Power Automate: build 2 cloud flows — PDF → Claude AI drafting → approval → email

Request for Quote AI-drafted service reports & quotes — Power Automate + Claude API + AroFlo · 1. Who we are and the problem Bindara Group is a commercial trades and facilities maintenance business in Victoria, Australia. We run 20–50 reactive maintenance jobs a day through AroFlo (job management software), with office systems on Microsoft 365. Today, when a technician finishes a job, someone in the office manually turns the raw job notes and photos into a client-ready service report, and the same happens for quotes. It is slow, it queues up, and quality depends on who writes it. We want the drafting automated, with a human approving every document before it goes anywhere. 2. The workflow we want built Five steps, end to end: 1. Job completed (or quote ready) in AroFlo — AroFlo automatically emails the task sheet as a PDF, photos included, to a dedicated inbox. This part is already built and working. 2. Power Automate picks it up — grabs the PDF from the email automatically. 3. Claude (AI) writes the document — reads the PDF including the photos, and drafts a professional service report (or quote) plus a short summary. We supply the tested prompts and the API key. 4. Ops Manager reviews — receives it with three buttons: Approve, Reject (with notes, which sends it back to Claude for one automatic rewrite and re-approval), or Send to Manager (escalates to the Senior Manager for a second review). 5. Approved work lands with admin — report, summary and job details arrive in the admin inbox ready to go, and a copy files itself to SharePoint. Admin attaches the report to the job in AroFlo, raises the invoice, and keys in the data. (For quotes: admin enters it in AroFlo and sends it to the client.) Two flows, one shape: the job-report flow and the quote flow are near-identical clones — different mailbox folder, different prompt, different output fields (report/summary/customer/job_ref vs quote/summary/customer/quote_ref). The one hard rule on quotes: a drafted quote must never contain a price that is not in the source PDF. 3. What is already done (you start from a proven input) • AroFlo Event Messages are configured and emailing the task PDF, with site photos rendering correctly. No AroFlo-side work is expected beyond routing the emails into the right folders. • The service-report prompt has been validated against a real production task PDF (job JN31412: 5 pages, 4 photos) directly on the Claude API — photos are read, photo-dependent detail appears in the draft, and the JSON fields parse cleanly. The prompts document, including that reference input/output pair, is attached. • A click-by-click build guide for the flows is attached as the functional specification. You are welcome to propose better implementation choices provided the behaviour is preserved. 4. What success looks like (acceptance criteria) • A completed-job email landing in the jobs folder produces an approval request to the Ops Manager — containing the drafted report and summary — within 10 minutes. • Approve delivers the full package to the admin inbox and files a copy in SharePoint. Reject with notes produces a rewritten draft and a fresh approval request without anyone touching the flow. Send to Manager routes to the Senior Manager and behaves the same on their decision. • The quote flow meets the same criteria with quote fields, and drafted quotes contain no prices absent from the source PDF. • Five end-to-end test passes per flow, on real PDFs we supply: happy-path approve, reject-and-rewrite, escalation, a photo-heavy PDF, and a malformed/missing attachment. Photo reading is demonstrated (photo detail appears in the draft). • Any failure (API error, unparseable response, missing attachment) notifies a nominated address — nothing fails silently, and nothing unparseable ever reaches the approver. • The Claude API key is never visible in flow run history, exports or documentation (Secure Inputs or better). • Model name, prompts, approver addresses and admin inbox are held in one obvious, editable place so we can change them without a developer — switching the model or endpoint must not require rebuilding the flow. 5. Scope In: the dedicated shared mailbox and folder routing; the two Power Automate flows (trigger → Claude HTTP call with the PDF as a base64 document block → parse JSON → three-way approval → routing → SharePoint copy); retry logic and failure notifications; testing per section 4; a short admin document (how to change prompts/recipients, read a failed run, pause/resume) and one recorded handover session of up to an hour. Out: any AroFlo API integration (this stage is email-in, email-out only); creating invoices, jobs or documents inside AroFlo; changes to AroFlo report templates or Event Messages (already built); Xero or client-portal integration. A larger Stage 2 (AroFlo API middleware on Azure) is planned separately — if you have Azure Functions experience, say so, as Stage 1 performance will weigh heavily in awarding it. 6. What we provide • Microsoft 365 tenant access appropriate to the work, and the Power Automate Premium licence on the flow-owner account. • Anthropic Claude API key, tested base prompts (attached), and the validated reference sample with expected output. • Real completed-job and quote PDFs for testing, and answers to questions within one business day. 7. Before you quote Please skim the attached build guide first, and tell us in your proposal anything you would do differently and why — simplifications welcome. We value engineering judgement over blind implementation; the guide describes behaviour we need, not steps we insist on. 8. Commercials • Fixed price in AUD (ex GST), briefly itemised (setup / flows / hardening & testing / documentation), plus an hourly rate for variations. • Estimated 8–15 hours for an experienced Power Automate builder; expected elapsed timeline 1 week. State your availability. . • Payment terms: propose yours; we typically pay a deposit on commencement and the balance on acceptance. 9. Your proposal should include • One or two examples of Power Automate flows you have built using the HTTP action against an external REST API. • Your itemised fixed price, variation rate, timeline, and any assumptions or exclusions. • What you would change in the attached guide, and why. • Whether you also have Azure Functions capability (for Stage 2). Send quotes and questions to Ben Peynenborg —