DEV Community

arafeq
arafeq

Posted on

We Built an AI Platform Where Every Analysis Is a Reusable Skill, Not a Throwaway Prompt

#ai #analytics #productivity #showdev

We built Reasio because we kept running into the same problem at every organization we worked with: expert knowledge was trapped in people's heads, and every AI tool on the market was too generic to actually help.

The senior analyst knows which Shopify metrics to trust. The operations manager knows which database tables have clean data. The marketing lead knows which Meta Ads columns lie. When any of them are on vacation, the team is stuck.

So we built an AI platform where every analysis is a reusable, auditable skill — not a disposable prompt.

Here's every aspect of what Reasio does and why we think it matters.

The Problem

Every organization we talked to had the same pattern:

  • Tribal knowledge trapped in people's heads — The person who knows how to pull accurate revenue numbers, which filters to apply, which edge cases to exclude... when they leave or go on vacation, that knowledge goes with them.

  • Repetitive analysis eating entire days — "What's our ROAS this week?" "Which products are trending?" "Are we seeing fraud?" Same questions, every week, rebuilt from scratch.

  • Generic AI tools producing generic answers — ChatGPT doesn't know your database schema. It doesn't know that your orders table has a financial_status column you need to filter on. It can't connect to your Shopify store or Meta Ads account.

  • Zero enterprise controls — No role-based access. No audit logs. No way to say "the marketing team can see campaign data but not financial data."

What Makes Reasio Different: Skills, Not Prompts

The core idea is simple: a skill is a structured module that encodes expert knowledge as executable code.

Each skill bundles:

  • TypeScript query scripts
  • Database schema references
  • Domain knowledge and best practices
  • Execution instructions for the AI agent

Here's what that looks like in practice:

// attribution-analysis skill
export const skill = {
  name: 'attribution-analysis',
  description: 'Cross-platform attribution: Meta Ads spend → Shopify revenue',
  schemas: ['shopify.orders', 'meta_ads.campaigns'],
  scripts: [
    'src/match-orders-to-campaigns.ts',
    'src/calculate-true-roas.ts',
    'src/identify-tracking-gaps.ts'
  ],
  bestPractices: [
    'Use UTM parameters for matching, not Meta-reported conversions',
    'Exclude refunded orders from revenue calculations',
    'Apply 7-day attribution window for accurate ROAS'
  ]
};
Enter fullscreen mode Exit fullscreen mode

When a user asks "What's my true ROAS for each Meta campaign this month?", the agent doesn't hallucinate — it selects the attribution-analysis skill, runs the TypeScript against your live Shopify and Meta Ads data, and returns verified numbers with interactive charts.

Time saved: 2 hours → 30 seconds.

20+ Pre-Built Skills Across Industries

We ship with skills for e-commerce, EV charging operations, and general automation:

E-commerce & Marketing (12 skills)

Skill What it does
shopify_analytics Orders, customers, products, inventory analytics
meta_ads_analytics Campaign performance, ROAS, ad spend tracking
attribution-analysis Cross-platform attribution (Meta → Shopify)
campaign-optimizer Budget reallocation recommendations
traffic-source-analysis Traffic mix breakdown across all channels
geographic-analysis Orders/revenue/ROAS by state and city
temporal-analysis Order patterns by time of day/week
abandoned-checkout-analysis Cart abandonment rates and recovery
fraud-detection Suspicious order patterns and blocklists
cod-performance COD completion rates and financial impact
product-performance Top sellers, underperformers, inventory turnover
anomaly-detection Statistical anomaly detection for key metrics

EV Charging Operations (8 skills)

Skill What it does
fleet-readiness Were all vehicles charged overnight? Ready for morning routes?
charger-health-monitoring Pre-calculated health checks (faults, connectivity)
locations_analysis Location utilization, energy delivery, real-time status
sessions_analysis Session lookups by ID/vehicle/location, error analysis
chargestations_analysis Individual charger status, OCPP messages, heartbeat monitoring
vehicles_analysis Vehicle charging history, SoC analysis, port usage
soc-analysis State-of-Charge timestamp queries
power-module-analysis Identify underperforming chargers by power output

And you can build your own with the skill-creator framework.

Multi-Agent Architecture

Reasio isn't one chatbot. You create multiple specialized agents, each with controlled access:

  • Marketing agent → has shopify_analytics + meta_ads_analytics + attribution-analysis
  • Operations agent → has fleet-readiness + charger-health-monitoring
  • Finance agent → has read-only database access + reporting skills

Each agent lives in its own workspace with isolated data permissions. The marketing intern can't accidentally query the financial database. The ops manager can't see campaign spend.

This is real RBAC for AI, not just "share a ChatGPT link."

Connect to Everything

Reasio connects to your actual data sources — no CSV uploads, no copy-pasting.

PostgreSQL · MySQL · Elasticsearch · Shopify · Meta Ads · Slack · GitHub · Stripe · Google Analytics · HubSpot · Jira · Notion · Any REST API · and hundreds more.

OAuth-based authentication handled for you. No custom code needed to connect.

Ask a question that spans multiple sources in a single conversation: "Compare my Shopify revenue against Meta Ads spend for the last 30 days and show me which campaigns are actually profitable."

Background Jobs & Workflow Automation

This is where it gets powerful. Reasio isn't just for one-off questions — you can schedule TypeScript scripts to run on cron schedules and take actions.

Real examples:

  • "Email me a revenue report every Monday at 8am"
  • "Alert Slack if cart abandonment rate > 75%"
  • "Sync Shopify orders to data warehouse every 15 minutes"
  • "Generate client performance reports and email to 10 clients every Friday"
  • "Check for fraud patterns every hour, send blocklist to security@company.com" 
// Example: Scheduled fraud monitoring
export default async function fraudMonitor() {
  const suspicious = await db('orders')
    .where('created_at', '>', oneDayAgo)
    .whereRaw('financial_status = ?', ['pending'])
    .groupBy('email')
    .havingRaw('count(*) > 3');

  if (suspicious.length > 0) {
    await sendEmail({
      to: 'fraud@company.com',
      subject: `${suspicious.length} suspicious orders detected`,
      body: formatBlocklist(suspicious)
    });
    await postToSlack('#fraud-alerts', formatAlert(suspicious));
  }
}
// Runs every hour: 0 * * * *
Enter fullscreen mode Exit fullscreen mode

Scripts can send emails, post to Slack, create tickets, update CRMs, trigger webhooks — anything you can do with a REST API. Every run keeps full stdout/stderr logs (50 run history per task).

Self-Contained Agents You Actually Control

Unlike generic AI assistants, every Reasio agent is self-contained with full control over:

  • Which skills it can access — restrict to only relevant analytical capabilities
  • Which data sources it connects to — no accidental cross-database queries
  • Who can use it — team-level permissions with role-based access
  • What it logs — comprehensive audit trail for every query and execution
  • Where it runs — sandboxed workspace isolation per conversation

Deploy an agent, hand it to your team, and know exactly what it can and can't do.

Interactive Visualizations

Every analysis returns Plotly-powered interactive charts — not static images. Time series, bar charts, scatter plots, heatmaps, geographic maps. Download as PNG or explore the interactive HTML.

Results stream in real-time via Server-Sent Events, so you watch the analysis unfold live. No page refreshes, no waiting for a spinner.

Real-World Example: E-commerce Multi-Source Attribution

A marketing manager needs to know which Meta campaigns are actually profitable:

User: "What's my true ROAS for each Meta campaign in the last 30 days?"

Reasio runs the attribution-analysis skill:

  • Connects to Meta Ads API for spend data
  • Connects to Shopify for actual revenue
  • Matches orders to campaigns via UTM parameters
  • Calculates true ROAS (not Meta-reported)

Result in 30 seconds:

Campaign 'Spring Sale' has 4.2x ROAS (best performer).
Campaign 'Retargeting' has 1.8x ROAS (below target).
Recommendation: Scale 'Spring Sale' by 50%, pause 'Retargeting'.

That analysis used to take 2 hours of manual data pulling across two platforms.

Try It

We're in early access. If you're running an e-commerce store, managing operations, or just drowning in repetitive data questions, I'd genuinely love your feedback.

Website: reasio.com
App: app.reasio.ai

Happy to answer any questions about the architecture, the skills system, or how we handle enterprise security. Drop a comment below or reach out directly.

Top comments (0)