Myth one: AI writes your entire product catalogue while you sleep
I spoke to a fashion vendor in Surulere who'd paid ₦15,000 for a "full AI e-commerce setup" from someone on Twitter. The promise: feed the tool your inventory list and wake up to perfect product descriptions for Instagram and WhatsApp. What she actually got: 200 descriptions that all said "high-quality fabric, available in multiple sizes, perfect for any occasion." Every single one. She spent two days rewriting them by hand.
Here's what AI actually does for e-commerce in Nigeria: it handles the repetitive admin that clogs your day. Paystack dispute responses when a customer claims they never received an order even though your rider handed it over in Lekki Phase 1. Bulk replies to the same three questions you get on WhatsApp—"Do you have this in blue?", "What's your delivery fee to Abuja?", "Can I pay on delivery?" Template captions for product posts so you're not staring at a blank screen every time you photograph new stock. It doesn't write your brand voice. It doesn't know which detail makes your ankara print dresses different from the vendor two streets over. It saves you an hour a day on the boring parts so you can write the pitch that actually converts.
I tested this with a home décor seller who runs her entire business through WhatsApp status updates and DMs. She was spending ninety minutes every evening replying to variants of "Is this still available?" and "How much to deliver to Ikeja?" We set up a simple prompt in Claude Sonnet 4.5: take the product name, price, and delivery zone, then write a three-sentence reply confirming stock and giving a delivery estimate. She ran it for two weeks. Her reply time dropped from an hour to fifteen minutes. Her conversion rate didn't change, because the replies weren't the bottleneck—her product photos and pricing were. AI gave her time back. It didn't make her rich.
Myth two: healthcare AI diagnoses patients (and opens you to lawsuits)
A doctor I know in Abuja heard "AI in healthcare" and immediately thought liability. He imagined a chatbot telling a patient they had malaria based on three symptoms, the patient skipping a proper test, and a lawsuit landing on his desk. He wasn't wrong to be cautious, but he'd misunderstood what the software does.
AI in Nigerian healthcare admin doesn't touch clinical decisions. It handles prior authorization forms for HMOs, appointment reminders via WhatsApp, patient intake questionnaires before the consultation, and insurance claim documentation. These are the tasks that take up two hours of a nurse's morning and add zero clinical value. A private clinic in Victoria Island started using Gemini Flash to draft prior authorization letters for diagnostic tests—scans, blood panels, specialist referrals. The doctor still reviews and signs every letter, but the model pulls the patient history, formats the request, and lists the clinical justification in under a minute. The clinic went from submitting six authorization requests a day to fifteen, which directly increased how many patients they could see.
The myth costs you money because you either avoid AI entirely (and drown in paperwork) or you buy a tool that promises diagnostic magic and delivers nothing your clinic can legally use. The reality is boring and useful: AI is a very fast typist that knows medical documentation format. It's not a doctor.
Myth three: real estate AI generates legal contracts you can file tomorrow
A property agent in Port Harcourt told me he'd tried an AI contract generator he found online. He fed it details of a lease—property address, tenant name, monthly rent, deposit—and got back a three-page document that looked professional. He showed it to a lawyer friend, who pointed out that it was missing the mandatory clauses under Rivers State tenancy law, used British English phrasing that wouldn't hold up in a Nigerian court, and had the landlord's liability section copied from a template written for California.
AI doesn't write legal documents you can file. It writes first drafts that a lawyer turns into legal documents. The distinction matters because the cost and the risk sit in different places. If you're a real estate agent juggling ten properties and you need to send a rental agreement to a new tenant tonight, you have two options: pay a lawyer ₦25,000 to draft it from scratch, or spend twenty minutes with DeepSeek V3 generating a structure you then send to the lawyer to review and correct for ₦8,000. The second option is faster and cheaper, but only if you know you're buying a draft, not a contract.
The same applies to title verification summaries, tenancy dispute letters, and property sale agreements. AI pulls the standard clauses, formats the document, fills in the names and numbers. A lawyer checks it against current law, adds the jurisdiction-specific language, and confirms it won't get thrown out if you end up in court. I've watched agents waste money both ways—paying full legal fees for boilerplate work, or trusting an AI draft without review and then spending double to fix it later when the tenant's lawyer spots the gaps.
Myth four: restaurant AI replaces your cashier and takes orders on Instagram
A suya spot in Ikeja saw a demo video of an AI voice assistant taking orders over the phone. The owner thought: this could handle our Instagram DMs, our WhatsApp orders, and maybe even our in-person queue. He paid ₦50,000 for a setup. It lasted three days. The voice assistant couldn't understand Lagos accents, the Instagram integration didn't work because the vendor had shown him a US tool that doesn't support Nigerian payment rails, and the WhatsApp bot kept asking customers to "add to cart" when everyone just wanted to send a voice note saying "two wraps, extra pepper, GRA delivery."
AI in Nigerian restaurants does two things well: it tracks inventory so you know when you're low on chicken before the Friday rush, and it handles the repetitive customer service questions that clog your DMs. "Are you open now?", "Do you deliver to Magodo?", "What's on the menu today?" A bot can answer those in one second. It can't replace the person who knows a regular wants his jollof rice a little burnt, or who talks a first-time customer through the menu because they've never ordered amala before.
The cost myth here is that automation means fewer staff. It doesn't. It means your staff stop spending forty minutes a day typing "Yes, we're open until 10 p.m." and start spending that time on the work that actually needs a human—cooking, packing orders correctly, handling the customer who's annoyed their delivery is late because of Third Mainland Bridge traffic. If you're buying AI to cut your wage bill, you'll be disappointed. If you're buying it to let your team do better work, it pays for itself.
Myth five: Paystack dispute AI writes responses the card issuer will believe
Paystack disputes are where most Nigerian e-commerce sellers first think about AI, because the volume is exhausting and the stakes are real. A customer claims they never received the order. You have the rider's delivery photo, the customer's WhatsApp confirmation that they got it, and a tracking log. You need to write a response that convinces the card issuer to rule in your favour, and you need to do it for fifteen disputes this week.
The myth: AI writes a response, you submit it, you win. The reality: AI writes a response that includes all the right information in a structure the issuer expects, but you still have to check that the evidence matches the claim, the tone isn't defensive, and the timeline makes sense. I tested this with a Shopify seller in Lagos who gets eight to twelve disputes a month, mostly "item not received" and "unauthorized transaction." We used Claude Sonnet 4.5 to generate responses. The model pulled the order number, the delivery date, the tracking reference, and the customer's chat history, then formatted it into a three-paragraph reply with evidence attached.
The seller won eleven out of twelve disputes that month. The one she lost was because the tracking showed delivery to a different address than the customer had provided—the AI didn't catch that discrepancy, but she should have when she reviewed the draft. The tool saved her six hours that month. It didn't think for her.
The cost mistake is paying for a tool that promises "automated dispute resolution" when what you actually need is "faster drafting with human review." The first doesn't exist in any reliable form. The second costs you thirty minutes a week and a Kryotta subscription instead of a full evening writing replies from scratch.
Myth six: AI integration means rebuilding your entire workflow around new software
A CA in Ikoyi told me she'd looked into AI for client invoicing and bookkeeping reminders, but the demo she sat through assumed she was using QuickBooks, Xero, and Stripe. She uses Excel, bank transfer confirmations via email, and a WhatsApp group with her clients. She closed the tab and decided AI wasn't for her.
This is the myth that costs the most, because it stops people from starting. You don't need to rebuild your workflow. You need a model that works with the tools you already use. If your client communication happens on WhatsApp, you need AI that drafts WhatsApp messages. If your payment confirmations come through Paystack or Flutterwave, you need a model that reads those emails and updates your records. If your contracts live in Google Docs, you need AI that edits Google Docs.
I work with a legal researcher who spends half her day summarizing case law for senior lawyers. She was using Gemini Flash in a browser tab, copying text from PDF judgments, pasting it into the chat, then copying the summary back into Word. It worked, but it was slow. We moved her to a workspace in Kryotta where she could upload the PDF, run the summary, and export it directly. Same model, same output, half the time. She didn't change her filing system or learn new software. She just stopped doing the same copy-paste loop forty times a day.
The integration myth makes you think AI is for businesses that already run on software. The reality is that AI is most useful for businesses still running on WhatsApp, Excel, and hustle, because those are the businesses where an hour saved actually changes what you can earn that week.
Questions people ask
Does AI actually save money if I'm already paying Paystack fees and a VA? It depends what the VA is doing. If they're writing unique responses to complex customer questions, AI won't replace that. If they're copying and pasting the same reply to "Do you deliver to Ajah?" fifty times a day, AI does it faster and costs less. The Paystack fees don't change—AI just helps you win more disputes so you're not losing the transaction fee plus the product cost.
Which model should I start with if I've never used AI before? Claude Sonnet 4.5 or Gemini Flash. Sonnet is better at writing tone-accurate replies for customer service and disputes. Flash is faster and cheaper for high-volume repetitive tasks like appointment reminders or inventory summaries. Try both for a week and see which one feels easier.
Can I use AI if my business runs entirely on WhatsApp and Instagram DMs? Yes. You're not automating the DMs themselves—you're using AI to draft replies faster, write captions, summarize customer questions at the end of the day, or generate product descriptions before you post. You still send everything manually. You just stop staring at a blank screen for twenty minutes every time you need to write something.
Is it legal to use AI for healthcare admin or legal documents in Nigeria? For healthcare: yes, as long as a licensed professional reviews anything that touches patient care. For legal: yes, as long as a lawyer reviews and signs anything you're filing or sending to a client. AI is a drafting tool, not a replacement for professional liability.
AI won't write your brand voice or diagnose your patients, but it will give you back the hours you're losing to repetitive admin. If you're spending more time on Paystack disputes, WhatsApp replies, or document formatting than you are on the work that actually grows your business, you're paying for the myth instead of using the tool. Try one model for one task this week—dispute response, product caption, patient intake form—and see whether it saves you an hour. If it does, keep going. If it doesn't, you've lost thirty minutes instead of ₦50,000 on software you'll never open again.
Start with a task you're already doing badly or slowly, and let the AI draft it while you fix the parts it gets wrong. That's the workflow. You can try it today at Kryotta with the models that work for Nigerian businesses—Claude, Gemini, DeepSeek—without rebuilding anything you're already doing.
