AMDG Consulting
Services

Three services. One practice.

Most consulting sells you a deck. We sell you the thing the deck was supposed to describe. Diagnostic, build, operate. No surprises between them.

Service 01Two weeks

Automation audit

We sit with your operators. We map every repeatable task. We tell you honestly what's worth automating and what's not.

RANKREPEATABLE TASKOPERATOR HOURS / MONTHIMonthly reconciliation~14 hrsIIInvoice approval routing~11 hrsIIILead intake → CRM~9 hrsIVInventory sync~6 hrsVWeekly close review~5 hrsVIAppointment confirmations~3 hrsVIIVendor onboarding~2 hrsVIIIAnnual policy updates— recommend keeping manualIXFounder-written client notes— recommend keeping manual
Fig. iOperations map — every repeatable task, ranked.
What you get
  • A ranked list of every automation candidate across your operation.
  • Effort, cost, and risk estimates for each.
  • A recommended first build — the one that pays for the rest.
  • One written page you can hand to a board or a partner.
// A diagnostic deliverable is a ranked list.
// Each candidate carries its own estimates and risk.
type Candidate = {
  process: string;
  team: string;
  frequency: "daily" | "weekly" | "monthly";
  effort_to_automate_weeks: number;
  estimated_saving_hrs_per_month: number;
  risk: "low" | "medium" | "high";
  recommend: boolean;
};
Fig. ii · candidate.ts · a single candidate row
Service 024–12 weeks

Agent-backed workflows

We build the automations. Claude-backed agents where judgment is needed. Plain code where it isn't. Tied into your existing tools.

Email / APIInternal DBagentclaudeWrite DBNotify SlackSend emailoperator reviewambiguous cases · ~5%
Fig. iArchitecture — sources, agent, tools, and operator review.
What you get
  • Real code in a repository your team owns.
  • Documented MCP servers for any internal data sources.
  • Operator-facing UI where a person still sits in the loop.
  • Tests, observability, and a shutoff switch.
// agent.ts — intake bot
export const intakeAgent = {
  model: "claude-sonnet-4-6",
  tools: ["email.read", "crm.lookup",
          "crm.create", "slack.notify"],
  guardrails: {
    human_in_loop: "ambiguous",
    max_monthly_actions: 2_000,
  },
};
Fig. ii · agent.ts · sample intake agent definition
Service 03Monthly retainer

Quiet oversight

We stay until the operators trust the system. Monthly review of what's running. Tuning. Upgrades when the models change.

JFMAMJJASONDMONTHLY REVIEWQUARTERLY MODEL UPGRADE
Fig. iCadence — a year of monthly reviews, at a glance.
What you get
  • A monthly one-page report of what ran and what broke.
  • Model upgrades and prompt tuning as the landscape moves.
  • New automation candidates surfaced from the last month's data.
  • An exit — not a renewal trap. We leave when you're ready.