Case study · DotVA

DotVA: giving every VA a Claude Code seat made the team 2x more productive without adding hires

How we deployed Claude Code + a tight CLAUDE.md across DotVA's virtual assistant team. Same headcount, roughly double the throughput, happier VAs.

Client
DotVA
Industry
Virtual assistant agency
Timeline
6 weeks rollout (one VA per week)
Investment
~$80 AUD/month per VA in Claude API + training time
Challenge

DotVA's VAs were drowning in repetitive admin work, Xero coding, inbox triage, document processing, draft generation, while clients increasingly asked for more strategic work the VAs didn't have time to do.

Outcome

Every VA now operates with Claude Code as a co-pilot. Repetitive admin compressed 5-10x. Same headcount delivers roughly double the billable hours of high-value work. NPS up. Churn down.

The build, one sentence

Over six weeks we equipped every DotVA virtual assistant with Claude Code + a tight DotVA-specific CLAUDE.md, doubling effective throughput per VA, keeping headcount the same, and freeing them up to do the strategic work clients actually pay premium for.

The problem

DotVA is an Australian virtual assistant agency. We staff dedicated VAs to small businesses, mostly Australian-owned, who need help with operations, admin, customer service and bookkeeping support.

By late 2025, two trends had collided uncomfortably:

  1. Clients were asking VAs for more strategic work, campaign analysis, content drafting, supplier comparisons, prep for executive meetings.
  2. The VAs were buried in manual admin, inbox sorting, Xero coding, expense reconciliation, document processing, drafting routine emails.

The strategic asks couldn’t fit because the manual work consumed 50-70% of every VA’s day. Throwing more VAs at the problem would have worked but compressed margin. Replacing VAs with AI was the obvious wrong answer, clients buy DotVA for the relationship.

The right move was augmentation, not replacement.

What we did

A six-week rollout. One VA per week, starting with the most technically curious.

The kit

Every VA got:

  • Claude Code installed on their work laptop (the getting started guide is essentially the install doc we used)
  • A standardised DotVA CLAUDE.md, Australian English, AUD pricing, DD/MM/YYYY, our voice + tone guide, our document templates
  • 5 slash commands, /standup, /triage (inbox sort + draft replies), /quote, /xero (categorisation suggestions), /aud (rewrite text Australian)
  • MCP wired to client tools per-client, usually Xero + Gmail + sometimes Notion or Shopify
  • 6-10 hours of pair-prompting with Jenn in their first week, sat next to them, watched them use Claude on real client work, built their CLAUDE.md additions live

The rollout sequence

WeekVAOutcome
1Most tech-curious VA, two-client portfolioInbox-triage workflow shipped first; 6 hours/week back per client
2Next VA, four-client portfolioXero-coding workflow added; the week-1 VA helped train
3-6Remaining VAs, rolled out one per weekEach new VA had two prior VAs to learn from

The “one per week” cadence was important. Every cohort had peer support from the prior cohort. By week 6, our most experienced VA was teaching new VAs the workflow she’d helped design.

What costs money

ItemMonthly AUD per VA
Claude API (Anthropic Console, metered)$60-100
Claude Pro (chat interface, separate from Code)$31
Training time (one-off)~6-10 hours @ Jenn’s time
Total ongoing per VA~$90-130 AUD/month

For a VA charging $1,800-2,500 AUD/month to a client, the AI tooling is 4-7% of revenue. Margin trade-off pays for itself the moment the VA’s effective throughput moves more than 7%.

What it delivered

Across the team, 90 days post-rollout:

  • Effective throughput per VA: roughly 2x. Each VA delivers roughly twice the billable output per hour. Validated by tracking hours-to-deliverable across a sample of recurring tasks.
  • Client NPS: +14 points. Clients noticed faster turnaround + more strategic work.
  • VA churn: zero in the 6 months post-rollout (vs 2 VAs in the 6 months prior). The work got more interesting.
  • Margin per client: up ~20% as VAs absorbed more strategic asks without DotVA needing to add hours.

What we’d do differently

  1. Start with Jenn shadowing each VA, not running training sessions. Watching real client work is more effective training than running a workshop. We pivoted to this in week 3 and should have started this way.

  2. Standardise the CLAUDE.md template harder. We let VAs customise too much in the first 3 weeks. Now every VA’s CLAUDE.md is a fork of the same master template, with client-specific bits in a separate file. Much easier to debug + update.

  3. Build the “AI handover” doc earlier. When a VA goes on leave or rotates between clients, the AI workflows need to transfer with them. We now have a one-page handover template that the outgoing VA fills in with each client’s specific Claude workflows.

Why this matters for other agencies

If you run any AU-based services agency (VA, accounting, bookkeeping, marketing, design), the playbook transfers:

  • Get every operator a Claude seat
  • Standardise the system-prompt across the team
  • Augment the relationship; never replace it
  • Charge the same hourly rate; deliver more per hour

The result, in our case, was a team that’s still the same size but does more, with happier clients and lower churn.

If you’d like to see DotVA’s CLAUDE.md template or chat about rolling this out at your agency, the free audit is the starting point.

Stack we used
  • Claude Code
  • Claude Pro
  • Anthropic API
  • DotVA's internal CLAUDE.md template

Want a build like this in your business?

Book a free 30-minute audit. We'll map your business and show you which automation would land the biggest result first.

Book my free AI audit