On this page
Why Zimbabwean Businesses Cannot Answer Basic Questions About Their Own Data
On the first Monday of the month, the principal of a typical Harare estate agency wants three numbers before the management meeting: how many enquiries came in last month, how many became viewings and sales, and which portal or channel each one came from. The office manager spends the afternoon opening the portal dashboard, the CRM, asking three agents for their WhatsApp threads, a shared spreadsheet and the accounting system. By evening she has numbers, but they disagree, and some buyers appear in two places under different names.
That afternoon is the problem in miniature. Data fragmentation, not a shortage of AI tools, is the first obstacle to AI in Zimbabwean businesses: the data that matters exists, but it is spread across ten locations in forms that share no identifier. Answering the principal's questions means joining at least three of those locations, and in some businesses are owned by different people.
This is not a failure of diligence. Each tool was adopted to solve an immediate problem: WhatsApp solved communication, Excel solved tracking and the accounting package solved invoicing. Together they leave a business that knows a great deal and can prove very little.
Where Business Data Is Stored in Zimbabwe
Across the businesses we work with, customer and operational data ends up in the same ten places.
Communication Channels: WhatsApp, Email and Facebook Messenger
WhatsApp is where most business communication happens, so it is where most business data starts: enquiries, orders, payment confirmations, delivery photos and supplier price lists. Our analysis of WhatsApp as an operations channel documents why. The data stays in the thread, on the phones of the participants, in a form that cannot be queried or connected to anything else.
Email carries the formal layer: quotations, invoices, contracts and bank statements. Where staff use personal Gmail accounts for business mail, business records mix with personal correspondence on a platform the business does not control. Attachments arrive as PDFs, are saved somewhere and are rarely indexed.
Facebook Messenger and post comments add a third layer. In our experience a share of first-contact enquiries arrives there from customers who found the business through a post or community group. The conversation stays in the Facebook inbox unless someone copies it, and rarely does anyone.
Employee Phones and Shared Drives
The data on employee phones is the data a business is least aware it holds. A sales representative who has covered a territory for three years carries its relationship history in a personal device: WhatsApp threads, saved contacts, delivery-note photos and voice notes with special instructions. None of it is backed up to a business system, and when the representative leaves it leaves with them.
Shared drives create the appearance of organised data. Files accumulate under names chosen by whoever created them, in folders that made sense to that person, with no consistent taxonomy or version control. That becomes a document archive, not a data system.
Paper Records and PDFs
Paper persists for good reasons: delivery notes need customer signatures, goods received notes are produced where there is no connectivity, and some customers expect physical receipts. A paper record is a record of a real transaction, but it cannot be searched or aggregated without manual transcription.
A PDF is a photograph of structured data. An invoice exported from accounting software carries the same line items and totals as the original, but the structure is no longer machine-readable, and extracting it at scale needs optical character recognition. Most businesses that receive PDFs simply file them in shared drives at best.
Accounting Software and Point of Sale
Accounting software, whether Sage, Pastel, QuickBooks, Wave or Xero, is usually the most structured store a business has. Transactions are recorded consistently and the ledger gives a coherent record of money. The limit is scope: the client record holds billing history, not enquiry history, conversation notes or service interactions.
A point-of-sale system holds the richest transactional data a retailer generates, but it typically stays inside the POS as end-of-day reports. It rarely connects to accounting, purchasing or customer identification, so the hardware store customer who buys monthly for five years is a transaction with no name attached.
The Operational Spreadsheet
Where integrated systems are missing, the spreadsheet becomes the informal database of record: the customer list that is really a CRM, the stock tracker that is really inventory. As our analysis of spreadsheets as unofficial databases sets out, it holds everything inconsistently, with no enforced relationships between records, no queryable audit trail and no reliable way to tell which shared copy is current.
Estate Agency Data Fragmentation: Portal Enquiries, CRM Pipelines and Deal Records
Estate agencies show the problem most clearly because it is built into how the sector works. Agencies list on portals such as Propertyzone, Property.co.zw and Propertybook.co.zw, and each portal shows an enquiry table: who enquired, when, on which listing, and sometimes the message. A portal that instruments visitor sessions holds more than that, including which listings the buyer viewed, how often they returned and what price and suburb filters they set. That behaviour is the best available signal of intent, and our WhatsApp lead capture analysis shows how a platform can score enquiries with it. In most agency workflows only the enquiry itself arrives.
Agencies that want pipeline management adopt a CRM, and in our experience HubSpot and Zoho are common choices. They manage contact details, interaction history, deal stage and commission forecasts. The gap sits between the portal and the CRM: when an enquiry arrives, the agent creates a contact and types in a name and number, and everything the portal knew before the enquiry stays behind.
The journey then spreads further. The viewing is arranged on the agent's personal WhatsApp, price feedback arrives as a voice note, the negotiation happens verbally, and the commission appears in the accounting system against the client's legal name, which may not match how the agent logged them in the CRM. When the deal closes, no record says which portal the buyer came from, which listings they viewed, how many times they returned, what they offered and how many days the deal took.
Propertybook data reported by Techzim and summarised in our lead capture analysis shows about 13,000 monthly enquiries across more than 2,000 agents, roughly 2,400 of them unanswered, with an estimated 3 percent conversion rate. Because enquiries leave the portal for WhatsApp, the true end-to-end rate is unknown, and that is the data problem in one figure. Our real estate AI readiness analysis documents the same fragmentation at the listing level.
Why Data Fragmentation Blocks AI
Gartner predicts that through 2026 organisations will abandon 60 percent of AI projects unsupported by AI-ready data, and its survey of 248 data management leaders found that 63 percent either lack the right data management practices for AI or are unsure they have them. Its guidance is that if the data has issues, it is not ready for AI. That is a global finding, and Zimbabwean businesses meet it in a specific form.
Records Without a Common Identifier Cannot Train a Model
A useful prediction needs a training set: historical records linking inputs to outcomes. To predict which enquiries convert, a model needs past enquiries alongside the record of which became deals. In the estate agency, enquiries live in the portal and the CRM while outcomes live in the accounting system, and no field traces a commission back to a portal enquiry. Without a joined dataset there is no training set.
The same gap appears wherever the interaction and the outcome live apart. A retailer's customer conversations sit in WhatsApp and its sales in the POS. A wholesale distributor's orders arrive by WhatsApp and its invoices are raised in QuickBooks, with no programmatic bridge between them. The data a useful AI application needs exists in each of these businesses, but not in a form training can use.
The Attribution Gap: Which Channel Produced Which Sale?
Fragmented data destroys attribution, the ability to trace an outcome to its cause. A business that cannot say which channel produced which enquiry, and which enquiry became which sale, cannot measure the return on any marketing effort. Even well-tooled global teams struggle: in HubSpot's research, 33 percent of marketing leaders name proving ROI as their biggest challenge. For a business whose enquiries move to WhatsApp, attribution cannot be rebuilt afterwards, and the gap starts at the website button, as our post-click analysis shows.
AI optimisation of ad spend, lead scoring and channel allocation all take attribution data as input. A tool that promises to optimise what the business cannot measure will produce confident output built on a foundation that does not exist.
The Source-of-Truth Problem: One Customer, Four Records
The same customer usually exists in several systems at once. A property buyer is in the portal under the name they registered with, in the CRM as the agent typed it, on the agent's phone under a nickname and in accounting under their legal name. "J. Mutasa," "Joseph Mutasa," "Joe Borrowdale 3 Bed House" and "Mr J. Mutasa" are one person, but no system knows it, so an AI tool ranking the best current leads treats them as four leads with four partial histories.
No AI tool resolves this after the fact. It takes an architectural decision, made before deployment, about which system is the authoritative record for each type of entity and what happens when data arrives through a channel that is not the authoritative one. The decision pays off beyond AI: in Salesforce's survey of 4,050 sales professionals, 79 percent of high performers prioritised data hygiene against 54 percent of underperformers.
Data Protection Law Adds a Second Reason to Consolidate
Zimbabwe's Cyber and Data Protection Act treats a customer's name, address and telephone number as personal information, and treats recording or holding it as processing (section 3). The business that decides why and how the data is processed is the data controller. That controller must keep the data accurate and up to date, ensure that anyone working under its authority does the same (sections 7 and 13), protect it against loss, unauthorised alteration and access (section 18), and be able to demonstrate compliance to data subjects and to the Authority, which is POTRAZ (section 24).
Customer data held across personal phones, WhatsApp threads and unversioned spreadsheets makes each of those duties harder. Customers can ask for access to their personal information, for corrections and for false or misleading data to be deleted (section 14), and no request can be answered quickly if nobody can say where a customer's records live.
A security breach must be reported to POTRAZ within 24 hours (section 19), which a business cannot do well if it cannot list where its customer data sits. Section 28 adds a question for data held in foreign cloud services, because transferring personal information to a third party abroad requires adequate protection. Contravening sections 13, 24 or 28 is an offence carrying a fine, up to seven years' imprisonment or both (section 33), so consolidation serves compliance as well as AI.
How to Prepare Business Data for AI: Source of Truth, Identifiers and Integration
The work before AI is architectural. The question is not which AI tool to choose but which system is the source of truth for each type of data the business generates, and how data created elsewhere reaches it automatically. That takes three decisions:
- Identify where the data that matters most lives today, starting with customer enquiries and sales outcomes.
- Decide which location becomes the authoritative record for each data type, and give every customer, property or order a single identifier.
- Build the connections that route data from non-authoritative sources into the authoritative record without manual copying.
Data type | Candidate system of record | Where it arrives first | How it should reach the record |
|---|---|---|---|
Customers and enquiries | CRM or shared customer database | WhatsApp, portals, Facebook, email | Business API or form capture that creates the record at first contact |
Orders and invoices | Accounting or ERP system | WhatsApp messages, emailed PDFs | Order entry that posts to accounting instead of being retyped |
Sales transactions | Point of sale | The counter | Customer identifier captured at the sale and linked to the customer record |
Stock | Inventory system | Delivery notes, spreadsheets | Goods-received entry at the point of receipt |
Documents | Document store linked to the record | Email, WhatsApp, paper | Upload or automatic attachment to the customer or deal record |
A practical first step costs nothing. Take one deal closed last quarter and trace it backwards: where the enquiry first appeared, how many places the buyer was recorded, how many manual copy steps connected them and whether anyone can name the channel. The count of locations and copy steps is the baseline.
Manual copying is where quality degrades and fragmentation perpetuates itself. An agent who retypes a portal lead into a CRM introduces spelling errors, omitted fields and delays, and cannot transfer behavioural data at all, because the browsing history never appears in the enquiry table the agent reads. Not every business needs custom software to fix this: the spreadsheet analysis lays out a ladder from process discipline to cloud spreadsheets, database-backed tools and custom builds, and the right rung is the lightest one that removes the specific risk. In our experience, the businesses closest to AI readiness are those that chose a system of record for customer data and built the discipline, then the automation, of capturing every interaction there.
Data Architecture Before AI: Where the Work Starts
Data fragmentation is the sum of tools adopted one problem at a time, and it is fixed the same way it formed, one decision at a time, but in a deliberate order. Sparkline's solution architecture work maps where data is created, who owns it and which system should be authoritative. Integration and automation then connects the channels so a record forms at first contact, and systems engineering builds the system of record where none exists.
An AI application is a late step in that sequence, not the first. A business that treats AI readiness as a purchase will find the data is not there. One that treats it as a set of decisions about records will find the AI work that follows is straightforward.
Sources
- Gartner. (2025, February 26). Lack of AI-Ready Data Puts AI Projects at Risk. Gartner press release and analyst Q&A with Roxane Edjlali.
- Salesforce. (2026). State of Sales Report 2026: Key Takeaways. Salesforce. Survey of 4,050 sales professionals.
- HubSpot. (2026). The Top Challenges Marketing Leaders Expect to Face in 2026. HubSpot.
- Parliament of Zimbabwe. (2021; commenced 11 March 2022). Cyber and Data Protection Act, 2021 [Chapter 12:07]. Act No. 5 of 2021. ZimLII, Laws.Africa Legislation Commons.
- Sparkline Labs. (2026). Why Zimbabwean Businesses Use WhatsApp to Run Critical Business Operations.
- Sparkline Labs. (2026). Excel for Business in Zimbabwe: When a Spreadsheet Becomes an Unofficial Database.
- Sparkline Labs. (2026). Can WhatsApp Leads Be Captured and Scored Automatically in Zimbabwe? Yes. Here Is What That Looks Like. Cites Techzim (2026, March) on Propertybook platform data.
- Sparkline Labs. (2026). AI Readiness in Zimbabwe Real Estate 2026: Why Your Listing Data Is Blocking Every Application That Matters.
- Sparkline Labs. (2026). The Zimbabwean Business Website: What Should Happen After Someone Clicks "WhatsApp"?
- Sparkline Labs. (2026). Solution Architecture.
- Sparkline Labs. (2026). Integration and Automation.
- Sparkline Labs. (2026). Systems Engineering.
About the Author
Building Something Like This?
We build platforms, internal tools, and integrations for Zimbabwean and African businesses. Bring us the problem, the idea, or the system that is no longer working as it should.