Databook Coach Editor Guide

The Coach Editor is built around a clone and edit workflow. Databook provides base coaches — pre-built, verified agents that have been tested end-to-end for specific sales motions (e.g., prospecting, opportunity sizing). Rather than building a coach from scratch, you clone a base coach and customize it to fit your organization's needs.

Here's what that means in practice:

This approach gives you the freedom to experiment without disrupting anything that's already working.

When you open a cloned coach in the editor, you'll see several sections. Some are editable, some are locked. Here's what each one does.

The display name of the coach. Rename it to something that reflects the customizations you've made — for example, changing "Prospecting Sequence Coach" to "Enterprise Prospecting Coach" if you've tailored it for enterprise accounts.

The user-facing description that appears when someone views the coach. Update this to describe what your customized version does, so your team understands what they're getting when they launch it.

The trigger field is set to Manual and is locked. This means the coach starts when a user clicks the button to launch it — it doesn't run on its own.

In the future, coaches will support automated triggers such as:

Automated coaches will be able to run research, generate output, and push results to systems like your CRM, Slack, or email — without the user initiating anything. That capability is on the roadmap but not available today. For now, all coaches are manually triggered.

The type field is set to Sequential and is locked. This means the coach moves through its tasks in a defined order — one step at a time, from start to finish. The user progresses through the flow linearly.

The purpose field describes what the coach is designed to do at a fundamental level. This field is locked intentionally.

The clone-and-edit model is designed to let you customize a verified use case — one that's been built and tested to work end-to-end. Changing the purpose would fundamentally change what the agent is trying to accomplish, which defeats the benefit of starting from a proven foundation. The goal is to tailor how the coach executes its purpose, not to redefine the purpose itself.

Tasks are the core of the coach. Think of them as the steps the coach follows to get from start to finish. For example, a prospecting coach might have tasks like:

Each task has its own anatomy, covered in the next section.

The output section controls what the final deliverable looks like. Right now, output is limited to text (e.g., email copy, research briefs, call prep notes). Future formats will include PDFs and PowerPoints.

Within the output section, you can manage:

The output section is where many teams will want to make edits to match their organization's brand, voice, and outreach style.

Each task inside a coach has four components.

A short label that describes what the task does at a high level (e.g., "Company research and signal detection"). This is what you see when scanning the task list in the editor. You can rename it to anything that makes sense for your team.

An auto-generated summary based on the title and instructions. It gives a more detailed explanation of what the task does, so someone reviewing the coach in the editor can quickly understand the flow without reading the full instructions.

This is where the prompt engineering happens — the heart of each task. The instructions tell the coach:

A good way to think about writing instructions is to follow a logical business flow. For example, a "Target Account" task might work like this:

The purpose of this task is to correctly identify the target company. After resolving the company, confirm with the user that this is the account they intended — since many companies share similar names across the Databook database and web search results.

That's a simple, clear instruction set. You're defining the job (identify the account), the action (look it up and resolve it), and the outcome (confirm with the user before moving on).

Tools define the scope of information available to a task. They control what data sources and capabilities the coach can access when executing that step.

Some common tools you'll see:

You'll see how tools are referenced within the instructions of each task. The instructions tell the coach when and how to use a given tool. Don't worry about tools too much when getting started — the biggest gains come from editing the instructions themselves, not from changing the tool list.

Here's a summary of what's available to change and what's locked.

Edit existing task instructions — This is the highest-value, lowest-risk change. You're refining what a task does without changing the overall flow. Even small prompt additions can produce meaningful differences in the output.

Edit the output section — If your organization has a specific brand voice, email style, or messaging framework, this is where you encode it. Add examples of what great outreach looks like for your team.

Add new tasks — This is more advanced and adds complexity, but it's fully supported. A new task can leverage all the research and context from previous steps without re-doing any work.

Every time you save changes, a version is recorded. If an edit doesn't work out, you can:

This is your safety net. Edit freely knowing you can always roll back.

The Preview function lets you test the coach as a user would experience it. After making changes in the editor:

Use this after every meaningful edit to verify the flow still works the way you expect.

When you're ready to publish your customized coach:

When a custom coach goes live, the base coach is hidden and your customized version becomes the one your team sees and uses.

The Coach Editor has several built-in safeguards to keep things running smoothly.

Every time you edit a task's instructions and save, the content is run through a security validator. This checks for:

This happens automatically — you don't need to do anything. If the validator flags an issue, you'll be notified before the save completes.

When you set a coach live, the system runs a full evaluation to make sure the coach works from start to finish. This catches issues like broken task transitions, tool failures, or prompts that produce unusable output.

If the evaluation returns an error but you believe the coach is still functional (e.g., the error was a transient issue or a non-critical edge case), you have the option to override and publish anyway. Use this with judgment — it's there for flexibility, not as a default.

As covered above, every saved version is preserved. If a published coach has issues, you can restore a previous version at any time.

Before touching the instructions, think about what you want the task to accomplish from a business perspective. What should the seller walk away with after this step? Once you know the answer, the prompt writes itself.

You don't need to rewrite entire tasks. Adding 3-4 lines to an existing instruction set — like asking the coach to include competitive context in its research output — can produce a whole new section in the deliverable.

The most effective editing workflow is: edit → save → preview → adjust. Make a change, test it, see what happens, and refine. The preview function is there to be used frequently.

The output section is where you'll likely get the most immediate value. If your team has a specific way they write emails, structure LinkedIn messages, or frame value propositions — put examples in the output section. The AI performs significantly better when it has concrete examples of what "good" looks like for your organization.

Make one change at a time and preview it. This makes it easy to identify what's working and what isn't. If you change five things at once and something breaks, it's harder to pinpoint the cause — though you can always restore from history if needed.