AI for Small Business in Pakistan: Results and a Plain Start

What this looks like in a Pakistani inbox

Most AI advice assumes typed English arriving by email. That is not what a business in Karachi or Lahore actually receives. Customers send voice notes. They send a screenshot of a bank transfer rather than typing the reference number. A single message starts in English and finishes in Urdu. And because so much of this market answers price questions with a kindly "DM for price", the conversation that decides the sale happens in the inbox, not on the website.

So test any setup against a week of your own real messages before you trust it with a customer. If it cannot handle a voice note asking for a wholesale rate, it will not survive Monday morning. In that inbox, AI earns its place in three ways:

• Reading and sorting incoming messages, voice notes and payment screenshots included, then drafting a clear reply, as an AI receptionist does.
• Answering the common questions from your own price list and catalogue.
• Nudging the leads that went quiet before they go cold.

Start with the work, not the tool

Most AI advice starts with a tool and looks for a use. That is backwards. Start with the work your business repeats every week: the reports someone copies by hand, the same few questions customers ask, the follow-up messages that never get sent. If the hand-copied report is your pain, an AI reporting dashboard is the shape that fix usually takes. These repeated tasks are where AI pays for itself.

What does our record of work show?

Across 25+ engagements, Wobble has shipped 120+ automations in 6 countries, run from Karachi. The table lists some of those cases as the Work page records them. Each figure belongs only to its own row, and the clients are described by industry and country because we do not name them.

Business	What was built	Recorded result
RTO business, Australia	Delivery line rebuilt as one audited automation of 34 AI nodes	94% gross margin, from under 35%
Real estate business, USA	AI voice agents in place of a human cold calling team	$114k saved per year
Eye care business, USA	AI booking behind paid search and social	82 conversations via AI booking at $0.36 cost per click
Fashion business, Pakistan	Full AI operating system with paid growth	2.46m lifetime impressions and 68k clicks, a 2.76% CTR (click-through rate, the share of people who saw an ad and clicked it)

Read these with their limits. The $114k is a saving set against the cold calling team the voice agents replaced, which is a different claim from new revenue. The 94% is a gross margin measured on the same delivery line before and after, so the comparison is clean. The eye care and fashion rows pair automation with paid advertising, and a smaller shop should not expect the same numbers. They show the kind of result that is measurable, not a promise of yours.

What is AI genuinely good at for a small business right now?

Repeated work with a predictable shape: sorting and answering messages, turning messy sheets into a summary you can act on, and drafting first versions of documents in your voice. It is weak where the task has never been written down, or where the answer depends on a relationship only you hold.

What to ignore for now

Ignore anything that promises to run your whole business on its own. Ignore tools you cannot connect to WhatsApp, your sheets or your existing systems, because an island of AI helps no one. And ignore the urge to automate something you have never written down. If a process only lives in one person's head, fix that first.

A sensible order to start

1. Pick one painful, repeated task that costs you money or time every week.
2. Write down how it works today, step by step.
3. Automate that one task and measure the result for long enough to trust it.
4. Only then move to the next task.

What should a small business with a very small budget do first?

Spend nothing on a platform. Take the one task that eats the most hours, do it by hand with a written checklist for a few days, and only then ask what automating it would cost. If the saving is not obvious on paper, the build will not repay itself. Money that has not yet been spent is the cheapest part of the plan.

Is AI cheaper than hiring someone for repeated work?

It can be, when the work is repeated often enough, and it is never free. The record shows both sides. The US real estate business saved $114k a year against the cold calling team its voice agents replaced. The RTO business in Australia spent about $4 of AI cost on each full unit pack, the audit-ready training pack that had been built by hand before.

Setup takes time, and it works best when someone who understands your business shapes it, so budget for the build and not just the subscription. A task that runs a handful of times a month will not repay a build however irritating it is to do by hand. For a fuller way to weigh it, read whether AI is worth it for a small business.

How to tell if a task is worth automating

Not every repeated task earns the setup cost. Run it through these questions first and count the yeses. All yes means go ahead. Most yes means write the process down properly and check again next month. Fewer than that, pick something else.

• It happens weekly or more often, not once a quarter.
• One person can explain it start to finish without guessing at any step.
• The information arrives in roughly the same form each time: a WhatsApp message, a form, a sheet.
• Getting it late or wrong has a cost you can put a number on.

When is a small business better off not automating yet?

There are businesses where automation is the wrong move for now. If you get a handful of enquiries a week and you personally reply within minutes, a system adds a layer without removing a problem. If every price is negotiated against a relationship you carry in your head, an automated reply will read as cold to your regulars and they will say so. If your constraint is supply rather than demand, answering faster only puts more people on a waiting list.

The other honest limit is information. AI that answers from your own material needs that material to exist somewhere other than a chat history. No catalogue, no price list, no written policy on returns means nothing to answer from, and the setup stalls there.

Where AI sits in how we work

We run the work with AI, and it is a real edge, but it is not the whole job. The result comes first: a task chosen, written down, built, and measured against what it replaced. AI is how that is done quickly. People still decide what to build, check what comes out, and approve anything involving money, pricing or your name.

Questions to ask before you pay anyone

Most of the money wasted on AI in this market goes on the arrangement rather than the technology. Ask these four questions before you sign, and write the answers down. If you are comparing outside help, our page on AI consultants in Pakistan covers how to judge it.

Our view at Wobble is that whatever runs inside your business should belong to you, the same way your delivery van does. Renting output for a one-off campaign is reasonable. The conversations your customers have with your business are a different matter, so ask where the account sits before you ask what it costs.

• Whose account holds the automations, the data and the phone number? If it is the vendor's, you are renting.
• What happens to the system if you stop paying next month? Ask for the answer in writing.
• Who fixes it late at night when the replies stop going out, and how do you reach that person?
• What is being measured, and where can you see it yourself without asking anyone?

The four questions, answered by one supplier

Those questions deserve answers rather than agreement, so here are one company's, in the same order.

Whose account holds it: yours. Customers own what Wobble builds. The accounts are registered in your name from the first day, and the workflows, credentials and data sit under your logins, so you own the system whether or not we keep working together.

What happens if you stop paying: monthly services stop any month, with no lock-in, and what we built stays yours. How you are billed depends on the build. Some projects are a one-time build, some are a monthly service, and some are both, so we say which before you commit. A new idea that comes up along the way becomes the next phase, quoted first and never added quietly.

Who fixes it late at night: we will not give a response time here that we cannot show you, so ask for the named person and how to reach them, and have it written into the agreement before you sign. What the system does by design is hand a case it cannot handle to a person, with the record attached, rather than send a reassuring sentence. Money, pricing, anything published in your name and any serious complaint wait for human approval. In a business this size that person is usually you.

What is measured, and where you can see it: the number for the one task we chose, checked against what it replaced, as with the RTO business's gross margin before and after on the same delivery line. Because the workflows and data sit under your logins, you can look at it yourself without asking us.