowledge-foundation

19. Training and adoption

The agent only adds value when the team knows how to work on it

The implementation may include training for the people who will activate, review or govern it. We work like interpreting the traffic light, checking a source, accepting or discarding a recommendation, providing an exception and detecting when the agent should stop.

We can also form other teams to understand what an agent is, what data they can use, how to protect information and what decisions must remain under human supervision.

Recommended link /formacio-ia-empreses

20. Frequently Asked Questions

Is it a standard tool that we can activate immediately?

No. The review process can share a structure, but goals, sources, metrics, language, permissions, and approvals change according to each organization and each account. First we adapt and test the agent in a delimited environment.

Replace the person responsible for the account?

No. The agent prepares and reinforces the review. The person in charge provides context, interprets the data, validates the conclusions and decides what is applied or communicated to the client.

Can you publish or modify the calendar?

Not in the proposed configuration. Its mission is to review and recommend. Any capacity for future action should be assessed separately, with permits, tests and approvals proportional to risk.

Do you need access to all of our tools?

No. You only need the essential sources for your mission. During the diagnosis we decide if a connection, a limited consultation or a controlled export is appropriate.

What happens if two platforms show different figures?

The agent does not automatically assume that one is wrong. First differences periods, definitions, publications, stories, adaptations, manual content and exceptions. If the discrepancy continues, present it with the sources for a person to resolve.

Can you check any social networks?

It can work with the channels and criteria that have been defined and tested. Actual coverage will depend on available sources, permissions, the functionalities of each platform and the scope of the project.

Does the radar automatically incorporate all the trends?

No. Radar is a filter, not a publishing channel. Each opportunity is contrasted with the strategy, the public, the resources, the timing and the risk before proposing to apply or test it.

What if our documentation is messy?

It is a usual situation. We start with the sources needed for this process, identify contradictions and document the minimum criteria. You don’t have to order the whole company before you start.

How long does it take to implement it?

It depends on the accessibility of the sources, the clarity of the process, the integrations, the permissions and the number of cases that need to be tested. After diagnosis we can propose phases, scope and schedule without presenting a generic duration as if it were valid for everyone.

How much?

The budget depends on the knowledge base to be prepared, the connections, volume and complexity of revisions, controls, training and follow-up. First we define a pilot and success criteria; then we prepare a proportional proposal.

How is the data protected?

We define sources, purposes, permissions, users, excluded information, conservation and incident procedures before connecting the process. The agent should not consult more data than necessary to fulfill their mission.

Can we start with a single customer or team?

Yeah. In fact, a delimited pilot is usually the best way to compare results, detect exceptions and learn before expanding the scope.

21. Closure and conversion

Which part of the review takes more time or concentrates more risk?

Tell us how you prepare your calendars and reports today, what sources you use, and where the incidents appear. We will help you to assess if the Karmina Review Agent is a good first driver or if it is convenient to start with another process.

You don’t need to come up with the defined technology, integrations, or agent. Let’s start with the real work.

Recommended fields of the form

• Name and surname.
• Company and position.
• Mail and phone.
• Number of accounts or brands you manage.
• Main calendar tools and reporting.
• Moment of the process you want to improve.
• Error, blocking or repetitive task more common.
• Person or team that would validate the pilot.
• Message or additional context.

Privacy Microcopy We will use this information only to understand your need and prepare the contact. You do not share customer data, credentials or personal, confidential or sensitive information in this form.

Message after shipment Oh, thanks. We will review the context and contact you to understand the current process before proposing any agents.

3. Alternative short blogs for Ads campaigns

These entries allow the hero to adapt to the intent of each campaign without changing the rest of the landing page.

Variant A. Quality Control

Label LESS ERRORS BEFORE THE CUSTOMER

H1 A second consistent review for social media calendars and reports

Subtitle The Karmina Review Agent compares data, strategy, content and conclusions to point out what needs to be corrected, checked or decided before shipping.

Variant B. Reporting

Label REPORTING ASSISTED WITH AI

H1 Convert scattered data into a more consistent and easy to validate report

Subtitle The agent reviews periods, variations, highlights and conclusions, and prepares a first executive layer for the team to spend more time interpreting and less time ordering.

Variant C. Calendar and Strategy

Label EDITORIAL CALENDAR AUDIT

H1 Check if the calendar actually applies the strategy before sharing it

Subtitle It detects gaps, repetitions, imbalances and pending recommendations with the own criteria of each account and a final human review.

8. Replication criteria for the next agent

When preparing the following landing page, you will not have to copy this text and change its name. It will be necessary to keep the order of decision and reconstruct the content from the actual card of the agent:

1. Mission and specific problem.
2. User and responsible person.
3. Entries and authorized sources.
4. Actions, tools and permissions.
5. Observable results.
6. Moments of human review.
7. Limits, risks and escalations.
8. Test criteria and success indicators.
9. Minimum requirements before the pilot.
10. Connections with Knowledge Foundation and other agents.

Karmina blocks, method, knowledge base, training, and contact can share architecture. The promise, examples, functions, process, requirements, limits and measure must be specific to each agent.