Did I really send 150 reminders, or is this just a theory?
I sent 150 reminders. I run a small industrial supplies business in Pune—fasteners, adhesives, safety equipment—selling B2B to workshops, contractors, and a few Shopify stores that resell PPE kits. We invoice on 30-day terms. Between October and December last year, our overdue stack hit 150 invoices, most of them ₹15,000 to ₹80,000, a few above ₹2 lakh. Wedding season, Diwali bonuses, year-end inventory builds—everyone was buying, nobody was paying on time.
I'd been writing reminders by hand: "Dear Sir, your invoice is overdue, kindly remit at the earliest." Polite, vague, ineffective. I'd get a reply three weeks later saying they'd "process it soon." I decided to test whether AI could write reminders that actually moved money. I split the 150 overdue invoices into two groups of 75, matched by amount and days overdue. Group A got reminders written by Claude Sonnet 4.5. Group B got reminders written by Gemini Pro. I sent them all via WhatsApp Business over six weeks, tracked who paid within seven days of the reminder, and measured the difference.
Claude's reminders got paid 40 per cent faster. Not 40 per cent more people paid—40 per cent faster median time from reminder to payment received. Group A (Claude) averaged 6.8 days from reminder to payment hitting our account. Group B (Gemini) averaged 11.4 days. Both groups eventually paid about the same total amount, but Claude's tone got them to move sooner.
What made Claude's reminders different from Gemini's?
Claude wrote reminders that sounded like a patient human who knew the customer was busy but needed the money now. Gemini wrote reminders that sounded like a very polite robot reading from a collection agency script.
Here's a real example. Invoice: ₹43,200 for M8 hex bolts and nylon cable ties, 22 days overdue, customer is a small contractor in Hadapsar who's bought from us for two years.
Claude's version (I gave it the invoice details, customer history, and asked for a WhatsApp-ready reminder under 60 words):
Hi Prakash, hope the site work's going well. Our invoice #PUN-2847 for ₹43,200 (hex bolts + cable ties) is now 22 days overdue—we've got our own supplier payment due Friday and could really use this cleared. UPI to 9823xxxxxx or NEFT details below. Can you process it today? Let me know if there's a hold-up.
Gemini's version (same prompt):
Dear Mr. Prakash, this is a gentle reminder that Invoice #PUN-2847 dated 12th October 2024 for ₹43,200 remains outstanding. We kindly request you to settle this amount at your earliest convenience via UPI (9823xxxxxx) or NEFT. Please confirm receipt of this message and provide an expected payment date. Thank you for your cooperation.
Claude's reminder named the products, acknowledged the relationship, gave a reason (our own supplier payment), and asked a direct question. It felt like I'd typed it myself in a hurry between calls. Gemini's reminder was formal, passive ("remains outstanding"), and asked for a confirmation instead of the money. It felt like a template.
The difference showed up in replies. Claude's reminders got responses like "will send by EOD" or "bank's closed today, tomorrow morning confirm." Gemini's reminders got "noted" or no reply at all for three days, then a payment with no message.
Which model understood UPI and NEFT details better?
Both models knew what UPI and NEFT were, but Claude was better at deciding which payment method to emphasise based on the amount and customer type.
I gave each model a simple rule in the prompt: for invoices under ₹50,000, lead with UPI and put NEFT details at the end. For invoices over ₹50,000, lead with NEFT because most businesses hit their daily UPI limit and won't bother scrolling back up to find bank details. Claude followed this exactly. Gemini listed both methods in every reminder regardless of amount, in the same order, with the same weight. That sounds neutral, but it's not—it makes the customer choose, and choosing adds friction.
One reminder was for ₹1.87 lakh (industrial gloves, 31 days overdue, customer runs a small factory in Chakan). Claude's version said: "NEFT details: HDFC Current a/c 50200xxxxxx, IFSC HDFC0001234. UPI works too if you want to split it across two days." Gemini's version listed UPI first, then NEFT, with no acknowledgment that ₹1.87 lakh is over most UPI transaction limits. The customer replied to Claude's reminder within two hours saying they'd NEFT it the next morning. They did.
Claude also handled partial payments better. When I told it a customer had paid ₹30,000 of a ₹72,000 invoice and the balance was now 18 days overdue, it wrote: "Thanks for the ₹30K last month—that helped. The remaining ₹42,000 is still pending, though. Can we close this one out this week?" Gemini wrote: "We acknowledge receipt of partial payment of ₹30,000. The outstanding balance of ₹42,000 is now overdue. Please remit the balance amount urgently." Acknowledge receipt. Remit urgently. No human talks like that on WhatsApp.
Did tone actually matter, or was it just luck?
Tone mattered. I tested it deliberately.
Halfway through the experiment, I took ten of the slowest-paying customers from Group B (Gemini) and sent them a second reminder, this time written by Claude, using a firmer tone. I told Claude: "This is the second reminder, first one sent 9 days ago, no reply. Keep it polite but make it clear I need the payment this week."
Claude wrote things like: "Hi Suresh, following up on my message from last Tuesday—invoice #PUN-2891 for ₹28,400 is now 31 days overdue and I really need it cleared by Friday. If there's an issue with the amount or the products, let's sort it out, but I can't keep extending this. UPI or NEFT, whichever works." Seven of the ten paid within three days.
I tried the same thing with Gemini—same prompt, same instruction to be firmer. Gemini wrote: "This is a second reminder regarding outstanding Invoice #PUN-2891. Immediate payment is required to avoid further action. Please treat this as urgent." It sounded like a legal notice. Two of the ten paid. The others went quiet, and I had to call them. One said, "I got your message, but it felt very aggressive, so I thought I'd wait for you to call."
The lesson: firm is fine. Formal-firm is not. Claude found the line. Gemini crossed it.
Can I use this for Shopify or WooCommerce B2B orders?
Yes, and it's easier than you think. If you're running a Shopify or WooCommerce store that sells B2B (bulk orders, trade accounts, 15- or 30-day payment terms), you're already exporting overdue invoice lists as CSVs. You just need to feed that list into Claude or Gemini and generate reminders in bulk.
I tested this with a friend who runs a WooCommerce store selling restaurant supplies—plates, cutlery, napkins—mostly to small cafés and cloud kitchens in Bangalore and Hyderabad. She had 40 overdue invoices, all B2B trade accounts. We exported the WooCommerce order list (Order ID, customer name, amount, due date, email, phone), pasted it into a spreadsheet, and wrote a prompt for Claude:
You're writing payment reminders for overdue B2B invoices. Each reminder should be under 60 words, friendly but direct, include the order number and amount, mention UPI or NEFT depending on the amount (UPI for under ₹50K, NEFT for over ₹50K), and end with a question that prompts a reply. Write one reminder per row.
We pasted the spreadsheet into Claude (using Kryotta's multi-model workspace, which lets you compare Claude and Gemini side-by-side without switching tabs), and Claude generated 40 reminders in about 90 seconds. We copied them into a WhatsApp Business broadcast list and sent them out. Eighteen customers paid within four days. She'd been chasing some of those invoices for three weeks.
The key: don't just paste the invoice number and amount. Give Claude or Gemini the customer's name, what they bought, and any context (repeat customer, first-time buyer, previous late payment). The more context, the more human the reminder sounds.
Should I always use Claude for reminders, or does Gemini have a place?
Use Claude for reminders where tone matters—overdue invoices, payment follow-ups, anything sent to a human who might take offense or ignore you. Use Gemini for high-volume, low-stakes reminders where you just need to confirm something or nudge someone gently: "Your order shipped today, here's the tracking link" or "Your GST return is due in 5 days, here's the portal link."
I still use Gemini for one thing: generating the first reminder, the one you send on the due date before the invoice is actually overdue. Gemini's polite, neutral tone works well there. "Hi Ramesh, just a heads-up that invoice #PUN-2903 for ₹34,500 is due today. UPI to 9823xxxxxx or NEFT details below. Let me know if you need anything." It's not trying to collect money yet, just confirming. Gemini handles that fine, and it's faster (Gemini Flash processes prompts about 30 per cent quicker than Claude Sonnet 4.5 in my tests).
But once the invoice is overdue, switch to Claude. The 40 per cent faster payment rate I saw wasn't a fluke—it held across three months of follow-up reminders I sent after the original test. Claude's reminders consistently got replies sooner, and replies meant payments.
One more thing: if you're sending reminders in bulk (more than 20 at a time), use Kryotta's compare mode. Paste your prompt into Claude and Gemini at the same time, see both outputs, pick the better one for each customer. Some customers respond better to Gemini's formality (older business owners, CA firms, government contractors). Most respond better to Claude's directness. You'll learn the pattern after ten reminders.
Questions people ask
Can I just copy-paste these prompts, or do I need to change them for my business?
Change them. The prompts I used included my business type (industrial supplies), typical invoice amounts (₹15K–₹2L), customer types (contractors, workshops), and payment methods (UPI, NEFT). If you sell SaaS subscriptions or run a consulting practice, your reminders need different language. Give the model your context: what you sell, who owes you money, why they might be late, what payment methods you accept. The more specific your prompt, the less generic the reminder.
Does this work for retail customers, or just B2B?
B2B only. Retail customers (individuals buying on EMI, pay-later apps, credit cards) are covered by different rules and expectations. You're also probably not writing those reminders yourself—Razorpay, Paytm, or your lender's system handles it. This approach is for business-to-business invoices where you're writing the reminder, sending it yourself, and the customer knows you.
How do I avoid sounding like a robot when I send 50 reminders in one day?
Don't use the same reminder for everyone. Even if you're sending 50 reminders, group them by customer type (repeat customers, first-time buyers, large vs small invoices, overdue by 10 days vs 40 days) and write 4–5 different prompts. Let Claude or Gemini generate variations. If every reminder sounds identical, customers will notice, especially if they talk to each other. I learned this the hard way when two contractors in the same industrial estate compared my messages and one called to ask if I'd "automated collections."
Which Kryotta model should I use if I want to test this myself?
Start with Claude Sonnet 4.5 for the reminders themselves—it's the best at tone and context. Use Gemini Flash or Gemini Pro to generate the first-pass list of overdue invoices if you're pulling from a messy spreadsheet or need to clean up data quickly. If you're on a budget, try Llama 3.3 70B—it's free in Kryotta and handles straightforward reminders well, though it's less consistent with tone than Claude. Don't use the smallest models (Haiku, Flash Lite, GPT-OSS 20B) for reminders; they're too terse and sound automated.
I still write reminders by hand for our top five customers—the ones who account for 40 per cent of our revenue. But for everyone else, Claude writes them now, and I've cut my overdue collection time from three hours a week to about 30 minutes. If you're chasing overdue GST invoices and your current reminders aren't working, try this. Write two versions of your next ten reminders—one with Claude, one with Gemini—send them, and track who pays faster. You'll know which one works for your customers within a week.
Try Claude and Gemini side-by-side in Kryotta and see which one your customers actually respond to.
