The $20 Claude Mobile Setup That Replaced My $300/mo VA

By Lazar Milicevic · Published June 18, 2026 · 10 min read

Most Claude tutorials show desktop power-user setups — MCP servers, sub-agents, terminal workflows — because they look impressive on screen. Nobody shows the mobile workflow, which is where roughly 60% of an operator's decisions actually get made. This is the setup I run from my phone between client calls. Four Projects, voice mode, $20/month, measured savings of 3.2 hours/week on inbox alone.

Why the "hire a VA + buy three SaaS tools" stack is fake leverage

The default fix when a solopreneur is drowning looks like this: hire a part-time VA at around $280/month, add a proposal generator ($49/month), add an inbox prioritizer ($19/month). Total: $348/month for what feels like delegation.

It isn't. You still have to:

• Brief the VA on every new client context
• Review and edit their output before it goes out
• Sync three separate SaaS dashboards
• Maintain prompts and templates inside each tool

What you've actually built is a coordination tax. Every tool needs context you have to feed it. The leverage only shows up if you have enough volume to keep the VA fully utilized, which most 1–5 person operations don't. Below a certain volume, you spend more time managing the stack than the stack saves you.

The mobile Claude setup wins because the context lives in one place (Projects), the input is voice, and the output is structured well enough that you act on it immediately instead of queuing more review work for yourself.

The whole stack, in one screenshot's worth of detail

Here is the entire setup:

No MCP. No Claude Code. No API key. No n8n. No Zapier. The point isn't that those tools are bad — I build with them daily for clients. The point is they're overkill for the mobile decision layer, which is the layer that actually leaks hours from your week.

The four Projects, in the order I use them through a typical day:

1. Inbox Triage — morning and mid-afternoon
2. Proposal Draft — right after sales calls
3. Screenshot to Action — whenever a client sends an image
4. Daily Standup & Decision Log — end of day

Each Project is just a system prompt plus a couple of reference documents. Below is what goes in each one, copy-paste style.

Project 1 — Inbox Triage (saves ~45 min/day)

System prompt structure:

ROLE
You are my chief of staff. I run a solo consultancy. You triage
my inbox the way a senior EA would after 6 months on the job.

CONTEXT YOU MUST USE
- Active clients: [Client A — invoicing automation project, weekly
  Thursday checkin], [Client B — discovery phase, decisions pending
  on scope], [Client C — paid retainer, async only]
- Response SLAs: prospects <4h, active clients <24h, vendors <72h,
  cold outreach = Friday batch
- "Urgent" = money moving, client blocked, or legal/compliance.
  Everything else is not urgent.

INPUT FORMAT
I will paste 10-15 emails (subject + first 1-2 lines).

OUTPUT FORMAT — strict
Numbered list, sorted by priority. For each email:
  [#] [SENDER] — [one-line summary]
      Action: [respond now | template reply | delegate | defer to Fri]
      Decision needed from me? Y/N. If Y, state the decision in <12 words.

Do not write the replies. Do not editorialize. Do not ask follow-ups.

Workflow on the phone:

• Open Gmail, scroll the unread block
• Long-press → copy subject + preview for the 10–15 newest
• Paste into the Triage Project
• 30 seconds later I have a ranked action list
• Knock out the top 3 on the spot, defer the rest

The reason this works isn't the prompt — it's the SLA rules. The model now has a deterministic way to sort, so the output is the same whether I run it in an Uber or at my desk. I measured the old workflow (manual triage, no system) against this one for two weeks before switching. Average savings: 3.2 hours/week, almost entirely from killing the "open email → feel anxious → close email → reopen later" loop.

Project 2 — Proposal Draft (2 days → under 4 hours)

This Project has the heaviest system prompt because brand voice and pricing logic live here.

What I load into the Project knowledge:

• My brand voice doc (1 page — sentence length, banned words, signature phrases)
• ICP definition (3 customer archetypes with examples)
• Pricing tiers with scope language (Starter / Standard / Custom)
• 3 won proposals as reference documents

System prompt (abbreviated):

ROLE
You draft first-version client proposals in my voice. You are
NOT writing final copy — you're getting me to 80% so I can
edit and send within the hour.

INPUT
I will speak (voice mode) for 30-90 seconds describing a
prospect call. Expect: company type, team size, the problem
they described, budget hints, timeline, decision maker, risks
or objections.

PROCESS
1. Map the prospect to one of the 3 ICP archetypes.
2. Pick the pricing tier that matches budget + scope signals.
3. Pull the matching scope language from knowledge.
4. Draft proposal using the structure in the 3 reference wins.

OUTPUT
Markdown proposal with: Summary, Scope, Timeline, Investment,
Next Step. Keep total length ~400 words. Match brand voice doc.
Flag any assumptions you made in a "REVIEW THESE" footer.

The voice mode part is what makes this mobile-native. The instant a call ends I open the Claude app, hit the mic, and ramble for 60 seconds: "Logistics company, 8 staff, want invoice automation across three systems, budget hinted around 4k, timeline 6 weeks, decision maker is the founder, mentioned compliance is a concern." Ninety seconds later I have a draft sitting in the right tier with compliance addressed in the scope section. I edit on the train. Sent before I'm home.

Old turnaround: ~2 days (because the proposal got pushed to "tonight" and tonight kept slipping). New turnaround: under 4 hours.

Project 3 — Screenshot to Action (the underrated one)

Clients send screenshots constantly — invoice PDFs photographed badly, error messages, contract clauses, dashboard exports, WhatsApp messages from their own clients. Without a system this is friction: you have to open the image, read it, figure out what they want, type a reply.

System prompt:

ROLE
You are an image-to-structured-data extractor for an operations
inbox. Every image is either: invoice, contract clause, error
message, dashboard/metrics screenshot, or chat screenshot.

PROCESS
1. Identify the document type.
2. Extract structured fields. For invoices: invoice #, vendor,
   amount, currency, issue date, due date, line items if visible.
   For errors: app, error code, stack trace summary.
   For contracts: clause type, key obligations, dates, $.
3. Compute one suggested next action.

OUTPUT — strict JSON-like block, then a one-line action

TYPE: <invoice|contract|error|metrics|chat>
FIELDS:
  field_1: value
  field_2: value
ACTION: <one sentence, imperative>

Forward the screenshot from my phone into the Claude app, get back something like:

TYPE: invoice
FIELDS:
  invoice_no: 2024-1142
  vendor: NovaPrint d.o.o.
  amount: 87,430 RSD
  due_date: 2024-11-08
  status: overdue (11 days)
ACTION: Send reminder template B to AP contact today.

If you want to make this stickier, pipe the JSON block into a Google Sheet or a Telegram bot on the back end and you've built a mobile invoice tracker without buying accounting SaaS. But you don't need to — even without backend wiring, the structured output is enough to act in 10 seconds instead of 2 minutes.

Project 4 — Daily Standup & Decision Log

End of day, voice mode on, I just dump:

"Today I shipped the second milestone for Client A, got blocked on Client B because I'm still waiting on their API credentials, had a discovery call with a new prospect in fintech, didn't get to the proposal yet, worried about cashflow because two invoices are late, tomorrow I want to push the Client A delivery and chase the invoices..."

System prompt key rules:

• Return a structured log with sections: SHIPPED, BLOCKED, NEW, WORRIES, TOMORROW
• Surface any decision I've mentioned-and-deferred more than once across the past week (the Project keeps prior logs as knowledge)
• Output exactly 3 priorities for tomorrow morning, ranked
• No coaching language, no "great job" — just the log

This replaced a journaling app, a productivity coach subscription I had briefly tried, and the Notion planning page I never opened on mobile.

The real numbers, logged over 6 weeks

I'm allergic to "saves you 10 hours a week!" claims, so I tracked my own usage for six weeks before writing any of this down:

The 14 hours/month figure is conservative — it only counts triage and proposal time, not the screenshot and standup wins, because those were harder to baseline.

The 90-minute setup playbook

If you want to copy this today:

• Open the Claude mobile app, go to Projects, create four with the names above
• For each Project, write a system prompt with three things: (1) the role Claude plays, (2) the exact output format, (3) 2–3 real examples of good output from your own past work
• Upload reference docs to the Projects that need them (brand voice, pricing, past proposals)
• Pin the four Projects to the top of your Projects list
• Turn on voice mode in settings
• Run each Project on real inputs for 3 days, then tighten the prompts based on where the output drifted

Total setup time if you do it carefully: ~90 minutes. Payback: first week.

Why bizflowai.io helps with this

The four-Project mobile setup is the manual version of what I build into client systems at bizflowai.io. For clients who outgrow the copy-paste workflow — usually around the point where they're triaging 50+ emails/day or sending 4+ proposals/week — I wire the same logic into Gmail, Telegram, and their CRM directly: inbox triage runs automatically on inbound mail, screenshot extraction pipes into a Sheet, proposal drafts get generated from CRM call notes without anyone opening an app. Same logic, no copy-paste. The mobile Projects setup is where most operators should start; the automated version is where it ends up when the volume justifies the build.

Frequently asked questions

What is the Claude mobile AEO stack for solopreneurs?

It's a setup using the Claude mobile app on the $20/month Pro plan with voice mode enabled and four pre-built Projects pinned at the top: inbox triage, proposal drafting, screenshot-to-action extraction, and a daily standup/decision log. It replaces a $300/month virtual assistant plus $80-150 in SaaS tools like proposal generators, inbox prioritizers, and CRM add-ons for running a 2-to-5 person operation.

How do I set up an inbox triage Project in Claude?

Create a Claude Project with a system prompt that defines Claude as your chief of staff, including your active client contexts, response SLAs, and what counts as urgent versus batchable. Instruct it to return a numbered priority-sorted list with one-line suggested actions when you paste email batches, flagging decisions versus delegatable or template responses. Paste subject lines and first lines from 10-15 unread emails to get a ranked action list in 30 seconds.

How do I use Claude voice mode to draft proposals after a call?

Build a proposal Project loaded with your brand voice, ideal customer profile, pricing tiers, standard scope language, and three winning example proposals. Immediately after a prospect call, open Claude voice mode and verbally dump key details: company type, staff size, scope, budget hint, timeline, decision maker, and concerns. Claude transcribes and produces a first-draft proposal in your brand voice with correct pricing tier and scope in about 90 seconds.

Why does a screenshot-to-action Project matter for mobile workflows?

Clients constantly send screenshots of error messages, invoices, contracts, and dashboards. A dedicated Project with a system prompt that extracts structured data, identifies the document type, and suggests a next action turns those images into usable records. For example, forwarding an invoice screenshot returns the invoice number, vendor, amount, due date, and a recommended action like sending a reminder template—eliminating the need for accounting SaaS when wired into Telegram or Google Sheets.

When should I use a daily standup Project versus a productivity app?

Use a Claude standup Project when you want one tool to replace a productivity coach, journal, and planning app. At end of day, open voice mode and dump what you did, what's blocked, what worries you, and tomorrow's goals. Claude returns a structured log, flags postponed decisions, and gives three prioritized tasks for the morning—consolidating reflection and planning into a single voice session instead of juggling multiple apps.

Want more like this?

I publish practical AI automation, GenAI engineering, and faceless content workflows on YouTube every week.

Subscribe to bizflowai.io on YouTube — never miss a new tutorial.

Planning an AI automation project or need a second opinion on your architecture?

Connect with me on LinkedIn — Lazar Milicevic, GenAI Engineer & bizflowai.io Founder.

Visit bizflowai.io for our services, case studies, and AI consulting.