Track Earnings Season on Autopilot With an AI Agent
On this page
Earnings season used to mean tabs. Lots of tabs. One for the calendar, one for each company’s investor relations page, one for the transcript, one for the analyst notes. I’d spend 20 minutes hunting down NVDA’s report time, another 30 reading through a transcript to find the one paragraph that actually mattered. Then do it again for MSFT. Then META.
So I built a workflow to handle it. An AI agent earnings tracker that wakes up every Sunday, scans the week’s calendar, posts the relevant companies to my phone, waits for me to pick which ones I want to follow, then schedules itself to fire after each report drops — searching for results, formatting a summary, and sending it to Telegram before I’ve had my second coffee.
The setup took about 15 minutes. The part that surprised me comes later — and it’s the reason I kept using it after the first week.
Why Earnings Season Breaks Manual Tracking
The problem isn’t that earnings data is hard to find. It’s that earnings season is relentless. Dozens of tech and AI companies report in overlapping windows — sometimes four or five in the same week. Missing a report date is easy. Forgetting which quarter GOOGL reports versus AMD is easier.
The deeper problem is what happens after you find the data. Research professionals spend nearly 60% of their time cleaning and organizing data, with another 19% just gathering it. For earnings specifically, that means reading through 30-page transcripts to extract five numbers that matter. That’s not analysis — that’s data janitorial work.
Professional tools exist for this. AlphaSense built a product that aggregates earnings call transcripts and benchmarks company AI outcomes using its Deep Research and Generative Grid tools. EarningsCall.ai positions its whole value proposition around saving the hours you’d spend reading transcripts manually. These are solid tools — built for equity analysts, portfolio managers, and M&A teams.
But if you want something that works on YOUR schedule, delivers to YOUR phone, and tracks exactly the companies YOU care about — you can build that yourself. For free. In 15 minutes. If you’ve been looking for a starting point for AI automation in your research workflow, this is one of the cleaner use cases I’ve seen.
How the Automated Earnings Tracker Works
The workflow runs in two phases: the Sunday preview and the post-report delivery.
Every Sunday at 6 PM, a scheduled job fires. The agent searches the upcoming earnings calendar, filters for the tech and AI companies on your watchlist — NVDA, MSFT, GOOGL, META, AMZN, TSLA, AMD, and whatever else you care about — then posts the list to a dedicated Telegram topic called ‘earnings.’ You reply with which ones you want to track that week.
That’s where it gets interesting. Once you confirm your picks, the agent schedules individual one-shot cron jobs for each company’s earnings date. Not a recurring job — a targeted job that fires once, after that specific report drops. Each one runs a web search for the results, formats a structured summary, and delivers it to Telegram.
The whole system runs on three things: the agent’s built-in web search capability, cron job support in OpenClaw, and a Telegram topic you set up as the delivery channel. No external APIs to configure. No dashboards to check. The information comes to you.
This is one of the cleaner examples of what agentic AI actually looks like in practice — not a chatbot you have to visit, but a system that takes action on a schedule and delivers results where you already spend your time.
The Part That Surprised Me: Memory
Here’s the thing I teased in the opening. The summary delivery is useful. The calendar scanning is useful. But the feature that made this genuinely sticky for me is the memory.
You can instruct the agent to remember which companies you tracked in previous weeks — and use that history to auto-suggest your watchlist the following Sunday. After three or four cycles, the agent stops presenting you with a generic list of 50 companies and starts leading with the eight you actually care about.
That’s a fundamentally different behavior than a search or a calendar reminder. A reminder tells you something is happening. An agent that remembers your preferences and adapts its behavior over time is doing something closer to what a good research assistant actually does — learning your focus areas so you don’t have to repeat yourself every week.
It’s a small thing. But after four weeks of it working correctly, I realized I’d stopped thinking about earnings season as a logistics problem. The logistics were handled.
What You Need Before You Start
The requirements are minimal. Here’s what you need:
- OpenClaw with web_search enabled — this is built-in, no additional configuration needed
- Cron job support in OpenClaw — used to schedule both the Sunday preview and the per-company post-report jobs
- A Telegram account with a topic called ‘earnings’ — this is your delivery channel; topics are available in Telegram groups, not direct messages
- A watchlist of companies — NVDA, MSFT, GOOGL, META, AMZN, TSLA, AMD are the typical tech/AI names to start with
That’s it. No paid data feeds. No brokerage API keys. No financial data subscriptions. The agent uses web search to find calendar data and earnings results from public sources.
Setting Up the Earnings Tracker
The setup is a single prompt. Do this once and the system runs itself from there.
Beacon’s got earnings season covered — no late nights, no missed reports, just steady light on every number that matters.
Step one: create a Telegram topic called ‘earnings’ in a group where your OpenClaw agent has posting permissions.
Step two: paste this prompt into OpenClaw:
Every Sunday at 6 PM, run a cron job to:
1. Search for the upcoming week's earnings calendar for tech and AI companies
2. Filter for companies I care about (NVDA, MSFT, GOOGL, META, AMZN, TSLA, AMD, etc.)
3. Post the list to my Telegram "earnings" topic
4. Wait for me to confirm which ones I want to track
When I reply with which companies to track:
1. Schedule one-shot cron jobs for each earnings date/time
2. After each report drops, search for earnings results
3. Format a summary including: beat/miss, revenue, EPS, key metrics,
AI-related highlights, guidance
4. Post to Telegram "earnings" topic
Keep a memory of which companies I typically track so you can
auto-suggest them each week.
Step three: on Sunday evening, reply to the agent’s Telegram message with your picks for the week. Something like: ‘Track NVDA, MSFT, META this week.’
The agent schedules the jobs. You do nothing else until the summaries arrive.
This is also a good companion to building a broader AI content pipeline — if you’re already having your agent generate research digests, the earnings summaries slot in naturally as a weekly input.
Where This Workflow Can Break
This setup works well for the common case. There are a few edge cases worth knowing about before you rely on it for anything time-sensitive.
- Earnings timing uncertainty — companies occasionally shift their report times (pre-market vs. after-hours) without much notice. The cron job fires based on the scheduled time, so if a report is delayed, the agent may search before results are available. The summary will reflect whatever is public at that moment.
- Web search quality variance — the agent uses public web search to find results. Immediately after a report drops, some sources are faster than others. If you’re watching high-volatility names, wait 30-45 minutes after the official report time before expecting a complete summary.
- Telegram topic permissions — if the agent loses posting permissions to your earnings topic, the Sunday preview silently fails. Check that bot permissions are intact when you first notice missing updates.
- Memory drift — the auto-suggest feature works well when your watchlist is stable. If you shift focus significantly (say, from large-caps to small-caps), explicitly update your watchlist in the prompt rather than relying on the agent to infer the change.
- No real-time data — this is not a trading tool. The summaries are delivered after reports are public, not during calls or pre-announcement. For options positioning around earnings, you still need a tool like Options AI’s earnings calendar, which surfaces expected move data from options market pricing.
How to Know It’s Working
- You receive a Telegram message in your ‘earnings’ topic every Sunday between 6 PM and 6:30 PM with a list of upcoming tech earnings
- After replying with your picks, you see a confirmation message from the agent (not just silence)
- On earnings days, a summary arrives in Telegram within 45-60 minutes of the report being public — it includes beat/miss, revenue, EPS, and at least one AI-specific callout if relevant
- By week three or four, the Sunday preview leads with your frequently-tracked companies rather than a generic alphabetical list
- The summaries include guidance figures, not just backward-looking metrics — if guidance is missing, the web search likely ran before the full release was indexed
Your Monday Morning Earnings Tracker Checklist
Here’s the sequence to get this running this week:
- Create a Telegram group (if you don’t have one) and add your OpenClaw agent as an admin with posting permissions. Then create a topic called ‘earnings’ inside that group.
- Paste the full prompt from the ‘Setting Up the Earnings Tracker’ section above into OpenClaw. Customize the watchlist to your actual companies — the default list (NVDA, MSFT, GOOGL, META, AMZN, TSLA, AMD) is a solid starting point.
- If today is before Sunday 6 PM, wait for the first automated preview. If you want to test immediately, ask the agent to ‘run the earnings preview now for the current week’ — it’ll execute the same job on demand.
- On the first Sunday, reply to the Telegram preview with your picks. Keep your initial list to 3-5 companies — this makes it easier to verify the summaries are arriving correctly before expanding.
- After your first earnings report lands, check the summary against the actual results. Verify it includes beat/miss, revenue, EPS, and guidance. If any field is missing, add an explicit instruction to the prompt: ‘Always include forward guidance even if the company only provides qualitative commentary.’
- After 3 weeks, check whether the Sunday preview is auto-suggesting your regular companies. If it’s not, add the line: ‘Prioritize companies I tracked in the previous 3 weeks in your weekly suggestions.’
- Optional: if you’re tracking names with significant options activity, cross-reference the expected move data from a tool like Options AI’s earnings calendar before each report to add context to the agent’s summary.
What This Means for Your Research Workflow
- The core setup requires three things: web_search capability in OpenClaw, cron job support, and a Telegram earnings topic. All three are available without external API configuration.
- The agent delivers two types of output: a Sunday preview you respond to, and automated post-report summaries covering beat/miss, revenue, EPS, AI highlights, and guidance.
- Persistent memory is the underrated feature — after a few weeks, the agent learns your watchlist and stops requiring you to rediscover it each Sunday.
- This is not a substitute for real-time data during earnings calls. Build in a 30-45 minute buffer after the official report time before expecting a complete web-sourced summary.
- The same pattern — scheduled job, user confirmation, one-shot follow-up jobs — applies to other recurring research workflows. Earnings is just one use case for this type of AI automation architecture.
Frequently Asked Questions
Do I need a paid data subscription to run this?
No. The agent uses its built-in web search to find earnings calendar data and post-report results from public sources. You don’t need Bloomberg, FactSet, or any paid data feed. The tradeoff is that the data arrives slightly later than premium sources — typically within 30-45 minutes of the public release.
What if I don't use Telegram?
Telegram is the delivery channel in this specific setup, but the same workflow logic can be adapted to other messaging platforms that your OpenClaw agent can post to. The core pattern — Sunday preview, user confirmation, one-shot post-report jobs — doesn’t depend on Telegram specifically. If you’re already using a different channel for AI digests, you can route the earnings summaries there instead.
How accurate are the AI-generated summaries?
Accuracy depends on two things: the quality of public sources the web search returns, and how quickly those sources are indexed after the report drops. For large-cap names like NVDA or MSFT, reliable summaries appear quickly because financial press covers them within minutes. For smaller names, give it more time. Always treat the summary as a starting point, not a replacement for reading the actual press release for material decisions.
Can I track non-US companies or international earnings?
You can add international names to your watchlist, but coverage quality will vary. The web search pulls from English-language sources, so US and major ADR-listed companies will have better coverage. European or Asian companies reporting in non-English may produce incomplete summaries. Test a few names manually before relying on them.
Is this available on BrainRoad?
BrainRoad is a personal AI agent hosting platform built on OpenClaw — which is exactly what this workflow runs on. If you don’t want to manage your own OpenClaw instance, BrainRoad gives you a hosted agent with web search and cron job support already configured. You bring the prompt; we handle the infrastructure.
