AI CLOCKHere is a number worth sitting with before you spend another naira, rand or cedi on an AI tool. In a new report published this week, Iron Mountain found that 86% of organisations had an AI project delayed or cancelled because of data concerns in the past 12 months. Nearly all of them, 95%, said using data to improve business outcomes was a top priority. The ambition is there. The records are not ready.
The report, The Information Supercycle: Aligning Data, Infrastructure and Capital for AI Scale, was released at the Gartner IT Symposium in Orlando on 19 October 2026, and covered in the days since by Express Computer, CXO Today and Compare the Cloud. All three report the same findings, so the figures below come from one survey, reported consistently across independent outlets. The report does not publish a sample size in the coverage, so treat the percentages as the survey respondents reporting their own experience, not as a census of all businesses.
That distinction matters, because the pattern is still hard to ignore. Four in ten organisations spend at least 41% of their AI project timelines on manual data preparation. Only 39% fall into the report’s “digitized and structured” category, where information is centrally managed, indexed and ready for analytics and AI. More than half cite information spread across disconnected systems or departmental silos (56%), and a similar share point to legacy formats or physical archives (53%). Seven in ten say they cannot effectively track the lifecycle of the data used in their AI models — where it came from, or where it went.
Why this hits African small businesses first
You do not need a corporate data warehouse to recognise this problem. You only need to look at how a typical African small business actually keeps its records. Prices live in a notebook. Orders live in WhatsApp chats. Customer names live in a phone contact list. Stock counts live in someone’s head. The delivery rider has one version of the price list, the shop has another, and the Instagram page has a third.
That is the same fragmentation the report describes, at a smaller scale. And it matters more now than it did two years ago, because AI is changing what it does with your records. A chatbot that gives a wrong opening time is embarrassing. An AI agent that quotes an old price, promises stock you do not have, or sends a refund you did not approve is expensive. Iron Mountain’s senior vice president for product and solutions marketing, Varun Gupta, put it plainly: “AI readiness is about much more than deploying the latest technology. Enterprises need to know that the information powering their AI is accessible, current and governed, with clear ownership and permissions.” As AI agents move from generating answers to taking actions across business workflows, he added, those foundations become essential to scaling AI with confidence.
The report also warns that fragmented information creates security exposure, compliance and audit risk, higher discovery costs and loss of institutional knowledge. In plain terms: when only one person knows where things are, the business loses that knowledge the day that person leaves, falls ill or changes phones.
The audit: five questions, honest answers
Run this audit before your next AI purchase. Answer each question with a yes or a no. A single no tells you where to start.
1. Can you find every price, product and customer record in one place, in under five minutes?
If your prices are in three places, you do not have a price list. You have three guesses. Pick one master location for each kind of record: one price list, one product list, one customer list. Everything else is a copy, and copies get updated last. This is the small-business version of the silo problem that 56% of the surveyed organisations reported.
2. Is your information current, or is it a museum?
Open your price list and check five items against what you charged this week. If any of them are wrong, your records are already costing you money, with or without AI. Retire old versions instead of keeping them “just in case”. An old price list that an AI tool can still read is not an archive. It is a trap.
3. Who owns each record, and who is allowed to see it?
Name one person responsible for keeping each list up to date. That person does not have to do all the typing, but someone has to notice when a list goes stale. Then decide what stays private. Customer phone numbers, supplier costs and staff pay details should never sit in the same unguarded document as your public price list. The report’s point about clear ownership and permissions applies to a five-person shop just as it applies to a corporation.
4. Can you trace where an answer came from?
Ask your assistant, or yourself, a simple question: where did that price come from? If the answer is “someone told me once”, you have the problem that 70% of surveyed organisations reported. You cannot effectively track the lifecycle of your data. Fix this by writing the source next to anything that changes often: the date a price changed, the supplier who set it, the person who approved a discount. One extra column saves arguments later.
5. Would you let AI act on these records without checking first?
This is the question that separates a safe AI pilot from a risky one. If you would not trust an AI tool to send a message, quote a price or update stock from your records unsupervised, then do not automate those actions yet. Start with AI jobs where a human checks the output: drafting replies, summarising customer messages, writing product descriptions. Move to unsupervised actions only after questions one to four are all yes. That order — draft first, act later — is the cheapest insurance in this whole story.
What to do this week
You do not need new software to pass this audit. You need one tidy afternoon.
- Merge your price lists. One master list, in one place, with a date on it. Delete or clearly label every older copy.
- Move one week of WhatsApp orders into a single sheet. Customer name, item, price, delivery status. That sheet becomes the record your future AI tool reads.
- Assign an owner to each list. Prices, stock, customers, suppliers. One name per list, written at the top.
- Remove customer details from shared documents. Keep a separate, private customer list and share only what each staff member needs.
- Pick one AI job, not five. Answering repeat customer questions is the safest first job, because you can check every answer before it goes out.
The bottom line
The Iron Mountain findings flip the usual AI sales pitch on its head. The bottleneck is rarely the model. It is the mess underneath it. Businesses that spend a week tidying their records will get more from a basic AI tool than businesses that buy an advanced one and feed it chaos.
Run the five questions. Fix the first no you find. Then buy the tool. In that order, your AI investment has a record to stand on.