AI adoption beyond the Silicon Valley playbook

Most of the AI advice you read online was written for one kind of company.

The power never goes off. Everyone works on a fast laptop with fibre internet. Software is priced in the same currency the company earns in. Customer records sit neatly in a CRM. And there is an IT team to sort out the details.

That company exists. But it is not most companies.

Across Africa, South and Southeast Asia, Latin America and the Middle East, and in plenty of small businesses everywhere else, the picture looks different. Power and networks are not always there when you need them. Much of the team works from phones. Dollar-priced software is a real budget decision. And important business information lives in WhatsApp chats, spreadsheets on one person's laptop, and paper files.

None of this means AI is not for these businesses. It means the way you adopt it has to fit the way they actually work. Copying a playbook built for a well-funded office usually ends in tools nobody uses and subscriptions nobody can justify.

We work from Nigeria, which has its share of these challenges. This post is about what we have learned about AI adoption when you take those realities seriously, wherever you are.

1. When the power or network fails, so does your AI

Almost every useful AI tool runs in the cloud. It needs a working device, a working connection, and power for both.

In many markets, none of these is guaranteed. Outages are not only a problem in places with low electricity access. In Morocco, which has full electrification, a World Bank survey still found that more than 20% of firms had experienced power outages in the past year. In Sub-Saharan Africa, more than half of firms rely on their own generators. In Nigeria, about 70.7% of firms own or share one, and outages cost businesses about 3% of their annual sales.

If a key process depends on an AI tool being available every minute, it will break whenever the power or network does. A support team that can only reply with AI help is stuck the moment the connection drops.

What works better:

• Use AI for work that can wait a few hours. Drafting reports, summarising documents, preparing proposals and cleaning data do not need to happen live. Schedule them for when power and network are stable.
• Keep a manual fallback. Any process that uses AI should still work, slower, without it. AI should speed work up, not hold it hostage.
• Save outputs, not just chats. When AI produces something useful, a template, a reply script, a summary, store it where your team can reach it offline.

2. Your team works on phones, not laptops

According to the GSMA, mobile is the main way, and often the only way, that people in low- and middle-income countries get online. In Nigeria, for example, over 99% of broadband connections are mobile.

Yet a lot of AI training quietly assumes everyone sits at a desk with a big screen. Demos show three browser tabs, a spreadsheet and a document side by side. That is hard to copy on a phone with a weak signal and a data plan that has to last until the end of the month.

What works better:

• Choose tools that work well on mobile. Test them on the phones and networks your team actually uses, not on the manager's laptop.
• Train on the device people use. If your field or sales team works from phones, teach them AI on their phones.
• Think about data costs. Large uploads and long recordings eat data. Short, focused tasks are cheaper and more reliable.

3. Dollar pricing is a local cost problem

Most AI tools are priced in US dollars. For a company earning dollars, $20 a month per person is small. For a company earning naira, rupees, pesos or rupiah, the same price can be a serious line item. And when the local currency weakens, the bill goes up even though nothing about the tool has changed.

The AI companies know this. OpenAI launched ChatGPT Go, a cheaper plan at about $4.50 a month, first in India and Indonesia and then across 18 countries in Asia. Google introduced a similar lower-priced plan in more than 40 countries. Those plans exist because $20 a month does not work everywhere.

To make it concrete: in Nigeria, ChatGPT Plus cost ₦33,862.50 a month after VAT in late 2025, against a national minimum wage of ₦70,000. One person's subscription was close to half of a minimum wage salary. For a team of ten, that is over ₦4 million a year for a single tool.

Taxes add another layer. More and more countries now charge VAT or similar taxes on foreign digital services, and in some places the business paying the bill is responsible for handling that tax. It is worth checking with your accountant how this affects your AI spend.

What works better:

• Do not buy seats for everyone on day one. Start with the few people whose work will clearly benefit. Let them prove the value before you scale.
• Use free and regional tiers first. Many tools have free plans or lower-priced local options that are enough to test with. Learn what you need before paying for the top plan.
• Price AI against the work it replaces. If a subscription saves someone ten hours a month, it may be a good deal. If nobody can say what it saves, it is a cost, not an investment.
• Review subscriptions every quarter. Cancel what nobody uses. Currency swings make this more important, not less.

4. Your data lives in chats and paper files

This is the constraint people talk about least, and it may matter the most.

AI is only as useful as the information it can see. A tool that analyses your sales, answers customer questions or spots problems in operations needs your sales, your customer questions and your operations data.

In many businesses around the world, that information is scattered. Orders come in on WhatsApp. Prices are agreed in voice notes. Stock is tracked in a notebook. When WhatsApp launched its business app in 2018, it said more than 80% of small businesses in Brazil and India already used WhatsApp to reach customers. In Nigeria, a GSMA study found that more than half of small businesses selling online relied only on social media platforms like Facebook, Instagram and WhatsApp to do it.

Messaging apps are a great way to talk to customers. They are a poor place to keep business records. Information there is hard to search, hard to share across a team, and almost impossible for an AI tool to use in a structured way.

So for many businesses, the most valuable first AI project is not really an AI project. It is getting key information out of chats and notebooks and into one simple, organised place.

The good news is that AI can help with that step:

• Turn paper into data. Staff can photograph handwritten records, and AI can turn them into tables. Someone should still check the results, but it is far faster than typing.
• Pull structure out of chats. Most messaging apps let you export conversations. AI can extract orders, complaints, prices and follow-ups from those exports.
• Build better habits going forward. Once you know what information matters, set up a shared sheet or simple tool to capture it as it comes in.

A business with organised data and no AI is in a better position than a business with expensive AI and no organised data.

5. Be careful what you paste into AI tools

When data is scattered and AI tools are easy to use, people copy and paste. A customer list goes into a chatbot to draft messages. A contract goes in to be summarised. A payroll sheet goes in to be analysed.

That convenience carries real legal risk. By March 2025, 172 countries had data privacy laws, from Europe's GDPR to Brazil's LGPD to Nigeria's Data Protection Act. Many of these laws limit sending personal data to other countries without proper safeguards, and most popular AI tools are run by companies in a handful of countries, on servers far from where your customers live.

This does not mean you cannot use AI with business information. It means you need a few clear rules:

• Which tools staff are allowed to use for work
• What kind of information must never go into them, such as customer details, staff records or financial data
• Who to ask when someone is not sure

A one-page AI policy is enough for most small and mid-sized businesses. What matters is that it exists and people know about it. For anything involving sensitive data, get proper legal advice for each country you operate in.

What good adoption looks like

Put this together and a clear pattern forms. AI adoption that works in the real world tends to be:

1. Offline-tolerant. It speeds work up when power and network are there, without breaking the business when they are not.
2. Phone-first. It fits the devices and data plans your team actually has.
3. Locally budgeted. It is priced in the currency you earn in, starts small, and grows only when the value is clear.
4. Data-first. It begins by getting important information into one organised place, so AI has something useful to work with.
5. Rule-guided. It comes with simple rules about what goes into AI tools and what does not.

None of this is glamorous. But it is how AI actually becomes useful in businesses that run on generators, mobile data and messaging apps, which describes a very large share of the world's businesses.

There is also an upside that rarely gets mentioned. Businesses that operate under these constraints are already good at working around them. Their teams improvise, adapt and get things done with what is available. That same skill, applied with some structure, is exactly what good AI adoption needs.

The question is not whether AI can work for your business. It is whether you adopt it in a way that fits how your business really runs.

If you are trying to work that out, talk to us at Lampstand. We help businesses figure out where AI fits, what is worth paying for, and how to make it stick.