The part number often does not exist yet: it has to be configured
Actuators, drives and motion components are not picked off a shelf, they are specified. Rapidflare reads the requirement in plain text, runs the config, and creates the SKU in your system. Then, it takes the purchase order and enters it into your ERP.
Product selection, where the SKU has to be built
Many of the components you carry are configurable. There’s no fixed part sitting in a catalog waiting to be found, someone has to specify it.
- As natural language. A customer describes their specification in an email or over the phone, not as parametric fields.
- As an image. A photo or drawing, followed by a "what’s compatible with this?" query.
- Someone interprets it, looks it up, and runs the configuration and spec tools before quoting anything.
- Reads the spoken/written specification and deciphers what is being asked for.
- Runs the configuration and spec tools as part of the conversation, instead of handing it off to a person.
- Recommends the right part or creates the configured SKU directly in SAP or the CRM.
- Answers from an image, identifying any compatible parts.
There’s no existing part number to look up. Someone has to specify it, every time.
Once the order arrives, it stops being typed in by hand
Tolomatic processes over 1,400 purchase orders a month, which took around 150 hours of manual entry into SyteLine.
Purchase orders arrive by email from distributors, direct customers and Tolomatic’s own storefront, and no two customers lay one out the same way. Historically that meant a person reading a PDF and typing it in: looking up the customer, matching part numbers, resolving the ship-to address, entering it, one order at a time. It doesn’t scale or tolerate the real-world mess of purchase orders: embedded-image PDFs instead of text, non-standard SKU columns, several separate orders bundled into a single email.
Any system can read a PO. The harder part is knowing when to stop and hand it to a person
This system writes into a live order pipeline, so every order passes four independent validation checks before anything is created.
Staged, against real order traffic
Nobody should flip a switch on a system that creates customer orders. We prove each stage on live traffic before expanding scope.
Manual-assisted extraction
Narrow scope, submitted by hand, so extraction and validation logic can be proven against real orders before automation.
Automated flow
Orders are picked up and processed automatically within a limited scope. Running against live traffic is what surfaces the formatting problems a generic PO parser misses.
Production
Full deployment, with exceptions still routing to a person. Tolomatic went from kickoff to a board-level demo of the complete flow in roughly four months.
It removes the repetitive work, not the judgment calls, so the team spends their time on the orders that need a person.
Technical support, on the same product knowledge
Sizing questions, compatibility, installation, troubleshooting, answered from your documentation, on your site, in your product or inside Slack and Teams for your own team. The same catalog that drives selection drives support.
Tolomatic can be referenced for product selection. The order entry system runs against their live SyteLine order pipeline.
Common questions from automation companies
Our products are configured, not selected from a list. Does that work?
Yes, that works. The agent interprets the natural-language specification, runs your tools as part of the conversation, and either recommends the right part or creates the configured SKU in SAP or your CRM.
What if our ERP doesn’t have a clean API for order creation?
This is a common occurrence. SyteLine for example, exposes no single unified API for order creation: it took working across multiple REST endpoints and IDO methods with an incomplete attribute set. Accurate extraction still has to be stitched into the ERP carefully, and that integration work is part of what we deliver.
What happens with a badly formatted purchase order?
Most orders are this way. We handle line items delivered as scanned images rather than text, customers whose "SKU" column is not the part number, and single emails containing several separate orders. Where a value cannot be resolved, an unrecognized ship-via for instance, it routes to review rather than guessing the nearest match.
How do we know it will not create a wrong order?
Four independent validation checks run before anything is written, pricing is never written in directly, no new ship-to address is ever created automatically, and any unresolved exception routes to Sales or IT. The system is designed around knowing when to stop.
Can a customer just send a photo of a part?
Yes. Image-based compatibility lookup is supported: a customer sends a photo or drawing and asks what is compatible, and the agent answers from your catalog.
Related
Send us thirty purchase orders
We’ll run them through extraction and validation and show you exactly what we get right, what we flag, and why.
SOC 2 Type II · Every answer traceable to its source