Why there is no fixed price list for AI automation
Website work can be packaged because the deliverable is predictable: a number of pages, a design, a contact form. Automation cannot, because the cost is driven almost entirely by what already exists inside the business. Two companies can ask for the same outcome and need completely different amounts of work:
- A company whose leads all arrive in one web form and go into one spreadsheet is a short build.
- A company whose leads arrive by WhatsApp, phone, Instagram DM and a form, and need to reach three different people depending on the product, is a much bigger one.
The automation itself is often the small part. Connecting reliably to the systems already in use is the work.
What actually drives the cost
| Cost driver | Cheaper | More expensive |
|---|---|---|
| Number of systems | One source, one destination | Several tools that must stay in sync |
| Integration type | Tools with a documented API | Legacy software, or no API at all |
| Decision-making | Fixed rules ("if X then Y") | An agent that judges each case |
| Data quality | Clean, structured records | Free text that must be interpreted first |
| Volume | Dozens of events a day | Thousands, needing queues and retries |
| Failure tolerance | Internal admin tasks | Anything touching payments or customers |
There is also a running cost that is easy to miss at quoting time. Automations that call an AI model pay per use, so a workflow processing 50 enquiries a day costs more to run than one processing five. Any honest quote should separate build cost from monthly running cost.
A realistic way to budget for a first project
The most reliable approach is not to budget for "AI" at all. Budget for a single, measurable outcome, and pick the one where manual effort is most obviously wasted:
- Pick one workflow that repeats daily. If it happens once a month, automating it rarely pays back.
- Count the hours it currently consumes across everyone who touches it, per week.
- Multiply by an hourly cost to get what the manual version costs you per year.
- Compare that to the build quote plus 12 months of running cost. If the payback period is under a year, it is usually worth doing.
This is also the fastest way to tell whether a vendor is being straight with you. Anyone who cannot tell you which workflow they are automating and what it currently costs you is selling a tool, not a result.
Where the money is usually well spent in Rwanda
Across the businesses we speak to in Kigali, the same handful of automations come up repeatedly because the manual version is genuinely expensive:
- Lead capture and routing — especially where enquiries arrive on WhatsApp and get lost in a personal phone. See our guide on automating WhatsApp lead follow-up.
- Follow-up sequences — most lost sales in small businesses are not rejections, they are enquiries nobody chased a second time.
- Reporting — weekly numbers assembled by hand from two or three systems.
- Appointment and scheduling flows — including reminders, which cut no-shows.
- Invoice and payment reconciliation — particularly where mobile money and bank records have to be matched manually.
Automation versus an AI agent
These are priced differently because they are different things. An automation follows a fixed path you defined. An agent decides what to do based on the situation, which means it needs more careful design, testing and guardrails — and costs more to build. Most businesses should start with automation and only move to an agent where the decision genuinely varies case by case. We cover the distinction in what is an AI agent.
Questions to ask before you sign anything
- Which exact workflow are you automating, and what does it cost us manually today?
- What is the build cost, and separately, what will it cost per month to run?
- What happens when it fails — do we get alerted, or does it fail silently?
- Who owns the accounts and the workflow files if we part ways?
- Can you show it running against our real data before we pay the balance?
That last one matters more than anything else. An automation that works in a demo and breaks on real data is the most common way this goes wrong.