Home / Insights / AI adoption vs AI implementation
The two get sold as the same thing and they are not. One gets your people using AI well. The other builds AI into a process so it does the work. Buy the wrong one first and you either own licences nobody uses or a system nobody is allowed to feed.
AI adoption is getting a team to use AI well: a usage policy, approved tools on the right tier, training with examples from your own work, and a view of what it costs. AI implementation is building AI into a specific business process so it runs in production: an agent that handles account admin, a pipeline that reads invoices, a system that drafts replies for a person to send.
Adoption changes how people work. Implementation changes what the software does. Most businesses need adoption first and implementation second, and the mistake in both directions is expensive.
| AI adoption | AI implementation | |
|---|---|---|
| The question it answers | How should our people use AI? | Which process should AI do for us? |
| What you end up with | A policy, approved tools, trained staff, visible spend, one pilot | A system running on your data, measured in hours saved |
| Who does the work | Your staff, with the tool | The software, with a person approving exceptions |
| Typical first step | A one-page policy and a business-tier assistant | Document capture, enquiry triage or repetitive admin |
| Timeframe | Weeks | Six to twelve weeks for a first system |
| Main risk if skipped | Client data in free tools; subscriptions nobody uses | People stay the bottleneck; the gain never compounds |
Implementation on top of chaos adds risk. If staff are already pasting client information into whichever free tool they found, building an agent that touches the same data makes the POPIA position worse, not better. The policy and the approved tool tier have to come first, because they decide what the implementation is allowed to see.
Adoption without implementation stalls. A team that has a policy and licences but no process that AI actually runs will drift back within a quarter. The tools become a better autocomplete. The hours do not come back, because nobody built the thing that gives them back.
Four questions. Answer them honestly and the order falls out.
Two or more answers pointing at adoption: start there. Clear answers to all four: implementation can start now.
A 60-person firm comes to us because staff are using ChatGPT on client work and the directors are nervous. That is an adoption problem. We set up the one-page policy, move the team to a business tier that does not train on their data, and run one pilot on the task that annoys people most. Weeks, not months.
Three months later the pilot has shown that reading supplier invoices and keying them into the accounting system is the single biggest time sink. That is an implementation problem. We build the document pipeline against real invoices, with a person approving anything unusual, and measure the hours against the figure we took in the pilot.
Same business, two different jobs, in the right order. The AI Adoption programme is the first job. AI implementation is the second.
This article is practical guidance from implementation work, not legal advice. The worked example is a composite of typical engagements, not a single client.
Sometimes. If your team already has a usage policy, approved tools on a business tier and one clear process to fix, implementation can start now. If staff are pasting client information into free tools and nobody has agreed which tools are allowed, building something on top of that adds risk rather than removing it.
It is the easiest part of it. Adoption also means a short usage policy, a decision on what information may and may not go into the tool, training with examples from your own work, and a way to see what the licences cost and whether they are used. Licences without those are the subscriptions nobody uses.
Adoption is weeks: a policy, tool set-up and a pilot. A first implementation is typically six to twelve weeks from scope to a system running in production, and a single well-defined process can be faster. Both are quoted in rands before anything starts.
Take the two-minute Quick Check, or tell us where you are and we will say which comes first, and what it costs in rands.
Book a call if you want one, or just send the form. We reply within one business day. POPIA-compliant.