Start with the hours, not the technology
Before you speak to anyone, pick the one task that annoys your team most and count it. How many times a week does it happen, how long does it take each time, and who does it? Multiply the hours by what that person costs you per hour. That number is the most a fix is worth to you each month. If an offer costs more than the hours it saves, it is a bad deal, however clever it sounds.
Six questions to ask any partner
- Which of my processes would you change first, and why that one?. A good partner wants to understand how work moves through your firm before proposing anything. If the answer is a product demo, they are selling a tool, not solving your problem.
- How will we measure whether it worked?. Hours saved, calls avoided, documents chased: agree the measure and the starting number before the build, so the result is not a matter of opinion.
- Who approves what reaches my clients?. Anything sent to a client in your name should be approved by a person, or follow wording you approved. Ask to see where that approval happens.
- Where will my clients' data live, and who can see it?. You remain responsible under POPIA. The partner should be able to tell you, in writing, which systems hold the data, which country they are in, and whether any of it is used to train AI models.
- How do I switch it off?. Every automation should have one place where you can stop it, and the work should carry on by hand. If nobody can show you the off-switch, do not go live.
- What do I own when we stop working together?. Accounts, data, workflows and documentation should be in your name. Avoid anything that only runs inside the partner's own account.
Red flags
- Big percentage savings promised before anyone has looked at your process.
- A long contract or a large monthly fee before a single task has been proved.
- "Fully automated" client communication with no person approving it.
- Vague answers about where data is stored, or claims to be "POPIA certified" (there is no such certificate).
- Logos of AI companies presented as partners or clients without explanation.
- No named client you could phone for a reference.
What a sensible first project looks like
One task, done by hand with the new process first, then automated. A fixed price agreed up front. A measure agreed before the work starts, and a check a few weeks after go-live. If it does not do what was agreed, you stop there, having spent a small amount to find out. If it works, you have proof, and the next task is an easy decision.
Where AI helps, and where it does not
AI is useful for reading and summarising documents, drafting replies for a person to check, and answering routine questions from information you control. It is not a substitute for professional judgement. Attorneys should be especially careful: South African courts have referred practitioners to the Legal Practice Council over AI-invented case citations. Use AI to save time on admin, and keep decisions with qualified people.
Questions owners ask
- Should we buy software or hire a consultant?
- If an off-the-shelf tool already does the task well, buy it, and only pay someone to set it up properly. A consultant is worth it when the problem sits between several tools, or when the process itself needs redesigning before anything is automated.
- How long should a first project take?
- A single task should be live and measurable within weeks, not months. If the timeline for a first step runs past a quarter, the scope is too big.
- Is it safe to put client data into AI tools?
- It can be, with the right settings and agreements: data kept in known locations, not used for training, access limited and logged. Ask for this in writing before any client data is used.