Why African SMEs need custom AI, not SaaS.
Off-the-shelf tools assume your tax code, your currency, your branch structure, and your support hours. Most don't match what an Accra retailer or a Lagos clinic actually needs. Here's the real arithmetic.
The pitch for SaaS sounds great in a Stripe ad. Pay a monthly fee. Log into a dashboard. Your problem is solved. For an Accra retailer with five branches, three internet outages a week, and revenue counted in cedis, that promise quietly falls apart on contact with reality. The contract is in dollars. The tax module doesn't know what SSNIT is. The cashier app freezes when fibre drops at 11am. And when something breaks at month-end, the support team is asleep in California.
This essay is the case for the alternative. Not a takedown of SaaS — there are real cases where SaaS is the right answer, and I'll get to those. But for most African small and medium businesses, the default assumption that "we should buy a SaaS tool" is wrong. The default should be a custom system, hosted on your own infrastructure, with AI agents woven through the parts that used to need a person.
Here's why.
The arithmetic doesn't work
Most enterprise SaaS prices in dollars per seat per month. The numbers vary, but the shape is consistent: $30 to $150 per seat, on the lighter end for HR and CRM, on the heavier end for ERP and BI. A 50-person company easily lands at $1,500 to $7,500 per month, per tool. And nobody runs on one tool. A typical mid-sized retailer needs an HR system, a payroll system, a POS, an accounting system, a CRM, and a customer-support tool. Six tools at six subscriptions.
The annual bill, in dollars, often clears $25,000–60,000. Translated into cedis at current rates, that's GHS 270,000 to 700,000 a year. Recurring. Forever.
For comparison: one mid-level developer in Accra building a custom integrated stack on a $50/month VPS — POS, payroll, attendance, CRM, with AI agents on top — costs roughly $8,000 to $25,000 once, plus a few thousand a year in maintenance and infrastructure. The break-even, even being generous to SaaS, is somewhere around month nine.
The break-even between SaaS and custom build, even being generous to SaaS, is somewhere around month nine. After that, every month the gap widens in your favour.
This isn't theoretical. This is the math we run for every prospect that walks in. The numbers don't lie, and they almost always tell the same story.
SaaS doesn't speak your country's language
Tax law is the cleanest example. In Ghana, payroll has to compute PAYE on a graduated scale, SSNIT at 5.5% employee and 13% employer (with a Tier 2 mandatory pension on top), plus optional Tier 3 contributions. Overtime has weekday and weekend bands. Allowances are taxed differently from basic salary depending on type. Every June, the Ghana Revenue Authority publishes new bands.
Now go open a popular international payroll SaaS and try to configure that. You'll find a "tax rules" section that lets you set a flat percentage. You'll find a place to upload a CSV of employee benefits. You will not find SSNIT Tier 1 vs. Tier 2 vs. Tier 3 fields. You will spend a week trying to bend the system into shape, and the result will still be wrong — because the system was built for a tax regime that doesn't exist here.
The same is true going country by country. Nigerian PAYE works differently. Senegalese payroll works differently. Kenyan NHIF and NSSF look nothing like SSNIT. Most international SaaS handles the US, the UK, parts of the EU, and "everywhere else" lumped into one generic "international" mode that doesn't actually compute anyone's taxes correctly.
A custom system computes whatever you tell it to compute. When the GRA publishes new bands in June, you change four lines of code and you're done.
SaaS assumes always-on internet
Most modern SaaS is browser-based with no real offline mode. If the internet drops in your branch, the cashier can't take payments. The receptionist can't pull patient records. The driver can't print a delivery note. You're not just slow — you're stopped.
An Accra outage that lasts forty minutes is forty minutes of zero revenue at every checkout. Multiply by branches, multiply by frequency, and the cost in lost transactions exceeds the entire SaaS subscription many times over.
Custom systems built for this environment cache aggressively. The POS keeps a local copy of the product list, queues transactions while offline, and syncs when connectivity returns. The payroll app runs on your VPS and you can hit it over the local network even when the international link is down. This isn't exotic engineering — it's just engineering that respects where the user actually is.
Your data is hostage
SaaS providers will tell you the data is yours. Try to leave one and find out what that means in practice. The export is a CSV that loses every relationship between tables. Attachments don't come down. History gets flattened. Custom fields disappear. You spend three months reconstructing what you had, and the new vendor charges you migration fees on top.
When the system is yours, on your VPS, the database is a real PostgreSQL or MySQL instance with all the foreign keys intact. You can move it to a new server in an afternoon. You can run analytics on it directly. You can give your accountant SQL access if you want. Nobody is gatekeeping your own information.
Support lives in the wrong time zone
"24/7 support" is one of those phrases that sounds reassuring until you need it. In practice it usually means a US or EU helpdesk that responds in eight hours. When payroll is due tomorrow morning and the import is broken tonight, eight hours is a lifetime.
The team that built your custom system is the team that fixes it. In Accra. Same time zone. They know your data because they designed your data. A bug that would take a SaaS support team three days of escalation across two continents takes the actual developer twenty minutes.
When SaaS is the right answer
I want to be honest here, because the goal isn't to pretend SaaS is always wrong. It isn't. There are real cases where SaaS wins, and a consultancy that won't admit them isn't being straight with you.
- Genuinely commodity problems. Email hosting. Calendar. Video calls. File storage. Use Google Workspace or Microsoft 365. Don't reinvent these.
- Truly global operations. If your business is multi-country with steady internet, dollar-denominated revenue, and a budget that doesn't blink at $50k of subscriptions, SaaS abstraction is often worth the price.
- Highly regulated, audited workflows. If you need SOC 2 compliance for international clients tomorrow, building it yourself is slower than buying a tool that already has it.
- Pre-revenue startups. Don't build payroll software for a four-person team. Use Paystack and a spreadsheet. Build only when scale forces you to.
The pattern: SaaS wins when the problem is commodity, the budget is dollars, and your operating environment matches the assumptions of the vendor. The moment any of those conditions break — you're operating in cedis, you have country-specific compliance, your internet is intermittent, your team scales differently — the math flips.
The custom AI path, in concrete terms
"Custom AI" doesn't mean building everything from scratch. That's a different mistake — the kind that takes two years and never ships. The path that actually works is much more boring:
- A standard open-source backend stack: PostgreSQL or MySQL, Node.js or Python, served by Nginx on a VPS that costs $50 to $200 a month.
- A custom front-end built specifically for your workflow — your branches, your roles, your forms, your reports — not a generic dashboard you have to configure.
- AI agents wrapped around the parts that used to need a human: a customer support agent on your support inbox, an HR agent on your attendance database, a finance agent on your daily reconciliation.
- Everything talking to one chat layer your team already uses, so the AI fits into existing habits instead of demanding new ones.
That's it. That's the stack. There's nothing magic about it. The reason it works is that every component is sized to the actual business, not to a generic mid-market American company that doesn't exist here.
What to do this week
If you're running a 30 to 200-person African business and currently writing checks for SaaS subscriptions every month, the honest exercise is fifteen minutes long: list every monthly software subscription, multiply by twelve, convert to your local currency. That number is the budget you have available to commission a custom system. Most owners are surprised by how big it is.
Then ask: would I rather rent that capability forever, or own it once?
For a growing business, the answer is increasingly the second one. The first generation of African companies that figure this out are about to have a structural cost advantage their SaaS-renting competitors will not be able to close.