For pulling specs out of PDFs, the schema matters more than the model. Extraction looks like a model problem. It's a contract problem. Define every field. Set ranges that make physical sense. Return null when the value isn't in the document. Reject anything else. With a tight schema, even smaller models do this job well. Without one, the biggest model will still guess. #AI #DataQuality #PIM

Clean the list first. Then write a single word. Remove closed businesses. Remove bounced domains. Pick one contact per role per company. Drop anyone who already declined. A smaller, cleaner list with real research beats a big list with mail merge, every time. #Prospecting #LeadGen

"Local Wikipedia," Wikipedia but only for things a half hour walk from you, is on Hacker News: 748 points. Same encyclopedia, filtered by where you stand, and people love it. B2B buyers want that from a catalog: stock at the branch they can drive to, lead times from the nearest warehouse, their own prices. That data usually already sits in the ERP. The storefront just has to ask. Does your storefront show stock by location, or one total number? #HackerNews #B2Bcommerce

When a B2B quote converts to an order, your ERP should not reprice it. The order comes from a checkout built on the quote, so it carries the agreed prices. The ERP should create a sales order with those prices and skip its own pricing engine. One thing to check: in B2B Edition, a fixed-amount discount on the whole quote is not folded into the line prices. Test how it lands in your ERP. This is one of the few places we routinely build custom. Every ERP's sales-order API has quirks. #B2Becommerce #ERP #BigCommerce

NVIDIA's OpenShell, a runtime for fleets of autonomous AI agents, is on Hacker News: 254 points, 133 comments. Whatever runtime you choose, the real work is deciding what each agent may touch. Give every agent one job, its own credentials and a log someone reads. Start read-only. Widen access only after it has earned it. If you run AI agents in your business, what was the first thing you trusted one to change? #OpenShell #AIagents

Low-voltage and smart-home shops: the walkthrough is where the bid is won or lost. What you capture standing in the house decides what you can price at the desk. On every walkthrough, write down: 1. Every room and what it needs. 2. Where cable has to run, and what's in the way. 3. What the customer said they care about, in their own words. Then build the bid from that, not from memory three days later. #LowVoltage #SmartHome #Estimating

Customer master cleanup is not an IT project. Staff it like it isn't. The right owners are your AR lead and your sales-ops lead. The tech team brings the scripts, the duplicate detection and the test harness. The business decides which record is real. Engaged from week one: it finishes on time. Looped in at week ten: the project slips. Same work. Different calendar. #B2BEcommerce #DataQuality #Replatforming

A small decision with a big effect on who can use it. BidMagic's QuickBooks Desktop link needed qbXML 14.0 on US company files, which meant 2018 R16-class releases and later. BidGlory targets qbXML 13.0, the oldest schema every QuickBooks Desktop release from 2015 on understands. Designed for US, Canadian and UK files where Intuit's SDK supports them. Designed for, not yet live-tested. That's next. #QuickBooksDesktop #BidGlory

Every project has one department that hears about it last and objects first. In discovery we interview IT, e-commerce, ops, marketing and finance where relevant. Not just the sponsor. The group left out of week one tends to surface at launch with a requirement nobody planned for: a tax rule, a credit process, a report someone depends on. In your company, which team is most often looped in too late? #ProjectManagement #Discovery #B2B

"A farewell to Arc browser" is on Hacker News: 98 points, 80 comments. When a tool people rely on says goodbye, the talk turns fast to one question: how do I get my stuff out? The same is true for business software. Ask every vendor to show you the export before you sign, not after. What's one tool your business would have the hardest time leaving? #Arc #SaaS

Short prompt, short answer? You're probably paying for a bigger model than the job needs. Category placement and attribute mapping are classification jobs. Small input, small output. We run those on a fast, cheaper model tier and save the larger model for writing and reasoning. Same pipeline, different engines per step. Ask of every AI step: does this need the smartest model, or just a fast one? #AI #LLM #CostControl

A reply is the end of automation, not the start of a sequence. The moment someone answers, every scheduled follow up stops. A human reads it. A human responds. Nothing feels more robotic than a third nudge arriving after someone already said "let's talk." Does your system stop on reply? #SalesAutomation #B2BSales

The B2B feature buyers love most in our user testing isn't flashy. It shortens their week. Shopping Lists. Warehouses and MRO buyers order the same SKUs every month. Let them save a list, share it with teammates and move it to the cart in a few clicks. Lists built by a Junior Buyer go to a Senior Buyer or Admin for approval first. Add reorder from a past order and from previously purchased products in Quick Order. If your buyers reorder, design the reorder path before the homepage. #B2Becommerce #UX #BigCommerce

"No more icons in every menu by default" is on the Hacker News front page: 381 points, 220 comments. B2B storefront menus have the same habit: an icon on every category and every account link. Buyers ordering by part number scan words. Decoration they have to look past slows them down. Keep icons where they carry meaning, like cart and search, and let text do the rest. Which do your buyers use more: the menu or the search box? #UXDesign #B2BCommerce

The Wolves traded for LaMelo Ball, and ESPN says the Anthony Edwards pairing opens with chemistry and hope. Adding a second star to a team that already has one is an integration project. Same with a new platform beside your ERP: decide early who owns which job, the ball or the data. Chemistry in October is easy. The test is the first rough stretch. When you added a big new system, what did you settle first? #Timberwolves #LaMeloBall

Give your customer a "request changes" button. It's how hesitant deals stay alive. If the only options are sign or ignore, a hesitant customer picks ignore. Offer a clear second path: 1. Accept with a signature. 2. Or request changes, in writing, on the bid itself. A change request is a buyer still at the table. Silence is a buyer who already left. BidGlory's portal gives the client both buttons. See it in BidGlory: bidglory.com/demo?utm_source… #Contractors #Sales #Proposals

Show your AR team the duplicate-customer report. Their first sentence tells you the real project scope. If they say: "Yep, the ones we knew about. We'll merge them." You're in good shape. If they say: "Wait, we have three records for that account? Let me check with someone." You just found a long tail of work nobody scoped. Run this report in week one. Not week ten. #B2BEcommerce #DataQuality #ERP

The most useful line in a status report is the one about what didn't ship. Every Friday our clients get four things in writing: What shipped. What didn't, and why. What's next. What's at risk. The second line is the one vendors quietly skip. It's also the one that lets you act early instead of finding out at the deadline. Which of those four do you most often have to chase your vendors for? #ProjectManagement #SoftwareDevelopment #B2B

DuckDB 2.0.0 LTS "Cinnamon Teal" is on the Hacker News front page with 243 points. For business teams, the word that matters is LTS: a version you can build on without chasing every update. A lot of ERP reporting still runs on a nightly export and a spreadsheet. Tools that run your operations should sit on long-term releases, not whatever shipped last week. Does your team pin to LTS releases or ride the latest version? #DuckDB

Four questions IT should ask before any AI gets near the ERP. 1. Where is our data allowed to go, and which model hosts respect that? 2. What does each call cost, and who watches the bill? 3. Is every AI action in the audit trail? 4. If the AI floods the integration, does the ERP slow down for everyone? Good answers sound boring. That's the point. #AI #ERP #ITLeadership