> ## Documentation Index > Fetch the complete documentation index at: https://docs.runspark.ai/llms.txt > Use this file to discover all available pages before exploring further. # Best Practices > Tips and strategies for getting the most out of Spark OS This guide distills the most effective patterns we've seen from teams getting real results with Spark. Whether you're just getting started or looking to level up, these tips will help. ## Setting Up Your Agent ### Give It a Real Identity Your agent performs better when it has a clear identity, not just a name. During setup, be specific: **Name:** Alex\ **Title:** Sales Development Rep\ **Personality:** Professional but friendly. Responds quickly and concisely. Always suggests a next step. Follows up persistently but never pushes too hard. **Name:** Assistant\ **Title:** AI\ **Personality:** Helpful A well-defined personality leads to consistent communication style across emails, chat, and task handling. ### Start with Knowledge, Not Prompts Before asking your agent to do complex tasks, invest time in your knowledge files. An agent with strong knowledge files and simple prompts outperforms one with weak knowledge files and elaborate prompts every time. **Priority order for new setups:** 1. Pricing guide (if applicable) 2. Product/service overview 3. Lead handling process 4. FAQ document 5. Email templates See [Writing Effective Knowledge Files](/guides/knowledge-files) for a deep dive. ### Write Instructions Like a Playbook Your agent's instructions should read like a playbook — clear rules for common situations: ```markdown theme={null} ## Lead Handling Rules - Always respond to new leads within 1 business day - Hot leads get a personalized email + follow-up task for same day - Warm leads get a template email + follow-up task for next day - Never quote custom pricing without checking with the team first - If a lead asks about a competitor, acknowledge it honestly and focus on our strengths — don't badmouth ``` ## CRM Management ### Keep Pipeline Stages Clean * Use 5–8 meaningful stages * Every stage = something changed * Review pipeline weekly * Move stale contacts to Lost or Archive * Create 15+ granular stages * Leave contacts in the same stage for months * Use stages as tags (that's what tags are for) * Skip stages — the data gets unreliable ### Use Tags for Segmentation Tags are lightweight labels. Use them for categories that don't fit in your pipeline: * **Source:** Website, Referral, Trade Show, Cold Outreach * **Industry:** Healthcare, Technology, Finance, Retail * **Interest:** Product A, Product B, Custom Solution * **Status:** VIP, Do Not Contact, Past Customer ### Keep Contacts Updated Stale data is worse than no data — your agent makes decisions based on what's in the CRM. Build habits: * Update contact notes after every interaction * Change pipeline stages promptly when something happens * Remove or archive dead leads quarterly * Use scheduled tasks to flag contacts that haven't been touched in X days ## Schedules That Work ### The Essential Three If you only set up three schedules, make it these: ``` Review the CRM and give me a morning briefing: 1. New leads that came in since yesterday 2. Tasks due today 3. Overdue tasks 4. Deals expected to close this week 5. Any contacts that haven't been touched in 7+ days Post the summary to chat. Keep it concise but actionable. ``` ``` Run a weekly pipeline review: 1. Total pipeline value vs. last week 2. Deals that moved stages this week 3. Deals that stalled (no activity in 7+ days) 4. New leads added this week 5. Win/loss count for the week Create a report document with the full analysis. Post a brief summary to chat. ``` ``` Review all contacts in Contacted and Qualified stages. For any contact whose last activity was more than 3 days ago, create a follow-up task for today with a note about what to follow up on based on their history. Don't create duplicate tasks — check if a follow-up task already exists before creating a new one. ``` ### Schedule Hygiene * **Be specific about output.** Always tell the schedule what to produce (post to chat, create a document, create tasks). If you don't, the agent might do the work but you won't see the results. * **Include duplicate prevention.** For schedules that create tasks or contacts, add "check if it already exists first" to avoid clutter. * **Review run history.** Check your schedule run history weekly to make sure things are working as expected. * **Start conservative.** Run schedules at low frequency first (daily, not hourly). Increase frequency once you've verified the output is useful. ## Working with Your Agent ### Teach as You Go Every correction is a training opportunity: > **You:** "That email was too formal. We're a casual brand — use first names, shorter sentences, and feel free to use exclamation marks." > > **Agent:** "Got it — I'll keep emails conversational and friendly going forward. I've noted this preference." Your agent logs these corrections and applies them in future interactions. ### Use the Work Tab When you need something more than a chat message — a report, a proposal, a detailed analysis — ask your agent to create a document: > "Create a quarterly business review document with pipeline analysis, win/loss data, and top accounts" This goes to the [Work tab](/agent/work-tab) where you can review, edit, and share it. Much better than scrolling through chat history to find that one time your agent gave you a breakdown. ### Give Feedback Explicitly Your agent won't know it got something wrong unless you say so: * ✅ "That quote was wrong — we don't charge shipping on orders over \$500. Update your knowledge." * ✅ "Great email draft, but remove the last paragraph — we never promise timelines." * ❌ Silently fixing the output yourself (the agent learns nothing) ## Integrations ### Connect Early Set up integrations before you start working with your agent intensively: 1. **Gmail** — so your agent can send and read emails 2. **Google Calendar** — so it can check availability and create events 3. **Webhooks** — so leads flow in automatically 4. **Phone** — if you want calling capability An agent with integrations is dramatically more useful than one limited to chat. ### Test Webhooks Manually Before connecting your production forms, test your webhook with sample data: 1. Set up the webhook URL in Spark 2. Send a test payload using a tool like curl or Postman 3. Verify your agent processes it correctly 4. Then connect your real form See [Webhooks](/integrations/webhooks) for details. ## Team Usage ### Define Clear Ownership When you have team members, make sure it's clear: * Who follows up with which leads (assign contacts) * Who reviews which pipeline stages * What your agent handles vs. what humans handle * When to escalate to a person vs. let the agent continue ### Use Views for Different Roles Create saved views tailored to each role: * **Sales rep:** "My Active Leads" — table filtered to their assigned contacts * **Manager:** "Pipeline Overview" — dashboard with team metrics * **Everyone:** "Overdue Tasks" — filtered task list ## Common Pitfalls **Avoid these common mistakes that reduce your agent's effectiveness:** | Pitfall | Why It Hurts | Fix | | -------------------------- | ----------------------------------- | -------------------------------------------- | | No knowledge files | Agent guesses instead of knowing | Upload your core docs first | | Vague schedule prompts | Agent does work you can't see | Always specify what to output | | Never updating the CRM | Agent uses stale data for decisions | Build update habits, use automated flags | | Too many pipeline stages | Contacts get stuck in limbo | Keep it to 5–8 meaningful stages | | No follow-up tasks | Leads go cold | Use schedules to auto-generate tasks | | Ignoring agent corrections | Same mistakes repeat | Give explicit feedback when things are wrong | ## The 30-Day Ramp If you're new to Spark, here's a realistic timeline: | Week | Focus | | ---------- | ----------------------------------------------------------------------------------------------------------------------------- | | **Week 1** | Set up workspace, upload 3–5 knowledge files, connect integrations, configure pipeline stages | | **Week 2** | Start chatting daily, correct mistakes, add more knowledge files, set up first 2–3 schedules | | **Week 3** | Connect lead capture (webhooks), set up automations, create dashboard, refine agent instructions based on what you've learned | | **Week 4** | Review everything — pipeline accuracy, schedule output quality, knowledge gaps. Optimize and fill holes | By the end of month one, your agent should be handling lead processing, daily reports, and follow-up management with minimal intervention.