The question no one asks before "trying AI"
I've watched three business owners this month sign up for AI tools, use them twice, then forget they exist. A boutique owner in Lekki who wanted help with Instagram captions. A clinic administrator in Abuja who thought AI could "handle scheduling". A restaurant manager in Ibadan who heard AI writes menus now.
All three had the same problem: they didn't know what AI actually does in their industry. They'd seen the demos—chatbots writing essays, models generating images—but no one showed them what a property agent in Lagos or a chartered accountant in Port Harcourt does with these tools on a Tuesday afternoon.
So here's the comparison no one's written yet. Six Nigerian industries, side by side, with the tasks AI genuinely saves time on and the ones where you're better off doing it yourself.
What AI actually does in six industries (the table everyone needs)
| Industry | Task AI handles well | Time saved per week | Task AI struggles with |
|---|---|---|---|
| E-commerce | Product descriptions for 50 listings; WhatsApp reply templates for common questions ("Is this in stock?", "Do you deliver to Enugu?") | 4–6 hours | Negotiating bulk orders; handling angry customers who want refunds |
| Education | Lesson plan outlines for SS2 Biology; summarising 30 student essays into grade categories and common mistakes | 3–5 hours | Actual grading with nuance; parent-teacher meeting notes that need empathy |
| Real estate | Property descriptions from photos and specs; tenant screening summaries from application forms | 2–4 hours | Viewing schedules across Lagos traffic; rent negotiation conversations |
| Healthcare admin | Appointment reminder messages; patient record summaries for handover between shifts | 3–4 hours | Triage decisions; explaining diagnoses to worried relatives |
| CA / Legal | Contract review checklists (missing clauses, inconsistent terms); compliance document summaries | 5–8 hours | Legal strategy; client advice on grey-area tax questions |
| Restaurants | Menu descriptions and daily specials for Instagram; delivery coordination messages to riders | 2–3 hours | Supplier negotiations; managing kitchen staff schedules |
The pattern: AI is excellent at turning structured information into clear text and drafting repetitive messages that still need a human check. It's terrible at anything involving negotiation, emotion, or real-time decisions where context shifts every minute.
E-commerce: where AI pays for itself in week one
If you sell on Instagram or WhatsApp—clothes, skincare, phone accessories, anything—you already know the bottleneck. You've got 200 product photos on your phone. Writing descriptions for each one takes an hour you don't have, so half your posts say "Available, DM for price" and you lose sales to sellers who actually describe the thing.
What works: Upload a product photo to Gemini Flash with this prompt: "Describe this product for an Instagram post. Include material, size, colour, and one benefit. Keep it under 40 words. Tone: friendly, Lagos vendor." You'll get something like: "Chiffon midi dress in burnt orange, fits UK 10–14. Breathable fabric, perfect for Lagos heat. Side zip, lined. ₦8,500. Free delivery within Lekki–Ajah."
You edit the price, check the size, post. Thirty seconds per product instead of five minutes.
Where it saves time: Product listings, WhatsApp auto-replies for stock checks, Instagram captions for new arrivals. One e-commerce seller I know processes 40 listings every Sunday night in two hours instead of six.
Where it doesn't: Customer replies that need judgment. Someone asks, "Is this dress true to size or should I go up?" AI doesn't know your supplier's quirks. You do.
Healthcare admin: the unglamorous time-saver
Doctors and clinic admins don't need AI to diagnose. They need it to stop drowning in appointment confirmations, patient handover notes, and insurance paperwork.
What works: Patient record summaries. A clinic administrator in Abuja uses Claude Sonnet 4.5 to summarise ten patient files at the end of each shift—symptoms, medications, follow-up needed—so the next shift doesn't waste twenty minutes reading notes. Prompt: "Summarise this patient record for handover. Include current symptoms, medications, and any follow-up tasks. Keep it under 100 words."
Appointment reminders: instead of typing "Your appointment with Dr Adeola is tomorrow at 3pm, please confirm", she generates twenty reminders in one go through Llama 3.3 70B and copies them into WhatsApp. Five minutes instead of forty.
Where it saves time: Routine admin text that follows a template but needs personalising. Record summaries, reminders, referral letters to other doctors.
Where it doesn't: Anything clinical. Triage, diagnosis, patient conversations. AI can draft a reminder; it can't tell a patient their test results over the phone.
Restaurants: menu updates and rider coordination
Restaurant owners spend more time on WhatsApp than in the kitchen. Coordinating delivery riders, updating daily specials, replying to reservation requests. The food part is fine. The admin part eats the day.
What works: Menu descriptions. A restaurant in Ibadan uses Mistral Large to write Instagram captions for daily specials. Prompt: "Write a 30-word description for [dish name]. Mention key ingredients and why it's good today. Tone: warm, makes people hungry."
Delivery coordination: instead of typing "Please pick up order #47 from our Bodija branch and deliver to [address] by 7pm", the manager generates five messages at once, checks addresses, sends. GPT-OSS 20B handles this—it's fast and cheap for short, repetitive text.
Where it saves time: Menu updates, delivery messages, Instagram posts for specials. Two to three hours a week, easily.
Where it doesn't: Supplier negotiations, staff schedules, handling complaints. A customer says their jollof was cold—you're not handing that to a model.
CA and legal: the surprise winner
Chartered accountants and lawyers have the most to gain and the least faith it'll work. They've seen too many tools promise to "automate legal work" and deliver generic rubbish.
But document review—the actual reading and checking, not the legal judgment—is where AI is quietly brilliant.
What works: Contract review checklists. A CA in Lagos uses DeepSeek R1 to review client contracts before she does. Prompt: "Review this contract. List any missing clauses, inconsistent terms, or unclear obligations. Don't give legal advice—just flag what needs a closer look."
R1 is a reasoning model, so it shows its working. It'll say, "Clause 5 mentions payment within 30 days, but Clause 9 says 45 days—inconsistency." She fixes it in two minutes instead of twenty.
Compliance checklists: "Does this document meet FIRS requirements for expense claims?" Claude Sonnet 4.5 reads the doc, checks it against a compliance template she provides, flags gaps. She reviews, signs off.
Where it saves time: First-pass document review, compliance checks, summarising 60-page contracts into two-page briefs for clients. Five to eight hours a week.
Where it doesn't: Legal strategy, client advice, anything involving judgment. AI spots the inconsistency; you decide if it matters.
When to use which model (the verdict no one gives you)
You don't need to test fifteen models. You need three, matched to task type.
Customer-facing text (product listings, Instagram captions, WhatsApp templates): Use Gemini Flash or Llama 3.3 70B. Both are fast, cheap, and good at short, friendly text. Flash is slightly better at reading images if you're generating descriptions from photos.
Document analysis (contracts, patient records, compliance checks): Use Claude Sonnet 4.5 or DeepSeek R1. Sonnet reads long files well and writes clear summaries. R1 is better for spotting logical problems—inconsistencies, missing steps, unclear terms—because it reasons out loud.
Repetitive admin (reminders, coordination messages, daily updates): Use GPT-OSS 20B or Mistral Large. Both handle short, templated text at speed. OSS is cheaper; Mistral is slightly better at tone.
One more thing: don't route everything through one model because it's the "best". The best model for writing a property description is not the best model for reviewing a 40-page contract. Match the tool to the task.
Questions people ask
Can AI actually understand Nigerian business contexts, or does it default to US examples?
It defaults unless you tell it not to. Add "Lagos vendor", "Nigerian clinic", or "Abuja-based CA" to your prompts. Mention naira, WhatsApp, Paystack if relevant. The model adjusts. I've tested this with property descriptions—"Lekki apartment" gets you better results than "apartment" alone.
Which industries save the most time with AI?
CA and legal work, by hours saved per week. E-commerce saves less total time but pays for itself faster because product listings directly drive sales. Healthcare admin and restaurants sit in the middle—solid time savings, but the tasks aren't revenue-critical.
Do I need to learn prompt engineering, or can I just type normal requests?
Somewhere in between. You don't need to learn a new language, but you do need to be specific. "Write a product description" gets you generic fluff. "Write a 40-word product description for a chiffon dress, include material and size, tone: friendly Lagos vendor" gets you something you can use. Specific beats clever.
Is this actually cheaper than hiring someone part-time?
For document review, record summaries, and product listings—yes, much cheaper. For customer service, delivery coordination, and anything involving judgment—no. You still need the person; AI just handles the repetitive 40% so they can focus on the rest.
I've worked with small business owners across Lagos, Abuja, and Ibadan who've built these workflows without a tech team or a big budget. Most of them use Kryotta because it gives them access to Claude, Gemini, Llama, DeepSeek, and Mistral in one place—no juggling subscriptions or hitting model limits mid-task. You can compare outputs side by side, route tasks to the right model, and pay in naira. Try it here and see which tasks in your industry actually save time.
