Two Months with WorkBuddy: Breaking the Messiest Post-Trade-Show Work into Three Stages

Third day back from the trade show, and I've still got two stacks of stuff on my desk. One stack is business cards, about sixty of them, half handwritten by the clients themselves in handwriting I can't read. The other stack is quote materials shoved at me by suppliers, some Excel some PDF, mixed Chinese and English, and the currency units aren't even consistent.

In the past, at times like this, I knew the next two days were shot.

Monday morning 7:30, as usual I check WhatsApp first. A Vietnamese client asking if that model we discussed last time is still in stock, a Brazilian sending three model numbers for me to quote, and an Egyptian one whose voice-to-text came out as total gibberish that I have to guess my way through. In my inbox, over forty new emails, and fewer than ten are actually clients asking for quotes. Ten years in foreign trade, I long ago stopped expecting inquiries to sit there all clean waiting for you — the usable ones are always buried in the junk.

Scrolling idly, I hit a push notification: Taiwan has an AI hands-on course on October 6th, six hours, at NCCU's Public and Business Administration Education Center, 3,800 NT a ticket, which comes to over 800 RMB. The instructor came from data visualization, and one line in the promo I recognized: "Solve the pain point of cross-tool integration, automate data consolidation, achieve cross-file, cross-process collaboration efficiency."

True enough. But the course teaches thinking, and thinking can't type sixty business cards into a spreadsheet for me.

So I still open WorkBuddy and break this mess into three stages.

A lot of people skip this step. I've stepped in that pit.

The first time I used WorkBuddy, I took the lazy route, dumped a pile of business card photos and supplier Excel files in all at once, and told it "organize into a table I can follow up on." What came out looked neat, but using it was all problems. The same Vietnamese company appeared twice, because on one card it was written as an abbreviation and in the Excel as the full name, and it treated them as two companies. The quote column was even messier — some filled in RMB, some in Vietnamese dong, and it converted them itself using online exchange rates, and after converting they didn't match the supplier's original quotes.

Later I got smart: rules have to be locked down first, then let it work.

I created two folders, one called raw, where all the raw material goes — card photos, Excel, PDF in three subdirectories. One called output, where only results go. This is my own isolation rule, so it doesn't touch the original files, and if something goes wrong I can start over.

Then I write the rules as a paragraph and paste it to it before every session: company names kept in original text, untranslated; contact column bilingual; country as two-letter code; all quotes unified to USD, using a fixed rate I provide, no looking it up online itself; model numbers must be copied exactly, no merging. The last one is the most critical — it's too smart, smart enough to "merge similar items" for you, but in foreign trade a model number differing by one letter is two different things.

The first stage is cleaning. Drop the card photos in, have it extract company, contact, phone, email, country, and generate a table directly. Of the sixty-plus, about ninety percent of fields auto-fill correctly. The crooked, backlit ones, seven or eight need me to fix myself. One card had a line handwritten on the back, and it didn't recognize it — that I understand, I'd have to lean in close to read it myself.

The second stage is merging. Combine the card table and the supplier quote Excel into one master table. For this step I have it do one thing at a time: first align field names, then merge, then dedupe. Before, I wanted it done in one go, and it forced the column names of the two tables together, and all the data got misaligned.

The third stage is producing output. Take the master table and generate an English quote sheet, then format it into a PPT, and finally write follow-up email drafts by client country.

All three stages together, the first version took about two-plus hours. The next morning I spent another forty minutes going through it, focusing on the remarks column and model numbers. The remarks column it often misses — this flaw has been there since I started using it and still isn't fixed. I guess it's because the remarks column format is too irregular, written every which way.

In the past, this kind of work took me two days alone. Day one organizing data, day two making the quote sheet and PPT. Now compressed to two-plus hours plus forty minutes, and what's saved is mainly that I don't have to type spreadsheets while replying to client messages those two days, my brain doesn't have to switch back and forth.

The good parts are clear: batch, repetitive, uniform format, multilingual. Have it modify a hundred rows of data by the same rule, and it won't get annoyed, won't miss anything, won't start spacing out at 3pm. From my testing, for pure format-type work its accuracy is higher than mine.

The bad parts, let me be clear too.

PPT is its weak spot. The generated layout is viewable, but not what you'd call good-looking, font sizes and line spacing are on the conservative side. My current approach is to have it produce the content framework and text, and I apply the template layout myself. Wanting it to one-shot a PPT ready to send directly to clients — that's not possible yet.

Another is no memory. Every new session, the rules have to be restated. I later saved that rules paragraph as a text file and paste it in at the start each time, which sort of works around it, but this isn't how it should be. A two-month veteran user still has to introduce himself every time — that experience isn't great.

The most fatal is that judgment-type work it can't do. Which is a real client and which is just price-shopping, which supplier to squeeze and which not to offend — it doesn't know, and shouldn't be made to know.

Last month I wrote a piece saying WorkBuddy only does lead cleaning, organizing messy stuff into a follow-up-able table, don't expect it to replace CRM. Now my thinking has moved forward a notch. It doesn't just clean leads — post-show material organization, quote sheet generation, follow-up email drafting, this whole stretch it can handle. But the boundary hasn't changed: it manages "turning materials into something usable," not "whether to pursue this client." On the CRM side I've used it for two weeks, and it still manages closing.

Back to that course. 3,800 NT, six hours, teaching you how to combine AI tools, how to build a second brain, how to do automated workflows. This kind of course has value, especially for beginners. But I've been in this trade ten years, and ways to combine tools are a dime a dozen online. The hard part is that set of rules in your own business that can't be clearly explained. No one can teach you that set of rules — you can only step through them one by one yourself, then feed them to the tool.

Whether WorkBuddy is worth it depends on whether you have that set of rules. If you do, it can take over the dirtiest stretch of work. If not, what it produces you'll still have to redo, and you'd be better off doing it yourself.

In the end, it just takes away the stretch I least want to do. The remaining judgment, I still have to do myself.