Dashboard Design / AI
AI Dashboard Design
AI dashboard design is the design of the operating views a team relies on to run an AI product: token usage and cost by model and feature, response times, evaluation results across model versions and the review queue where people check what the model produced. Gestures designs AI dashboards with cost, quality and speed on the same screen, because switching a model or a prompt moves all three at once.
AI dashboard design with Gestures costs $2,250 per week, $4,000 for a two-week sprint or $7,500 per month for ongoing work as models, features and customers change. Every plan covers strategy, design and review on a scope agreed before work starts.
Measures
Cost, quality and speed on one screen
An AI product team decides on trade-offs, such as a cheaper model that answers worse or a faster one that costs more, so the dashboard shows the measures side by side instead of on separate pages.
| Measure | What to show | Breakdown that matters |
|---|---|---|
| Token usage | What to showInput and output tokens, with cached input counted separately | Breakdown that mattersBy model, feature and customer |
| Cost | What to showSpend calculated from usage and the provider's prices | Breakdown that mattersBy feature, and as cost per completed task |
| Speed | What to showTime to first token and total duration, as percentiles | Breakdown that mattersBy model and request type |
| Quality | What to showEvaluation scores and the outcome of human review | Breakdown that mattersBy model version and prompt version |
| Failures | What to showRefusals, timeouts, rate limits and blocked requests | Breakdown that mattersBy error type and feature |
Telemetry
Charts built on the metric names your engineers already record
The OpenTelemetry semantic conventions for generative AI define shared metric names, such as gen_ai.client.token.usage for tokens and gen_ai.server.time_to_first_token for the wait before the first token appears. A dashboard designed around these names works across model providers, and we label each chart with the metric it reads so an engineer can check the source.
- Each chart titled in plain words, with the underlying metric name in its details.
- Cost charts that show which price list was applied and when it was last updated.
- A note on any chart whose data comes from sampling rather than every request.
- Gaps in telemetry shown as gaps, never drawn as zero usage.
Versions
Compare model and prompt versions before and after a change
Every model or prompt release changes several numbers at once, so the comparison view lines up versions on the same evaluation set and the same period of traffic.
| Column | What it contains |
|---|---|
| Version | What it containsThe model and prompt version, with its release date and owner |
| Evaluation score | What it containsThe score on the shared evaluation set, with the number of cases |
| Human corrections | What it containsThe share of outputs that reviewers edited or rejected |
| Cost per task | What it containsSpend divided by completed tasks, not by requests |
| Time to first token | What it containsThe p50 and p95 values for the same request types |
Releases also appear as markers on every time series, so a change in cost or quality can be tied to the version that caused it.
Review queue
A review queue people can work through
A human review queue is part of the dashboard when an AI product needs people to approve or correct outputs, and its design decides how many items a reviewer can clear in an hour.
- Items sorted by risk score or age, with the score explained in one line.
- The source and the model output side by side, so the reviewer never switches screens.
- Clear states such as pending review, approved and changes requested, with the reviewer's note attached.
- Queue health at the top: items waiting, the age of the oldest item and the items cleared today.

Agent runs
Agent runs as timelines, not only totals
An agent run is a sequence of model calls and tool calls, so a total for duration or cost cannot explain why a run failed. We design each run as a timeline that shows every step in order.
| Part of the view | Question it answers |
|---|---|
| Run status | Question it answersDid the run finish, stop for approval or fail, and when |
| Step timeline | Question it answersWhich steps ran, in what order and how long each took |
| Tool calls | Question it answersWhich tools were called, with which inputs, and what they returned |
| Tokens and cost per run | Question it answersWhat this run cost and which step used the most |
| Handoff to a person | Question it answersWhere a person approved, corrected or took over the task |
AI inside dashboards
When the dashboard itself answers with AI
A generated summary or a natural-language answer inside a dashboard is trusted only when people can see the data behind it, so every generated answer shows the query it ran, the period it covered and the numbers it used.
- Generated text marked as generated, with the time it was produced.
- The chart or table behind the answer, one click away.
- An edit control for the question, so a wrong reading can be corrected instead of repeated.
- No generated summary for a period with missing data, and a message that says why.
Ragip Diler, who leads Gestures, designed analytics workflows for AI products as Senior Product Designer at Venice, an independent design studio for AI and venture-backed companies.
Scope and cost
AI dashboard design scope and cost with Gestures
AI dashboard design with Gestures costs $2,250 per week, $4,000 for a two-week sprint or $7,500 per month for ongoing work. The plan depends on which of the views below you need first.
| Scope | What it covers | Plan |
|---|---|---|
| Usage and cost | What it coversTokens, spend and cost per task by model, feature and customer | PlanWeekly sprint, $2,250 per week |
| Quality and versions | What it coversEvaluation results, human corrections and version comparison | PlanWeekly sprint, $2,250 per week |
| Review queue and agent runs | What it coversThe review workflow and the run timeline | PlanTwo-week sprint, $4,000 |
| An AI operations suite | What it coversAll of the above, plus customer-facing usage pages and a shared chart system | PlanMonthly partnership, $7,500 per month |
The screens people use inside the AI product, such as drafting, approvals and correction, are covered by AI product design, and the website that sells it by AI website design. For an overview of AI work, see our AI and LLM startups page. API-level usage, latency and error charts for developer products are covered by developer tools dashboard design.
Monthly partnership
$7,500$10,500USD / month
Studio $7,500 + Growth $3,000 / month
For an AI product that changes models, prompts and features every month. Set the current focus each month and review what needs attention next.
Strategy · Design · Review
Work starts within 1 day of payment.
Cancel anytime.
Start your projectWeekly sprint
$2,250$4,000USD / week / 2 weeks
+ $3,000 / month for Growth
For one dashboard, such as cost and usage, the review queue or the model comparison view. Agree the metrics your engineers already record before starting.
Save $500 with two weeks: $4,000 instead of $4,500.
Strategy · Design · Review
Work starts within 1 day of payment.
Cancel anytime.
Start your projectAdd Growth to a studio plan
SEO, content strategy and content production, alongside your design team. Growth is billed monthly. Cancel anytime.
+$3,000 / month
Compare all plans or book a call to agree the first dashboard.
Questions
Questions about AI dashboard design
How much does AI dashboard design cost?
AI dashboard design with Gestures costs $2,250 per week, $4,000 for a two-week sprint or $7,500 per month. A usage and cost dashboard fits a weekly sprint. See pricing for every plan.
Do you design internal AI dashboards or customer-facing ones?
Both. Internal dashboards serve your engineering and evaluation teams, and customer-facing usage pages show each customer their own usage, spend and limits.
Do we need an observability tool before you start?
No. We can design from the logs and events you already record and list the measures your engineers would need to add, so the design and the instrumentation move together.
Do you set up evaluations or choose metrics?
No. Your team owns the evaluation sets and quality metrics. We design how results, versions and corrections appear, and how people act on them.
Can you design a dashboard that uses natural-language questions?
Yes. We design the question input, the answer with its sources, the chart behind it and the correction path when the answer is wrong.
What do you need from our team to start?
An anonymised usage export, the events and metrics you log, the evaluation results from one release and an engineer who can explain how tokens and costs are counted.
Is AI dashboard design part of AI product design?
They are separate scopes. AI product design covers how people use the model inside the product, and AI dashboard design covers how your team and your customers watch it run.
A conversation
Send us last month's usage export
Share an anonymised usage export and the question your team cannot answer from it today. We will suggest the first view to design and say whether Gestures fits the work. The wider service is on our product design page.