You get POs by email and key each one into your inventory system by hand. I build the automation that reads the order and drops it straight in — with you approving before anything is created.
A PO lands in Gmail. Someone reads it and types every line into the IMS, then sends the order back out from there.
Counts and adjustments get entered by hand, one item at a time, every cycle.
Each new product and its SKUs are set up manually across the systems that need them.
None of it is hard. There's just a lot of it, it repeats every day, and every step is a chance to mistype a quantity or a SKU.
You asked for help with Claude, and Claude is the right tool — for one half of the job. Here's the split.
Reads the order out of an email or a PDF attachment: the customer, the line items, the SKUs, the quantities. It copes with the fact that every customer formats their PO differently.
Connects Gmail to your IMS so the order is created without anyone keying it in. Matches each line to your internal SKU, and stops for your approval before it commits anything.
You need both halves. I build both.
Nothing is created until you approve it. Once it's proven on your real orders, the clean ones go through on their own — only the unusual ones get flagged for a look.
The PO is read, each line is matched to your SKU, and a Sales Order is prepared in your IMS — held for your approval until you've decided it's safe to let the clean ones through automatically.
Once Phase 1 is trusted, the same approach handles stocktake so counts and adjustments flow into the IMS without manual entry. We scope the exact shape once I've seen how you run it.
Adding a product and its SKUs becomes a single structured step that creates the record consistently across the systems that need it — instead of being re-entered in each one by hand.
We prove it on Phase 1, you see the time saved, then we expand. You're never betting everything on one big build.
B2B orders carry real money, so the first version never creates anything blindly. You see the parsed order next to the original email and approve it — or fix it — in a click.
Control stays with you. The typing goes away. As confidence builds, you decide how much runs on its own.
A short call plus five to ten real POs, including your messiest formats. Real examples are worth more than any spec.
I build the pipeline with the approval step in place and test it on your historical POs, so by the time it goes live it has already handled orders you recognise.
We run it on incoming POs, track the time saved per order, and tighten the auto-approval rules as confidence grows. Then we move to Phase 2 and Phase 3.
Built on lean infrastructure — no stack of monthly subscriptions to carry. Your main running cost is the Claude usage that reads each order, which is small per PO. I'll give you the number once I've seen your volume.
Which platform you use as your B2B IMS — and whether it has an API to create Sales Orders. This decides whether the build is clean and reliable or needs a more manual bridge. If the answer's bad, I tell you on the call, not after you've paid.
Five to ten real purchase orders, including your messiest formats.
How your SKUs are structured, and how a customer's SKU maps to yours.
I build automations for a living. I work across GoHighLevel, n8n, Make and Zapier for the wiring, and I bring in agentic tools like Claude when the job needs real judgement — reading a messy PO, mapping an odd SKU, deciding what to flag — rather than a rigid if-this-then-that rule.
I currently run data scraping, content automation and project delivery for a software team, so I'm used to owning a workflow end to end instead of handing over a script and disappearing. You deal with one person start to finish, and the page you're reading was built by me — not a template.
Aircon King sells aircon units and cleaning and maintenance services. When I came in, requests and orders arrived through every channel and were keyed in manually — the exact bottleneck you have with POs. I built the digital infrastructure that took them from that manual operation toward ₱100M-a-quarter scale: unstructured requests in, structured orders and service jobs out, with the team approving before anything committed.
Different industry, identical problem. Your POs are that same shape — email in, Sales Order out, you in control. I've already done the hard version of this.
Read the Aircon King case studyEach one is messy input turned into a clean, structured operation. Click through to the full case study on my site.
A ₱30M/month manual operation rebuilt into digital infrastructure for ₱100M/quarter scale.
View case study PhilippinesOrder pipelineFour landing pages plus a Messenger-commerce order pipeline that captures and routes every order.
View case study AustraliaAI automation$1.25M AUD of client data migrated with zero downtime, on an automated pipeline.
View case study PhilippinesCapture + attributionA single-page reseller funnel that captures, identity-resolves and attributes every applicant.
View case study Public toolData scrapingA Chrome extension that turns a Google Maps search into a structured lead list.
View case study PhilippinesDigital operationsReworked tooling and process for a 60% cut in monthly software cost.
View case studyFor more case studies, see bcdianco.com/work.
This is the outside view. The real solution is built on a call.
I wrote this from your message alone, so it's deliberately the shape of the work, not the final spec. The right build depends on your full current process — your channels, your IMS, how your SKUs really map, and the exceptions that never fit a tidy diagram. Give me 20 minutes to walk through how you actually operate today, and I'll come back with a solution catered to your business rather than a template.
Twenty minutes on a call and I'll tell you exactly how clean the Phase 1 build is — and roughly how long it takes.