In This Article: What makes an effective prompt, the four-element prompt anatomy with finance-specific examples, how to use the Prompt Library, and how to create, configure, apply, and manage Personas in Smith — including the Optimize with Smith workflow, the full 12-field persona anatomy, activation via the chat input bar, mid-conversation switching, managing personas from the detail panel, and best practices for crafting effective personas.
The Prompt Library
A prompt is the instruction you give Smith — the text that tells it what to do, how to reason, and what kind of output to produce. The quality of Smith's response is directly proportional to the quality of the prompt. A vague question gets a generic answer; a well-structured prompt with clear context and defined output gets analysis you can actually use.
Smith includes a built-in library of pre-written prompts designed for common financial analysis, reporting, and business intelligence tasks. These are ready to use immediately — no setup required. To access them, click the PROMPT LIBRARY tab in the Smith navigation bar. Prompts appear as browsable cards; clicking any card automatically populates the chat input field so you can review, adjust, and send.
Featured Prompts on the Custom Tab
A curated selection of prompts from the library also appears as cards on the CUSTOM tab homepage, visible before you start a new conversation. These are the same pre-built prompts — clicking them works identically to clicking from the PROMPT LIBRARY tab.
What Makes a Good Prompt
Even though custom prompt creation is coming soon, understanding prompt anatomy now helps you get more out of the pre-built library — and prepares you to write sharper prompts the moment custom creation launches. Every strong prompt answers four questions: who is Smith acting as, what should it do, what data should it use, and how should it format the output.
Prompt Tips for Finance and Operations Teams
The pre-built prompts in the library already follow good prompt structure. When you modify them — or type your own in the chat input — these habits consistently improve output quality.
Understanding Personas
A persona is a customizable AI configuration that tells Smith how to interpret your queries — defining its role, expertise, tone, and reasoning approach. When a persona is active, it shapes every response automatically, without you needing to re-explain context on each query.
Personas are most valuable when you return to the same domain repeatedly. A "Budget Variance Analyst" persona calibrates Smith's reasoning to financial measurement and corrective action framing. A "Technical Documentation Writer" persona shifts it toward structured, audience-aware prose. The persona does that work so you can focus on the question.
Anatomy of a Persona
The Build Your Persona modal contains 12 fields that work together to give Smith a complete picture of the role it should inhabit. The modal is scrollable and two-column. All fields beyond the first three are optional — the more you provide, the more consistent Smith's output will be.
Creating a Persona
Smith's Optimize with Smith feature lets you build a fully populated persona from just three starting fields. The recommended workflow is to enter the minimum required information, let Smith generate the remaining fields, then review and refine before saving.
Step 1: Open the PERSONAS Tab
In the Smith navigation bar, click the PERSONAS tab. If you haven't created any personas yet, the center of the screen shows a Create Personas button. Click it to open the Build Your Persona modal.
Step 2: Enter the Three Required Fields
Fill in Persona name, Professional/Personal title, and Persona primary goal. These three fields give Smith enough context to generate a complete, coherent persona definition. Entering more detail at this stage produces a more accurate first draft.
Step 3: Click "Optimize with Smith"
The Optimize with Smith button is located in the top right of the modal. Clicking it sends your three seed fields to Smith, which generates content for all remaining fields automatically. Review the generated output field by field — adjust any language that doesn't accurately reflect your intended persona before saving.
Building Manually
You can also fill in any combination of fields manually without using Optimize. All fields are optional beyond the minimum three. A persona with more complete fields produces more consistent, nuanced output — but even a persona with just a name and primary goal provides useful framing.
Sharing Permissions COMING SOON
Currently all personas are Personal — visible only to you. When the sharing feature launches, you will be able to mark a persona as Shared, making it accessible to everyone in your organization. Shared personas enable team-wide standardization of AI interaction patterns across departments.
Applying a Persona to a Conversation
Clicking a persona card in the PERSONAS tab opens its detail panel for management — it does not activate the persona for your conversation. To apply a persona, use the persona dropdown in the chat input bar.
Switching Personas Mid-Conversation
You can switch personas at any point during a conversation — simply open the Persona dropdown and select a different persona. The new persona applies from that query forward. Previous responses in the thread are not affected.
This is useful when a single conversation spans multiple domains. For example, you might start with a financial analysis persona to interpret a report, then switch to a content writing persona to draft a summary for a non-technical audience.
Managing Personas
All persona management actions — Edit, Copy, and Delete — are accessed from the persona detail panel. Click any persona card in the PERSONAS tab to open it.
Version Control Recommendation
Smith does not maintain a version history for personas. Before making significant changes to a persona you use regularly, use Copy to create a backup with a new name (e.g., "CFO Advisor — v2"). This preserves the original while you iterate on the new version.
Personas vs. Prompt Library
Both tools customize Smith's output, but they operate at different levels and serve distinct purposes. Understanding the distinction helps you use them together effectively.
- Persistent context — Remains active across all queries until deselected
- Role-based framing — Defines how Smith thinks and reasons
- Broad scope — Affects every query while active
- Best for: Ongoing work in a specific domain or role
- One-time use — Populates the input field for a single query
- Task-specific — Designed for a particular action or output type
- Quick access — Click to populate, edit if needed, send
- Best for: Recurring tasks that don't need persistent role framing
Using Personas and Prompts Together
The most effective Smith interactions layer both tools. Activate a persona to establish the role and reasoning frame, then use a prompt library entry to execute a specific task within that context.
Personas and Lakehouse Files
Personas become significantly more powerful when combined with Lakehouse files selected as context. A persona frames how Smith reasons; Lakehouse files provide the ground-truth data it reasons against. Together, they produce domain-specific analysis grounded in your actual documents.
Common Persona Use Cases
Personas are versatile and can be adapted to a wide range of professional functions. Click any use case below to see example field values.
Persona name: "Budget Variance Analyst"
Primary goal: Explain budget deviations and recommend corrective actions
Use case: Analyzing monthly departmental spend against forecasts, identifying anomalies, and suggesting reallocation strategies. Pair with your financial reporting files from the Lakehouse for grounded variance analysis.
Persona name: "Month-End Close Coordinator"
Primary goal: Accelerate close by identifying outstanding items, flagging inconsistencies, and producing review-ready summaries
Use case: Attach your actuals and prior-period reports from the Lakehouse. Ask Smith to flag journal entries that fall outside normal ranges, summarize outstanding reconciling items by account, or draft the close commentary for leadership review. This persona keeps Smith anchored in structured accounting logic rather than general financial reasoning.
Persona name: "Technical Documentation Specialist"
Primary goal: Produce precise, structured documentation for technical audiences
Use case: Drafting API guides, system specifications, runbooks, and data dictionaries. Define the intended reader (developers, architects, data engineers) in the Tone & Style field to calibrate depth and vocabulary.
Persona name: "Compliance Risk Advisor"
Primary goal: Identify regulatory risks and flag areas requiring legal review
Use case: Reviewing contracts for liability clauses, GDPR considerations, or regulatory gaps. Use guardrails in the Tone & Style field such as "Always flag areas requiring formal legal counsel" to keep outputs appropriately scoped.
Persona name: "Executive Summary Generator"
Primary goal: Distill complex analysis into concise, decision-ready executive summaries
Use case: Compressing detailed financial, operational, or strategic documents into board-appropriate summaries with clear recommendations and ROI framing. Define "Limit to 3 key insights with recommended action" in the Tone & Style field.
Persona name: "Profitability Analyst"
Primary goal: Identify margin drivers, surface declining profitability trends, and recommend corrective actions by product line or customer segment
Use case: CFOs and controllers use this persona to cut through revenue aggregates and get SKU-level or segment-level clarity. Attach your product margin files or customer P&L data from the Lakehouse, then ask Smith to rank segments by gross margin erosion, model the impact of a price change, or identify customers with declining profitability trends. Define guardrails in Tone & Style: "Present findings as a ranked table. Flag any data gaps explicitly."
Best Practices for Effective Personas
A well-built persona balances specificity with flexibility. Vague instructions produce generic output; overly rigid instructions can cause Smith to miss useful adjacent reasoning. These guidelines produce consistently high-quality results.
| BEST PRACTICE | WHY IT MATTERS |
|---|---|
| Be specific about expertise | "Financial analyst" is vague. "Financial analyst specializing in SaaS revenue recognition under ASC 606" gives Smith a precise lens. Specificity drives actionable output. |
| Define output format | Specify how you want information structured: bullet points, tables, narrative paragraphs, or executive summaries. Without format guidance, Smith makes its own judgment — which may not match your use case. |
| Set tone and audience | "Explain for technical audiences" vs. "Simplify for non-technical executives" produces fundamentally different outputs from the same underlying data. |
| Include guardrails | Add constraints like "Always cite ASC standards" or "Limit responses to 300 words." Guardrails prevent off-topic or overly verbose responses without narrowing the persona's reasoning too much. |
| Use Optimize as a starting point | Optimize with Smith generates a solid draft persona quickly. Treat the output as a first draft to refine, not a final configuration. Small adjustments after testing significantly improve consistency. |
| Test and iterate | Run the persona with several representative queries before relying on it for high-stakes work. Personas rarely perform perfectly on the first attempt — minor prompt adjustments often yield significant improvements. |
Troubleshooting
If a persona isn't producing the expected results, review these common issues and fixes.
| ISSUE | SOLUTION |
|---|---|
| Responses are too generic and don't reflect the persona | Add more specificity to the Context & Intent and Tone & Style fields. Include explicit expertise areas, output format requirements, and domain constraints. |
| Persona provides incorrect or irrelevant information | Ensure the right Lakehouse files are selected as context in the My Lakehouse Files panel. Also review the persona for unrealistic scope (e.g., expecting real-time data from static uploads). |
| Responses are overly verbose or off-topic | Add guardrails to the Tone & Style field: "Limit responses to 200 words" or "Focus only on [domain]; exclude operational details." |
| Prompt library entry doesn't match expected output | Click the prompt card to populate the input field, then modify the text before sending. Pre-built prompts are starting points — edit them to fit your specific situation. |
| Output is technically correct but feels too generic | Combine a persona (for role framing) with a prompt library entry (for task specificity). Layering both tools produces the most precisely targeted output. |
The most effective Smith users build a small library of well-crafted personas for their most common domains — financial analysis, executive reporting, compliance review — then layer in prompt library entries for specific tasks. Personas carry the context so your prompts can stay concise and task-focused. If you need help getting started, use the Support widget in the bottom-right corner of Compass.
Comments
0 comments
Article is closed for comments.