Two weeks after her midnight Google search, something lands in Sam’s inbox that makes the whole thing worse.
It’s a PDF from Henderson’s. Not a purchase order. A vendor scorecard. Henderson’s does these quarterly, rating every supplier on three metrics: order accuracy, delivery timeliness, and invoice consistency. Sam has received these before. She usually skims them.
This time, she reads every line.
MapleCo’s order accuracy: 91%. The threshold is 95%. Below threshold.
Delivery timeliness: 87%. The threshold is 92%. Below threshold.
Invoice consistency: 94%. Threshold is 95%. Below threshold, but barely.
At the bottom, in a box that Sam has never seen populated before, there’s a note: “Suppliers scoring below threshold on two or more metrics for two consecutive quarters will be placed on Vendor Performance Review. Please contact your account representative to discuss a remediation plan.”
Sam stares at the screen. She’s below threshold on all three metrics. Henderson’s hasn’t said anything yet. They haven’t called. They’ve sent a document. And data, Sam is learning, is how large companies say “we’re watching” without raising their voice.
She puts the scorecard in a drawer. She doesn’t show it to Lisa. She doesn’t show it to anyone.
She will think about this later. Right now, she has nineteen purchase orders in the shared inbox and no time to think about anything.
She doesn’t get to think about it later. Ten days after the scorecard lands, her phone rings. Greg Hollis, Henderson’s procurement manager. Five years of purchase orders, and she has heard his voice maybe four times.
“Sam. I sent you the scorecard on the third. The box at the bottom asks you to call me.”
“I know. I’ve been meaning to.”
“Here’s where we are.” He doesn’t sound angry. He sounds like a man reading the weather. “Two quarters under threshold and corporate puts you on Vendor Performance Review. Review means we re-bid your shelf space, and somebody cheaper is always waiting. Q4 is going to be ugly for everyone, so I asked them to look at the trend instead. They gave me until the end of June. If your numbers haven’t moved by then, it’s out of my hands.”
“They’ll move.”
“I hope so. I like your syrup.” He hangs up.
The next morning, Sam does what every overwhelmed business owner does.
She hires someone.
His name is Jordan. He’s 26, sharp, and eager, fresh from a tech startup that folded three weeks before his rent was due. He took the MapleCo job because he needed to eat, not because he dreamed of maple syrup distribution. A recruiter in London finds him in nine days, for a fee Sam tries not to think about. He starts at $48,000 a year. Sam’s logic is simple: Lisa is drowning, so add another pair of hands.
For the first two weeks, it feels like the right call. Jordan picks things up fast. He starts processing POs alongside Lisa. The backlog shrinks. Sam feels a flicker of relief.
By week six, the flicker is gone.
Sam has added a second person to a broken process. Jordan does the same manual data entry that Lisa does, copying PO data from PDFs into QuickBooks, updating the spreadsheet, emailing tracking numbers. He’s faster than Lisa at typing but slower at judgment calls, because he doesn’t know which customers are fussy about grades and which suppliers need a follow-up call on Wednesdays.
The bottleneck didn’t disappear. It got a second chair.
Sam is paying $48,000 a year, plus CPP, EI, benefits, equipment, and the hours Lisa spends training Jordan instead of entering orders. All-in, Jordan is costing the business about $65,000 a year. The error rate hasn’t dropped. Neither have Sam’s hours. Henderson’s is still shaky.
Sam thinks she just needs to give Jordan more time.
At the end of week six, on a Friday afternoon in early January, Jordan knocks on Sam’s door frame.
He’s polite about it. He’s taken a job at a software company in Kitchener. They want him Monday.
“I’m sorry,” he says. “I can’t spend forty hours a week copying numbers from one tab into another.”
At least he stayed through the gift baskets.
Sam tells Lisa before she goes home. Lisa takes it quietly. “Six weeks,” she says, and goes back to her screen. She doesn’t say the rest. She’s been copying numbers from one tab into another for eight years.
The math is simple, and Sam hates it. The recruiter’s fee. Six weeks of wages. The hours Lisa spent training him instead of entering orders. About $15,000 gone, and Lisa is back to doing it alone.
Two weeks later, in the middle of January, with Lisa underwater and Henderson’s clock running, Sam makes her second mistake.
She books a demo. A software company called CloudSyncPro has sent her a polished email about their “AI-powered supply chain platform.” Sam reserves the boardroom, puts the Zoom call on the big screen, and brings Lisa and Ray in to watch.
The salesperson is about twenty-four and wearing a quarter-zip vest. He says “seamless” eleven times in forty minutes. He clicks through animated dashboards showing real-time supply chain velocity, a phrase nobody in the room can define.
“It connects your ordering, inventory, shipping, and invoicing into one unified ecosystem,” he says, beaming. “Any questions?”
Lisa has one. “Our biggest customer sends purchase orders as PDFs with a cover sheet on page one that has to be ignored. Then they pay four invoices at a time with one USD wire. How does it match those in QuickBooks Online?”
The rep doesn’t stop smiling. “Great question. Our AI layer ingests omnichannel inputs seamlessly.”
“Right,” Lisa says. “Does it skip the cover sheet?”
“Our onboarding engineers can scope a custom connector for that.”
Ray leans forward from the far end of the table. “Restaurant in Niagara faxes us a PO on carbon paper. What does your ecosystem do with that?”
The rep laughs like Ray is making a joke. Ray is not making a joke.
Ray stands up, picks up his coffee mug, and gives his assessment in six words. “So it’s another thing that doesn’t.”
Then he goes back to the dock. He is not invited to the next software review meeting.
Sam signs anyway. She wants something purpose-built, and the dashboards were very nice. CloudSyncPro costs $2,500 a month: $30,000 a year, billed annually, up front.
By the end of January, the platform is shelfware.
CloudSyncPro was built for companies doing $200 million to $500 million in revenue with 200-person operations teams. It has features Sam will never use and is missing features she desperately needs. It doesn’t understand maple syrup grading. It doesn’t skip Henderson’s cover sheet. It only connects natively to QuickBooks Enterprise, a different product from the QuickBooks Online that MapleCo runs on. And the “custom connector” turns out to be a $15,000 setup fee that wasn’t in the original quote.
Ray, as it turned out, was completely correct.
Sam’s third mistake is the one that almost breaks something for real.
In the last week of January, with Jordan gone and CloudSyncPro gathering dust, she calls Dave.
Dave runs the three-person IT company in Ingersoll that has looked after MapleCo for nine years. Laptops, Microsoft 365, the firewall, the backups, the warehouse printers Ray swears at. When the Wi-Fi drops, Dave fixes it. When someone clicks a link they shouldn’t, Dave cleans it up. As far as MapleCo is concerned, Dave is the technology department.
“Can your team build us something with AI?” Sam asks. “For the purchase orders.”
“Sam, we’re an MSP. We keep the Wi-Fi up, the laptops patched, and the ransomware out. Custom AI isn’t our trade.”
“I’m not asking for custom AI. Customers keep calling to ask if we got their PO. I just want them to get a confirmation so they stop calling. Something small.”
Something small is how most bad weekends start. Dave knows this. He also likes Sam, and MapleCo has paid every invoice on time for nine years.
“Fine,” he says. “I’ll wire up a quick Zapier thing with ChatGPT. As a favour.”
On Monday he shows her on his laptop. “When a PO comes into the shared inbox, Zapier picks up the email, sends the PDF to ChatGPT to pull out the order details, and sends the customer a confirmation. Then it books the pickup with the carrier, so Ray doesn’t have to phone it in. If any step fails, it sends the email back into the inbox and starts over. I tested it with ten emails. It works.”
Sam is impressed. It took him one weekend. CloudSyncPro took her $30,000.
“Ship it,” she says.
Dave hesitates. “It’s a favour, Sam. Nobody’s watching it.”
“It’s a confirmation email. What’s it going to do?”
Dave turns it on.
For three weeks, it works beautifully. Customers send POs by email, and within minutes they get a professional confirmation: “Thank you for your order. We’ve received your purchase order #4721 for 24 cases of Grade A Amber and 12 cases of Grade A Dark.”
Sam is thrilled. She tells Lisa this is the future. Lisa nods politely but says nothing. Lisa has seen “the future” before. It usually has a shelf life of about a month.
Marcus is even more thrilled. He watched Dave’s demo over Sam’s shoulder, and by Wednesday he’s doing the same trick by hand in his own ChatGPT account: pasting in a customer’s POs to see what they ordered, drafting replies to buyers, asking it who’s due for a reorder call. He shows Sam a summary it wrote of Golden Dragon’s last six months. It’s good. She tells him so.
On a Sunday night in February, at 2:14 AM, the carrier’s API goes down for maintenance. The pickup step fails. Dave’s shortcut does exactly what he built it to do: it sends the email back into the inbox, and the zap starts again from the top. It reads the PO, extracts it, and sends the confirmation. Then the pickup step fails again. Nothing in the zap records that the customer already has a confirmation. For every PO that comes in overnight, the same email goes out again. And again. And again.
On Monday, Sam’s alarm goes off at 5:47, and she lies in bed and watches her inbox fill.
Got your confirmation. And again. And again?
Is this a phishing thing?
Please confirm we are NOT receiving eight pallets. Our dock takes two.
Fourteen customers, each with between three and eight identical confirmations. Some of them reply to every copy. By seven, Greg Hollis, the man who gave her until the end of June, has forwarded his six copies to his VP and copied Sam. His note is one line: FYI. MapleCo.
At 8:40, Ray is in her doorway holding a printout. The customer whose dock takes two has forwarded all eight confirmations to the shipping desk.
“Twenty-four cases of Amber,” Ray says, and fans the pages. “Eight times. Which one am I shipping?”
Dave is in Sam’s office before nine. He shuts the zap off, then scrolls through its history for a long minute without saying anything.
“I built you a favour,” he says finally, “not a system. What you’re after needs proper error handling, retries that don’t start over from the top, monitoring around the clock, and somebody who answers when it alerts. That’s not what we do, Sam. If you find someone who builds this stuff properly, I’ll hand it over happily.”
That same morning, while Sam fields the angry phone calls, Lisa finds something else. She goes to Marcus’s desk for a customer’s phone number, and his personal ChatGPT account is open on the screen. The sidebar is a list of customer names. For three weeks, he’s been pasting actual purchase orders into it, complete with company names, contact information, order quantities, and pricing. Lisa takes a photo of the screen with her phone.
Lisa mentions it quietly to Sam. “We don’t really have a process for this,” she says. “We don’t have a process for any of this. We have a habit, and the habit doesn’t include a section on what Marcus is allowed to paste into the internet.”
The next morning, Sam sits in her office with the door closed.
She doesn’t add up the damage. She can’t face the number yet. Instead, she writes apology emails. One per customer, fourteen of them, each one personal, because a form apology to someone who just got eight form emails would be its own kind of insult.
Somewhere around number nine, Marcus knocks and asks if she has a minute.
“No,” Sam says, in a voice she doesn’t recognize.
He backs out of the doorway. She feels bad about it for the rest of the day. The ChatGPT conversation she owes him slides to tomorrow, and then to next week.
The money will hurt. The trust hurts more. Henderson’s is quieter than usual. Mrs. Chen hasn’t placed an order this month.
That Friday, Sam calls Gerald. Not for accounting advice. Just to vent.
Gerald, being Gerald, listens patiently and then says: “Have you called that automation fellow I mentioned?”
Sam sighs. “Gerald, I just spent $30K on software that doesn’t work and had a robot email-bomb my customers at 2 AM. I’m not exactly in the market for more technology.”
“He’s not technology,” Gerald says. “He’s a person. A weird one. But the good kind of weird. He fixed my dental client’s invoicing problem in three weeks, and that practice had been a mess for years.”
Gerald sends Sam a link. She doesn’t open it. She has a stack of credit notes to approve and a family to feed.
At 7 PM, she’s at the kitchen counter with her laptop open beside a pot of pasta she has forgotten to stir. Ben is setting the table around her paperwork. Maya comes in and drops her goalie bag by the door. The bag is roughly the size of Maya.
Sam remembers the last game she actually saw, back in November: a penalty shot, Maya flat on the ice with the puck under her glove, then looking up into the stands to find her. Sam opens her mouth to ask about tomorrow. The laptop chimes. Another credit note.
“Dad’s driving me to the scrimmage tomorrow,” Maya says. She’s eleven. She doesn’t look up from her phone. “You don’t have to pretend you’re going to make it, Mom.”
“I’m not pretending. I’m going to try.”
Maya looks at the laptop, then at her mother. “Okay,” she says, and goes upstairs.
“Ten o’clock, rink two,” Ben says. “I’ll save you a seat.”
The pasta boils over. Sam doesn’t move to stop it. Ben does.
At 11 PM, after Ben and Maya are asleep, she opens Gerald’s link. It’s a blog post by someone named Oscar ONeill, called “Automation vs Hiring: The Real Math for Growing Canadian Businesses.”
The first line reads: You’re working 60-hour weeks and your business is growing. Your instinct says hire someone.
Sam reads the whole thing twice. Salary is just the start, it says. Add CPP and EI, benefits, equipment, and the months it takes before a new hire is actually productive. She doesn’t need to do the multiplication. The recruiter’s invoice is in the same drawer as Henderson’s scorecard.
One line stops her: These are tasks where a human is essentially acting as a slow, error-prone computer.
She thinks about Jordan in her doorway on a Friday afternoon. She thinks about Lisa, who has been doing a computer’s job for years without once complaining about it.
It doesn’t read like a pitch. It reads like someone has a camera in her office.
There’s a contact form at the bottom. She fills it out. A chat window pops up. A chatbot named Winston asks if he can help.
“I’m a maple syrup company,” she types, feeling ridiculous. “And I think my business is falling apart.”
Winston’s reply is immediate: “You’re not the first person to tell me that at 11 PM on a Friday. That’s usually when business owners finally have time to think. Can I ask what’s eating up most of your time right now?”
Sam looks at the question for a while.
Then she types for ten minutes straight.
Monday morning. Sam’s phone rings. The caller ID says London, Ontario.
“Sam? It’s Oscar. From DigitalStaff. I read what you sent Winston on Friday night.”
Sam braces for the pitch.
Oscar says: “Tell me about the last invoice that made you want to throw your computer out the window.”
It’s such a specific question that Sam answers it before she can decide not to. And then she keeps going. For forty-five minutes, she talks. About Lisa, about Jordan walking out, about Dave’s zap, about the six browser tabs, about Henderson’s, about the POs that arrive as PDFs and the tracking numbers she can’t find and the payment matching that only Lisa understands and the month-end that eats her weekends.
Oscar doesn’t interrupt. He asks sharp questions: “How many POs per week?” “What format do Henderson’s send?” “When you say Lisa enters them, you mean line by line into QuickBooks?” But mostly he just listens.
At the end, Sam waits for the pitch.
Instead, Oscar says: “You don’t have a technology problem, Sam.”
“I’m sorry?”
“You have a process problem that technology can solve. But only if we understand the process first.”
“So what do you start with?”
“I start with the dumbest possible question: What are you actually doing, step by step, right now? Not what you think you’re doing. Not what you wish you were doing. What actually happens when a purchase order hits your inbox at 9 AM on a Tuesday?”
“I told you. Lisa opens the email…”
“No. Slower. What email client? Where’s the inbox? Is it shared? Who sees it first? Does the subject line tell you anything? Is the PO always a PDF? Is it always an attachment, or sometimes inline? Does it always have the same fields?”
Sam has never thought about her business at this level of detail. It’s like someone asking her to describe how she breathes.
Sam hangs up and looks at her laptop. Six browser tabs, still open from this morning.
She has done this process thousands of times. She has never once described it. Not to the CloudSyncPro salesperson, who wanted to talk about features. Not to Dave, who built exactly what she asked for. Not even to herself.
She picks up her phone and books a meeting at Oscar’s office in London for the first week of March.