How to Read the Dashboard

Who Is This For?

This guide is for non-technical readers — founders, finance leads, and product managers who need to understand AI agent costs without reading code. Each section explains:

1. What you're looking at — plain English, not technical jargon
2. Why it matters — the business implication
3. What to do — actionable next steps

If you're a developer looking for API documentation, see the API Reference instead.

Overview — Your Total AI Picture

The Overview page shows your aggregate AI spend across all customers and all workflows.

What you're looking at:

• Total Spend — how much you've spent on AI this month (or selected period)
• Task Count — how many AI tasks ran in that period
• Avg Cost per Task — Total Spend ÷ Task Count
• Cost Efficiency — what percentage of spend went to productive calls vs. retries

Why it matters: This is your top-line number. If this number surprises you (higher than expected), you need to dig into Cost Health to understand where the waste is.

What to do:

• If Total Spend is higher than last month, check Cost Health for retry waste spikes
• If Avg Cost per Task is climbing, you may have expensive workflows eating into margins
• Use the date filters (7/30/90 days) to spot trends

Cost Health — Where Is My Money Going?

Cost Health breaks down your spend into productive work vs. waste.

What you're looking at:

• Retry Cost — money spent on failed AI calls that had to be retried
• Failure Cost — money spent on calls that failed completely
• Retry % of Spend — what percentage of your total AI bill is retry waste

The bar chart shows retry costs grouped by reason:

• rate_limit — you hit the API's speed limits
• timeout — the API took too long to respond
• parse_error — the model output couldn't be parsed, causing a retry
• Other reasons the system detected

Why it matters: A 6-8% retry rate is normal. If you're above 15%, you're burning money. The reasons tell you where to focus engineering effort.

What to do:

• If rate_limit is high: consider batching requests or upgrading your API tier
• If timeout is high: check if you're using the right model for the task
• If parse_error is high: your prompts may need work, or you need longer timeouts

Cost Intelligence — Which Customers Are Profitable?

Cost Intelligence shows you which customers cost the most — and helps you understand profitability.

What you're looking at:

• Total AI Spend — your total AI bill for the period
• Top Customer (% of total) — which customer drives the most spend
• Avg Cost / Customer — Total Spend ÷ number of active customers
• Avg Cost / Task — Total Spend ÷ number of tasks

The Pareto chart shows your customers ranked by cost, with a line showing cumulative percentage. Typically, 20% of customers drive 80% of costs.

The table shows each customer's:

• Total cost
• Number of tasks
• Average cost per task
• Retry percentage (high retry % = problem customer)

Why it matters: If Customer X pays you $500/month but generates $800/month in AI costs, they're losing you $300/month. You need to know this to make pricing decisions.

What to do:

• Look for customers with high retry % — they may need prompt optimization or a different AI strategy
• Compare avg cost/task across customers — similar customers should have similar costs
• If one customer dominates spend, you're concentrated on a small number of accounts

Note: Profitability requires revenue data. If you haven't imported your pricing/subscription data, the profitability numbers will be incomplete. See Settings to import revenue.

Customers — Per-Customer Breakdown

What you're looking at: A list of every customer with their AI costs, broken down by:

• Total cost (LLM + external services + compute)
• LLM cost specifically
• Non-LLM cost (API calls, vector DB, etc.)
• Retry cost
• Task count

Click any customer to see their task history and cost breakdown by task type.

Why it matters: You can see at a glance which customers are expensive to serve — and whether expensive correlates with valuable (based on what they pay you).

What to do:

• Review customers with retry % above 10% — something is wrong
• Compare task counts: a customer with 10x the tasks but 3x the cost is more efficient
• Use this to have informed conversations with customers about their AI usage

Tasks — Which Workflows Are Most Expensive?

What you're looking at: A breakdown of costs by task type (workflow):

• resolve_ticket — support ticket resolution
• generate_report — report generation
• classify_lead — lead classification
• etc.

Each row shows:

• Total cost for this task type
• How many times it ran (frequency)
• Average cost per run
• Retry lineage (if retries are common for this task type)

Why it matters: Some workflows are inherently expensive. If resolve_ticket costs $12/call and runs 1,000 times/month, that's $12,000/month. If you're pricing by seat or subscription, you need to know this.

What to do:

• Find your top 3 most expensive task types
• Ask: is the cost justified by the value? Can it be optimized?
• If a task has high retry cost, focus engineering on that specific workflow

Anomalies — When Costs Spike

What you're looking at: Automatic detection of unusual cost patterns:

• A customer whose costs suddenly doubled
• A task type that started costing 3x more than usual
• A day where retry waste spiked

Each anomaly shows:

• What changed (before vs. after)
• Which customer or task type is affected
• When it was detected

Why it matters: Cost anomalies often indicate problems before they become expensive. A sudden spike might be:

• A customer running a bulk job they shouldn't be
• A prompt that started failing and retrying
• A new workflow that wasn't accounted for in pricing

What to do:

• Review anomalies weekly (or set up Slack alerts)
• Investigate any anomaly that represents more than 5% of your monthly spend
• Use the "fix snippets" provided to address common causes

Statements — Monthly Cost Reports

What you're looking at: Pre-generated monthly reports you can send to customers (or keep for your own records) showing:

• Total AI cost for the period
• Breakdown by service (OpenAI, Pinecone, etc.)
• Task count and average cost
• Any credits or adjustments

You can download statements as CSV or PDF.

Why it matters: Statements give you auditable proof of what you spent on AI — useful for internal finance, customer reporting, and reconciliation.

What to do:

• Generate statements monthly for your records
• If you charge customers for AI costs, send them their statement
• Use the detailed variant for finance team review

Reconciliation — Did My Numbers Match the Invoice?

What you're looking at: A comparison between:

• What you tracked in Dexcost
• What the provider invoiced (OpenAI, Anthropic, AWS)

The variance is the difference, expressed as a percentage:

Variance % = (Tracked - Invoice) / Invoice × 100

A positive variance means you tracked more than you were billed. A negative variance means you under-tracked.

Why it matters: Provider invoices regularly disagree with internal telemetry by 2-15%. This isn't necessarily fraud — it's due to mid-month price changes, tier discounts, rounding, and tax treatment. But you need to know:

1. Is my tracking accurate?
2. Am I being overbilled?
3. Do I need to adjust my cost model?

What to do:

1. Upload your provider invoice (CSV) or connect your API key for auto-import
2. Run reconciliation — Dexcost will match line items
3. Review the gaps — if variance > 5%, investigate
4. Use the fix snippets to correct tracking gaps

Settings — Configuration

What you're looking at:

• API Keys — manage keys for SDK connections
• Workspace — rename your workspace, configure timezone
• Members & Roles — invite team members ( Owner, Admin, FinOps, Finance, Viewer)
• Integrations — connect OpenAI/Anthropic for reconciliation
• Data Export — download your raw data

Why it matters: This is where you set up the system for ongoing use.

What to do:

• Create separate API keys for each environment (dev/staging/prod)
• Invite your finance team with the Finance role (can see costs but not technical details)
• Connect provider API keys to enable one-click reconciliation

Getting Help

If you have questions about what you're seeing in the dashboard:

1. Talk to your engineering team — they can explain the technical details
2. Check the API Reference — for the technically curious
3. Contact Dexcost support — for billing, account, or technical issues