Key takeaways
- The Bloomberg-like Dashboard is a custom tool that tracks 6 assets (BTC, ETH, SOL, NVDA, TTE, CW8) and pushes 3 automatic briefs a day to Telegram, at 8:50 am, 1 pm and 5:30 pm Paris time.
- The briefs are written by Claude Haiku 4.5, Anthropic's small model, priced at $1 per million input tokens and $5 per million output tokens according to the CloudZero guide updated on August 20, 2026; a brief of a few hundred words costs a fraction of a cent.
- 29% of employees use the business intelligence tools their employer pays for, according to Gartner figures compiled by Business Analysis in 2026; the dashboard nobody opens is the norm, not the exception.
- The same setup transposes to the numbers of a small business: code collects and computes, a small model writes, and the brief goes out at the times when decisions are made.
The problem: a dashboard only works if someone opens it
The starting point is personal. I follow six assets across crypto, stocks and ETFs: BTC, ETH, SOL, NVDA, TotalEnergies and a world ETF. Following them seriously meant several browser tabs open at all times, indicators recalculated by hand, and decisions made in the heat of the moment, which is the worst time to decide.
Nothing about this problem is specific to markets. It is exactly the problem of a company's numbers. According to Gartner data compiled by Business Analysis in 2026, 29% of employees actually use the business intelligence tools their employer funds, and only 16% of organizations reach full adoption of their Power BI dashboards. The same research notes that 67% of business leaders do not fully trust the data they base decisions on.
In other words, the most likely fate of a dashboard, in a small business or anywhere else, is to go unopened. Mine solved that problem in a simple way: it does not wait for me to open it.
What was built: collect, compute, write, send
The tool, documented on my projects page, chains four steps, three times a day, at the times that matter for markets: 8:50 am before the European open, 1 pm at midday, 5:30 pm after the close.
- Collect. The system pulls prices for the six assets from market data sources.
- Compute. Code calculates the technical indicators; no AI is involved at this stage, because an indicator is a formula, not an opinion.
- Write. Claude Haiku 4.5 receives the numbers and the pre-computed indicators, and writes a brief: what moved, what deserves attention, what can wait.
- Send. The brief goes out on Telegram, with a three-level alert system depending on how urgent things are.
Around this core, the tool includes a scenario simulator, tracking for scheduled investments, a market events calendar and a trading journal. But the daily value lies in the ritual: three messages a day, at fixed times, read in one minute.
Why is a small model enough?
Because the division of labor is strict: code computes, the model writes. Claude Haiku 4.5 decides nothing and calculates nothing; it receives exact figures, produced by deterministic code, and turns them into three readable paragraphs. For this constrained writing task, the smallest model in the lineup does the same job as the largest.
The consequence is economic. According to the CloudZero pricing guide updated on August 20, 2026, Claude Haiku 4.5 costs $1 per million input tokens and $5 per million output tokens, against $5 and $25 for Claude Opus 5. At that rate, a brief of a few hundred words is a fraction of a cent; three briefs a day for a month stays below the price of a coffee. The running cost of the AI has become a non-issue; what you pay for is building the chain around it.
This is the point that spectacular AI demos make people forget: in a tool that runs every day, the model is the smallest part. Reliability comes from everything else, the data collection that does not break, the verifiable calculations, the scheduling, the logging.
What does this change for the numbers of a small business?
Replace the six assets with your numbers: yesterday's sales, cash collected, pending quotes, open support tickets, stock levels. The chain is identical. Code queries your tools, computes the gaps that matter to you, a small model writes the brief, and the message lands before your morning meeting. Nobody has a dashboard to open or a filter to set.
One important nuance: a pushed brief does not replace an exploration tool. The table below sums up how the options split for a small business.
| Option | Indicative cost | Strength | Limit |
|---|---|---|---|
| Hand-maintained spreadsheet | $0 in licenses, hours every week | Total flexibility | Depends on one person, rarely up to date |
| BI platform (Power BI, Looker Studio) | Free to a few dozen dollars per user per month | Free-form data exploration | 29% actual usage by employees |
| Custom AI brief, pushed at fixed times | Built once, quoted per project (Sprint from €400); marginal inference cost | Information arrives effortlessly, at chosen times | Does not replace on-demand exploration |
The last two rows are not rivals; they complement each other. The BI platform answers the questions you ask; the brief answers the one you never think to ask each morning: what happened that matters since yesterday?
When this kind of tool is the wrong choice
Three situations where I advise against ordering an automated brief:
- Your current dashboard actually gets opened every day. If your team is part of the 29% that uses its BI tool, the problem the brief solves does not exist for you. Keep your tool.
- Your underlying data is not reliable. An AI writing from wrong numbers produces wrong briefs, confidently. If you are among the 67% of leaders who doubt their data, the first project is fixing the sources, not adding a writing layer.
- Your decisions are monthly. A daily brief on a slow-moving business manufactures noise. A well-built monthly export, or simply a meeting with the right numbers, is enough.
The setup wins in one specific case: numbers that move every day, scattered across several tools, and a recurring decision that would benefit from being made in the morning rather than in reaction. It is the same reasoning as for KPI dashboards built without a BI platform: the tool follows the decision, never the other way around.
Frequently asked questions
How much does the AI cost in a dashboard like this?
Claude Haiku 4.5 is priced at $1 per million input tokens and $5 per million output tokens, according to the CloudZero pricing guide updated on August 20, 2026. At that rate, a brief of a few hundred words costs a fraction of a cent; the real cost is building the tool, not running it.
Does an automated brief replace a business intelligence tool?
No. A BI tool like Power BI or Looker Studio is for exploring data when you have a question. The automated brief covers the other half of the need: pushing the essentials at set times, without anyone having to open anything. The two complement each other, and a company that already opens its BI tool every day may not need a brief at all.
Do you need perfect data before plugging an AI into it?
You need reliable data on the scope of the brief, not perfect data everywhere. If the underlying numbers are wrong or contradictory, the brief will be wrong too; 67% of business leaders say they do not fully trust the data they base decisions on. In that case, the first project is fixing the data, not adding AI.
The takeaway. The most sophisticated dashboard loses to three paragraphs that show up on their own at 8:50 am. Before investing in yet another tool, ask the reverse question: what piece of information, delivered at what time, would change a decision you already make? If the answer is clear, the chain that produces it can be built.
If you want to see what a daily brief would look like on your numbers, the simplest way is to talk it through: book a 30-minute call, and see how I build this kind of chain on the AI agent developer page. At minimum, you will leave with an honest opinion on the best option for your case, including when that option is not me.