Home / AI implementation
We put AI to work on the process that eats the most hours: document processing, customer queries, reporting, repetitive account work. Built on your real data, with guardrails that keep client information inside POPIA, and measured in hours saved.
Book a call if you want one, or just send the form. We reply within one business day. POPIA-compliant.
AI implementation is the work of putting AI into a real business process so it runs in production: choosing the use case, connecting the data, building or configuring the tool, adding guardrails, and measuring the result in hours saved.
It is the step after AI adoption (policy, approved tools, training) and before scaling to the next use case. Buying ChatGPT seats is not implementation. Neither is a strategy document.
Saved by three AI agents our director built to handle repetitive account work at a NYSE-listed global SaaS company, supporting a portfolio worth $55M+ in client ARR.
Those agents ran against live account records, not samples. That is why the number is real and why the guardrails had to be designed before the first run.
We publish it because it shows how we scope and measure AI work, not to imply a client roster we do not yet have. Who did it, and where.
An AI implementation consultant picks a real process, connects your data, builds or configures the AI tool, adds guardrails for privacy and measures the hours saved. Unlike an adviser who delivers a report, an implementer stays until the system runs in production and your team can run it.
Working systems for the inside of a business. Each one replaces a person reading, re-typing or chasing something that software can now read, type and chase.
Invoices, contracts, forms and emails read by AI and written into your systems as structured data. The person who used to re-type them checks the exceptions instead.
Account admin, follow-ups, status updates, triage. The agent does the repetitive path; a human approves anything that leaves the business. This is what saved the 14 hours a week.
Email and WhatsApp enquiries classified, routed to the right person and drafted for reply. Your team sends the reply; it no longer writes it from scratch.
The monthly pack assembled from your systems and narrated automatically, with the numbers that moved flagged for a human to explain. Three days back to the finance team.
Answers from your own documents, policies and systems, respecting who is allowed to see what. The new hire stops asking the one person who knows.
The same fixed-scope, fortnightly-demo cycle we run on everything, with two steps AI makes non-negotiable: the POPIA line is drawn before the AI sees anything, and a person stays in the loop.
High volume, rule-based, costing real hours, and the data already exists. We put a rand figure on the time it takes today, so we can measure against it later.
What will the AI see? We strip or mask personal information that does not need to be there and choose a business-tier tool that does not train on your data. The traffic-light test we use.
Against live documents and the real edge cases, because that is where the exceptions hide. Working demos every fortnight, not a status report.
Anything that leaves the business is approved by a person. Logging, access control and a one-page usage policy are in place before go-live.
Hours saved against the number from step one. Only then do we move to the next use case. Spend stays visible in rands throughout.
Accounts, credentials, prompts and code in your name. Documentation for whoever maintains it next. 30 days of support after go-live.
All four are legitimate. They solve different problems, and the money goes wrong when a business buys one expecting another.
| Buy seats of a tool | AI strategy consulting | Agency build | AI implementation with us | |
|---|---|---|---|---|
| What you get | Licences; usage depends on each person | A roadmap and use-case list | A chatbot or automation, built to a brief | A process running in production, measured in hours saved |
| Who chooses the use case | Nobody, or everybody | The consultant, from interviews | You, in the brief | We do, with you, by sitting with the people doing the work |
| Data safety | Whatever each user pastes | A recommendation to have a policy | Depends on the brief | POPIA line drawn before the AI sees anything; human approves what leaves |
| When you see it working | Day one, for the people who try | After someone else builds it | At delivery | Every fortnight, on your real data |
| Who owns the result | The vendor owns the tool | You own the document | Depends on the contract | You: accounts, prompts, code and credentials in your name |
| Best when | Your team already knows what to do with it | You need board-level direction first | The brief is right and will not change | You want one real process fixed, then the next |
If staff are already using AI informally and there is no policy, no approved tools and no first use case, the build is not the first job. Our AI Adoption programme sets up the policy, the right tool tiers and a working first pilot in weeks. Implementation is what follows, and the pilot usually becomes the first production system.
Not sure which you need? The two-minute Quick Check will tell you, no email required.
Adoption vs implementation, explained →AI implementation is the work of putting AI into a real business process so it runs in production: choosing the use case, connecting the data, building or configuring the tool, adding guardrails, and measuring the result in hours saved. It is the step after AI adoption (policy, approved tools, training) and before scaling to the next use case.
AI adoption gets your team using AI well: a usage policy, approved tools on the right tier, training and spend control. AI implementation builds AI into a specific process so it does the work: an agent that handles account admin, a pipeline that reads invoices, a system that drafts replies. Most businesses need adoption first and implementation second, and we run both.
The one that is high volume, rule-based, costing real hours, and where the data already exists in a system. Document capture, email and enquiry triage, repetitive account admin and report assembly are the usual first candidates. If someone reads a document and re-types what is in it, that is the one.
It can be, if it is designed that way. We decide what information the AI sees before it sees any, mask personal information that is not needed, use business-tier tools that do not train on your data, log access, and keep a person approving anything that leaves the business. POPIA treats information inside an AI tool as processing, so the controls are built in, not retrofitted.
Two costs: the build, which we fix in rands before starting and scope small for a first engagement, and the running cost of the AI tools, which are priced in dollars per seat or per use. We publish a worked monthly AI budget for a 50-person firm so you can see the second number before you commit to the first.
Yes, if it is built for it. Most of the AI runs in the cloud; the risk is your side of the connection. We design the work to queue rather than fail, and to catch up when the connection returns, so an outage delays the process instead of losing what happened during it.
We will come back within one business day with whether AI is the right fix for it, how we would implement it, roughly what it costs in rands, and an honest answer if a simpler automation would do.
Book a call if you want one, or just send the form. We reply within one business day. POPIA-compliant.