Book a call
Services AutomationsComplex workflowsAI agents All services
Explore Solutions Work Pricing Insights
Company About Sample scope Evaluation report Example dashboard Contact
Book a discovery call
Example dashboard

The dashboard every client gets.

Every build ships with reporting like this for your systems alone: runs, hours removed, success rate, approvals waiting, and incidents. This page is a working demo with representative figures, so you can see exactly what you would be looking at after launch.

Demo data · refreshed on load

146 automations, workflows, and agents in production
64,374 runs in the last 30 days, across every system we operate
2,300 hrs of manual work removed in September
11 days median from kickoff to first live workflow, over the past 12 months
Volume

Runs, last 30 days.

One run is one triggered execution, from trigger to final result. Weekends dip because most of what we run follows office hours.

Today so far 1,964

Runs per day, all systems

Sep 1 to Sep 30 · 64,374 total
Runs per day across all systems, September 1 to September 30, 2026 Thirty bars between roughly 1,500 and 2,500 runs per day. Weekdays sit near 2,100 to 2,500, weekends near 1,500 to 1,700. The final bar is today and is still filling in. Tue Sep 1 · 2,284 runs Wed Sep 2 · 2,341 runs Thu Sep 3 · 2,297 runs Fri Sep 4 · 2,152 runs Sat Sep 5 · 1,588 runs Sun Sep 6 · 1,512 runs Mon Sep 7 · 2,104 runs Tue Sep 8 · 2,366 runs Wed Sep 9 · 2,412 runs Thu Sep 10 · 2,389 runs Fri Sep 11 · 2,231 runs Sat Sep 12 · 1,627 runs Sun Sep 13 · 1,544 runs Mon Sep 14 · 2,358 runs Tue Sep 15 · 2,447 runs Wed Sep 16 · 2,493 runs Thu Sep 17 · 2,418 runs Fri Sep 18 · 2,276 runs Sat Sep 19 · 1,661 runs Sun Sep 20 · 1,573 runs Mon Sep 21 · 2,401 runs Tue Sep 22 · 2,462 runs Wed Sep 23 · 2,508 runs Thu Sep 24 · 2,455 runs Fri Sep 25 · 2,302 runs Sat Sep 26 · 1,689 runs Sun Sep 27 · 1,602 runs Mon Sep 28 · 2,437 runs Tue Sep 29 · 2,481 runs Wed Sep 30 · 1,964 runs so far
  • Weekday
  • Weekend
  • Today, partial
Outcome

Hours removed per month, trailing 12.

We time the manual process at scoping, multiply by the volume each system handled that month, and subtract the time people still spend on reviews.

Manual hours removed, by month

Oct 2025 to Sep 2026
Manual hours removed per month, October 2025 to September 2026 A line rising steadily from about 900 hours in October 2025 to 2,300 hours in September 2026.
Success rate 99.3%

of 64,374 runs finished without an engineer touching them

Approvals awaiting a person 7

open at the last nightly summary, oldest waiting 14 minutes

Median agent accuracy 95.8%

on evaluation sets of past cases with known answers, across 23 agents, re-scored monthly

Fleet

Systems in production.

The ten highest-volume systems this month, with clients named by industry only. The remaining 136 are mostly small automations that run a few times a day.

Showing 10 of 146
ClientSystemTypeRuns / moStatus
Freight brokerage Quote extraction Workflow 2,318 Healthy
B2B SaaS First-line support agent Agent 4,860 Healthy
Dental group Recall reminders Automation 6,120 Healthy
Specialty retail Order exception routing Workflow 4,380 Healthy
Marine services Invoice to accounting sync Automation 3,940 Healthy
Property management Maintenance ticket triage Agent 3,420 Degraded The CRM vendor has limited request rates since Sep 26. Retries are extended and the backlog clears nightly.
Freight brokerage Carrier onboarding checks Automation 1,210 Healthy
Professional services Proposal drafting assistant Agent 890 Healthy
Home builder Permit document intake Workflow 640 Healthy
Insurance brokerage Policy renewal outreach Automation 0 Paused Seasonal. Resumes Nov 1 at the client’s request.
Reliability

Incidents, last 90 days.

Anything that stopped a production system for more than ten minutes, or caused a run to end in the wrong state. Three this quarter, all resolved, none with data loss.

  1. Sep 18, 2026 41 min

    CRM vendor change broke field mapping

    Resolved

    Cause. The CRM vendor renamed two properties in a scheduled release. The proposal drafting assistant began failing on contact lookups. Runs queued and none were lost.

    Fix. We updated the mapping within the hour. A nightly test now runs against the CRM test environment and alerts us to any field change before it reaches live systems.

  2. Aug 27, 2026 2 h 10 min

    AI provider outage triggered backup

    Resolved

    Cause. The primary AI model provider returned errors for just over two hours. The backup provider took over automatically after four minutes of elevated failures.

    Fix. No data was lost. Response time rose about 30% during the window, and sampled accuracy stayed within limits. We lowered the switch-over threshold from four minutes to ninety seconds.

  3. Jul 14, 2026 26 min

    Expired credential on an accounting sync

    Resolved

    Cause. A login token for the accounting system expired after an admin change on the client side. The first failed run alerted us at 06:12.

    Fix. We renewed the credential and replayed the queued batch. Every credential we hold now has an expiry alert that warns 14 days ahead.

Methodology

How these numbers are made.

What counts as a run

One run is one triggered execution of a system, from the trigger to its final result. A workflow started by a signal from another tool is one run, no matter how many steps follow. An agent handling a support ticket is one run per ticket, however many messages it takes. A scheduled update counts once each time it fires, even when it processes several hundred records.

How hours saved are measured

During scoping we time the manual process as it exists. We take the median of at least ten observed instances per step. We multiply that baseline by the volume the system handled in the month, then subtract the time people still spend on reviews and approvals in the new process. We re-measure at the 90-day review and keep whichever figure is lower.

What success means

A run succeeds when it reaches its intended final result without an engineer stepping in. Automatic retries that eventually complete count as success. Runs that end in a deliberate handoff to a person count as success when the handoff was the right call. We sample and check that weekly. Runs still waiting on an approval are excluded until they finish.

How agent accuracy is scored

Before launch, we score every agent against an evaluation set: at least 200 of the client's own past cases with known right answers. We re-score monthly as the model, the instructions, or the data change. The figure on this page is the median across all agents in production. Individual results are shared with each client, and one is public in the Halcyon case study.

Why some systems are excluded

Systems in their first 14 days are left out while their baseline settles. Systems paused at the client's request are listed but contribute no runs. Pilots that have not gone live are not counted, and neither are systems that clients have taken fully in-house, since we no longer see their data.

Rounding and timing

We round hours to the nearest hundred and percentages to one decimal place. Run counts are exact as of the nightly summary. The page is regenerated as demo data each time the page loads. We replace client names with industries and never publish a figure that could identify a single client's volume.

Want these numbers for your own operations?

Every build ships with a dashboard like this for your systems alone.

Book a discovery call See pricing
esc
PagesHomeAutomation and AI agents for growing companiesPagesServicesAutomations, workflows, and agentsServicesAutomationsRule-based handoffs that remove copy-paste work between tools.ServicesComplex workflowsProcesses that span several tools, with approvals, exceptions, and clean data.ServicesAI agentsAgents that read, decide, and act across your tools. People stay in the loop.PagesSolutions by industryLogistics, healthcare, professional services, e-commerce, SaaSPagesWorkCase studies with the numbersCase studiesLumen LogisticsQuote requests from inbox to booked load in nine minutesCase studiesHalcyonA first-line support agent that resolves 61% of tickets on its ownCase studiesMeridian Dental GroupRecall reminders that cut no-shows by a third in 90 daysPagesPricingFixed prices and the instant estimatorPagesSample scopeThe one-page scope every client receivesPagesSample evaluation reportWhat ships with every agent buildPagesExample dashboardThe reporting every client getsPagesAboutA small senior teamPagesInsightsField notes from the buildInsightsWhen not to use an AI agentMost automation value comes from rule-based work. A guide to choosing rules, workflows, or agents for each process.InsightsDesigning approval gates that people useHuman review fails when people are asked too often or too late. Here are the patterns we ship, with the numbers behind them.InsightsThe true monthly cost of an automationAI model usage, tool seats, hosting, and the upkeep nobody budgets for. A worked example from a real quote pipeline.PagesContactBook a discovery call
↑↓ navigate↵ open⌘K or / to open anywhere