Everyone is talking about AI. Fewer are talking about the data it runs on

Carla Roos, Strategic Account Manager at CloudSmiths.

---

In South Africa's automotive and logistics sectors, AI is the hot topic in every strategy session. Yet many of the same businesses cannot say with confidence how many real customers they have, where their stock should sit, or what last month's group numbers really were.

The conventional advice is to fix the data first and think about AI later. In practice, that often means a clean-up project that never ends, while the data drifts again the moment the project team leaves.

But data and AI are two sides of the same coin. AI needs good data to deliver value, and AI is now one of the best tools for getting data into shape. That points to a better order of operations: put AI to work on the groundwork itself. It can expose where data is broken, do most of the cleaning and give leaders a concrete view of what still needs fixing at the source.

Data in this sector sits in dealer systems, ERPs, finance tools, inboxes and countless spreadsheets. In logistics, much of the day-to-day runs on WhatsApp, drivers post delivery notes into groups and dispatchers confirm loads in chats. It works on the ground, but none of it is reportable until someone recaptures it by hand into a spreadsheet.

The result is disparate sources, manual recapture and processes that are slow and prone to human error. The fix has three parts: combine the sources, clean the data and automate the manual steps in between.

Book a test drive at a dealership and a record gets created. Sometimes it is a real customer. Sometimes it is a placeholder, typed in because the system demanded one. Multiply that across every retailer and every month, and the data becomes very hard to trust.

For a global vehicle manufacturer, we deployed an AI agent, built on CloudSmiths' ObjectiveFoundation, that assesses every retailer record as it arrives. Passing records go into a trusted golden record. Failing records go to a quarantine queue for a data engineer to review. That queue shows exactly which fields and retailers produce bad data, giving the business a concrete list of what to fix at the source. A natural-language layer then lets users query the clean data directly.

A South African automotive components group had subsidiaries that each reported in their own way, across planning tools, accounting systems, a data platform and many spreadsheets. Group reporting meant manual consolidation.

CloudSmiths brought those sources into a single data lakehouse, then added AI agents inside their team messaging workspace. These AI agents are able to explain financial variances, flag plant-level exceptions on stock and throughput, consolidate risk registers into one monthly view, produce ESG reports automatically, and flag payroll variances before each pay run. Each subsidiary sees only its own data.

For a motor retail and logistics group, CloudSmiths automated how one business handles requests for quote arriving by e-mail. Rule-based checks filter out internal mail, reply threads and system messages, so 76% of e-mails never reach AI. Valid requests get a lightweight AI check, and structured data is posted straight into the business's portal. In two months, it has processed about 8 200 e-mails, with 100% field-extraction accuracy on a test batch. Next is digitising paper-heavy dealership deal files, so missing documents are flagged automatically and deals are audit-ready.

A fuel distributor now has an agent that watches its WhatsApp groups and fills in its master sheet as each release note, bill of lading and proof of delivery arrives. And a delivery platform with 70 to 80 outsourced driver partners is automating fortnightly payouts that today take three to four people three to four days. Trip data is pulled from the admin platform and WhatsApp, de-duplicated, invoiced and paid per partner.

In each case, the data started off being scattered and messy. Cleaning it was part of the process, not a precondition for it.

In many automotive businesses, the CFO now has the final say on technology spend, and a traditional data warehouse project is a hard sell. Its return is real, but indirect.

Tying data work to a manual process changes that. The case can rest on measures finance already tracks: people-days saved, volume handled without extra headcount, reporting cycle time and errors avoided. Start with one process, prove the saving and use it to fund the next.

AI may be the hot topic, but data and AI are two sides of the same coin. The real business outcomes come from the data underneath, and AI is what helps get that data right. The good news is that this groundwork no longer has to be a separate, expensive project that comes first. AI can expose the gaps, do most of the cleaning and show leaders what to fix, while automating the work that was slowing them down.

So before signing off the next AI pilot, ask a simpler question: if the board asked tomorrow how many real customers you have, could you answer with confidence?