Scaling

Agencies lose margin when every client is treated as a new project. Productizing means deciding what the service is, what it includes and how it is delivered, so that it can be repeated. Define the client profile, package scope, onboarding, monitoring, action plan, reporting and review cadence.

Write the stages and who does what:

1. Sales and qualification.
2. Onboarding, question set and competitors.
3. Baseline and diagnosis.
4. Proposal and scope confirmation.
5. Monthly monitoring and analysis.
6. Recommendations, tasks and execution.
7. Reporting and client review.
8. Renewal and expansion.

For each stage, define an input, an owner, an output and a quality check.

Track the hours for each stage and for each client. Start with estimates, then replace them with real data after a few months. Separate one-time onboarding from recurring work, and note where client delays or custom requests add time.

For each client, calculate:

• Revenue per month.
• CiteMind usage cost from questions, assistants and days.
• Hours multiplied by internal cost.
• Gross margin.
• Hours of account-management and delivery capacity used.

A client can have a good price and still be unprofitable if the scope is uncontrolled.

Build a library so you do not start from zero:

• Question-set patterns by vertical.
• Onboarding checklist.
• Audit and proposal templates.
• Recommendation and task formats.
• Monthly report and meeting agenda.
• Renewal summary.

Keep room for client-specific judgment where it matters.

How many clients one person can manage depends on package scope, client responsiveness and how much work is standardized. Work it out from your own time data: available hours divided by hours per client. Separate analysis, execution, account management and quality review as volume grows.

• Revenue and gross margin per client.
• Tooling cost per client.
• Hours per client by stage.
• Recommendations completed.
• Renewal and expansion rate.
• Time from sale to first report.
• Scope changes and unbilled work.

• Selling unlimited scope.
• Customizing every process.
• Ignoring hours because the client pays a retainer.
• Growing before the quality check is clear.
• Automating a broken process.
• Keeping unprofitable accounts without a plan to fix them.

MCP, the Model Context Protocol, is a standard way for AI assistants and automation tools to connect to other systems through a defined set of tools. Instead of building a custom integration for each assistant or workflow, you connect once and any compatible client can use it.

Good candidates repeat often, follow clear rules and are easy to check:

• Refreshing data and preparing summaries.
• Watching for visibility or sentiment drops.
• Creating a task when a recommendation is ready.
• Updating task status from a workflow.
• Preparing a report draft from approved data.
• Sending an internal alert when a threshold is met.

Keep a person in the loop for anything that needs judgment or reaches a client:

• Choosing the strategy.
• Approving claims about the client.
• Publishing content or editing important pages.
• Sending a report or sales message.
• Deciding whether a result proves that an action worked.
• Acting on incomplete or unexpected data.

Each workflow has a trigger, an action and a human check.

Visibility drop: a new run crosses the agreed alert condition -> collect affected questions and assistants -> analyst checks whether the movement is real.

Recommendation to task: a recommendation is approved -> create a task with the question, finding, owner and due date -> strategist reviews the task.

Monthly report draft: the reporting date arrives -> collect approved metrics, actions and notes -> analyst checks every number and interpretation before delivery.

Renewal preparation: a term approaches its review date -> gather baseline, progress and open actions -> account manager writes the renewal summary.

• Use least-privilege access and separate test accounts where possible.
• Log every automated action.
• Set clear thresholds and stop conditions.
• Keep client data separated.
• Require human approval for external messages and publishing.
• Review outputs for facts, tone and scope.
• Revisit workflows when the product or process changes.