Asher's Blog

Many companies are developing an AI strategy that is really a collection of tools. They subscribe to assistants, add a chatbot, experiment with automated reports and encourage employees to use models in their work. These activities may improve productivity, but they do not yet create a defensible AI capability.

The quality of intelligence a company can build depends on the quality of the information it can lawfully access, understand and trust. If customer records are duplicated, transaction descriptions are inconsistent, operational decisions are undocumented and important knowledge lives inside private conversations, the company does not have an AI problem. It has a data problem.

Your AI strategy is therefore your data strategy. The companies that organise, own and govern useful information today will have a meaningful advantage throughout the next decade.

Models are increasingly shared; context is not

Most businesses will not train a frontier model. They will use models available to thousands of other companies. Access to the underlying intelligence will become widely distributed, which means the model itself may offer limited differentiation.

What distinguishes the output is context: the company’s customers, transaction history, product knowledge, operating procedures, performance patterns and accumulated decisions. A general model can explain payroll. A company-specific system can understand how a particular employer structures compensation, which compliance deadlines apply, what happened during previous runs and which anomaly deserves attention.

That context must be captured before it can be used. Years of operating experience create no AI advantage if the experience remains scattered across employee memories, inboxes and unlabelled spreadsheets.

Clean data is an operating discipline

Data quality is often treated as a technical clean-up project. In reality, it reflects how the company works.

Duplicate customer records may reveal weak onboarding. Inconsistent transaction categories may reveal unclear financial processes. Missing outcomes may show that teams measure activity without recording what happened afterwards. An AI implementation cannot permanently repair these problems because new poor-quality data will continue entering the system.

The company must define important entities, fields, owners and standards. What is a customer? Which record is authoritative? How is a transaction identified across internal and partner systems? Who corrects an error, and how is the correction audited? Which decisions need reasons recorded?

This work can feel less exciting than deploying an intelligent agent, but it determines whether that agent will be reliable.

Owned does not mean collected without limits

Companies sometimes interpret data ownership as permission to gather everything. That is both dangerous and strategically weak.

Useful data has a clear relationship to the service being provided. Customers should understand why important information is collected, how it is protected and which decisions it may influence. The company needs legal rights and appropriate consent, but it also needs moral legitimacy.

Collecting unnecessary sensitive information increases security risk and makes governance more difficult. The strongest data strategy is not the largest database; it is the most relevant, trustworthy and well-understood body of information.

Ownership should also include portability and independence. If all of the company’s intelligence is trapped inside a third-party tool, changing providers may mean losing history or control. Businesses need contracts, architecture and export processes that preserve access to the information they have legitimately created.

African companies must build the missing context

Many global models have limited exposure to the full diversity of African languages, regulations, addresses, informal business practices and cultural context. A generic system may be fluent while still being wrong in ways that matter locally.

African companies possess the opportunity to create domain-specific data from real workflows. A payroll platform can learn how compensation and statutory obligations operate across jurisdictions. A logistics company can build knowledge around addresses and movement patterns. An agricultural platform can collect local crop, weather and market information.

This should not become an excuse for extracting data from communities without returning value. Local data is an asset partly because it represents people whose rights and interests must be respected. Governance is not separate from the opportunity; it is what makes the opportunity sustainable.

History becomes valuable when it is structured

The most useful AI systems will not know only what happened. They will understand what the company did in response and what outcome followed.

A support ticket that records a complaint but not its root cause and resolution offers limited learning. A sales record that shows a lost deal without the reason cannot improve qualification. An incident report that documents the failure but not the control introduced afterwards leaves future teams without the lesson.

Companies should begin capturing decision trails. What information was available? Which choice was made? Who approved it? What result followed? This creates the raw material for systems that can recommend rather than merely retrieve.

It also improves accountability. When an AI system proposes an action, the organisation can compare it with prior cases and understand whether the recommendation fits established policy or represents a meaningful departure.

Data access needs boundaries

AI makes information easier to combine, which increases both value and risk. An employee who previously had access to one report may now ask a system a question whose answer combines payroll, performance and personal data from several sources.

Permissions must therefore apply to the question and the resulting output, not merely to the underlying database. The system should know who is asking, what they are allowed to see and how sensitive information should be masked or aggregated.

Financial regulators increasingly emphasise data governance, model risk, operational resilience and third-party dependence because AI can amplify weaknesses that already exist. A company cannot outsource responsibility simply because the model or infrastructure belongs to a respected provider.

Start with important decisions

An effective data strategy begins by identifying the decisions the company wants to improve. Which customers are at risk of leaving? Where is money leaking? Which transaction requires review? What prevents onboarding from completing? Which employee question appears repeatedly?

Once the decision is clear, the company can ask what data is required, whether it exists, how reliable it is and who owns it. This prevents teams from building enormous data platforms without a defined use.

The first projects should create value while improving the underlying discipline. A reconciliation assistant, for example, may reveal inconsistent references and force the company to standardise them. The application produces an immediate benefit while strengthening the foundation for future systems.

The next decade will reward memory

Companies forget more than they realise. Employees leave, decisions disappear into meetings and customer insight is repeatedly rediscovered. AI offers the possibility of making organisational memory searchable and useful, but only if the organisation records experience in a form intelligence can interpret.

The strategic work begins now: clean the records, define ownership, document decisions, preserve customer permission, secure access and make outcomes measurable. The model can change as technology improves. A well-governed body of proprietary knowledge remains valuable across models.

Your AI strategy is not the presentation announcing which tools the company will adopt. It is the long-term decision about what the organisation will know, how confidently it will know it and what it is permitted to do with that knowledge.

Companies with clean, owned and trustworthy data will not merely use AI better. They will become organisations that learn faster than their competitors—and retain what they learn.