AI projects are usually sold as model projects. In practice, they are data projects with a model attached. McKinsey's 2026 survey finds that nearly nine in ten organisations use AI in at least one function, but only 37 percent report a positive EBIT contribution. The gap is not only adoption. It is the quality, ownership and accessibility of the data the model depends on.
An AI-ready data foundation is not a data lake. It is a set of agreements: this data means this, this person owns it, this is its quality rule, this is how it is accessed and this is how long it is kept. Without those agreements, every AI project rebuilds the same plumbing and discovers the same contradictions.
Where things stand in 2026
Most SMEs have data in an accounting system, a spreadsheet, a CRM, an ecommerce platform, a support inbox and a few personal drives. The same customer may have three IDs. The same product may have two names. The same revenue number may be calculated differently by sales, finance and the founder. AI makes those contradictions visible very quickly.
The first step is not a warehouse. It is a definition of the critical data elements: customer, product, order, payment, supplier, employee and student. For each one, define the source of truth, the owner, the key, the quality rule and the access policy.
2027: ownership, definitions and quality
2027 should be the year of the data contract. For each critical element, write a one-page contract: what it is, where it comes from, who owns it, what "complete" and "correct" mean, who can access it and how long it is retained. Publish it where the business can find it.
Fix the worst quality problems at the source. Duplicate customer records, missing product categories, inconsistent payment statuses and free-text addresses are common. A cleaning script hides the problem; a validation rule at the point of entry prevents it.
- Owner: a business person, not only IT.
- Key: a stable identifier for customer, product and order.
- Quality rule: completeness, validity, uniqueness and timeliness.
- Lineage: where the data came from and where it is used.
2028: semantic layer and real-time access
By 2028, the business should have a semantic layer: common definitions for revenue, active customer, order, refund, margin and churn. The dashboard, the AI assistant and the finance report should use the same definition. Without that, every team has a different truth.
Real-time access becomes more important for operations, but not everything needs to be real time. Classify data by freshness requirement: real time for order status and fraud, hourly for inventory, daily for finance, monthly for strategy. The architecture should match the decision, not the fashion.
Evaluation data is part of the foundation. If the business is building an AI workflow, it needs a curated set of real cases with expected outcomes. That set should be owned, versioned and protected. It is as important as the production data.
2029: data products and observability
2029 is the year data products become a practical idea. A data product is a reusable, documented dataset with an owner, a quality contract and a consumer. "Customer 360", "order truth" and "inventory availability" are data products. They reduce the cost of every new AI workflow because the plumbing is shared.
Data observability matures. Monitor freshness, volume, schema changes and quality failures. Alert the owner, not the whole company. A pipeline that silently stops updating is worse than a pipeline that fails loudly.
Access governance becomes more granular. Role-based access is not enough when an AI assistant can retrieve any document. Attribute-based and purpose-based access may be necessary for sensitive data. Log access and review it.
2030: AI-ready by default
By 2030, new systems should be designed to produce AI-ready data from day one: stable IDs, clear definitions, quality checks, lineage and access metadata. The business should not need a separate "AI data project" for every new use case.
Stewardship becomes a recognized role, even in a small company. The data steward is not a full-time job in every SME; it can be a responsibility assigned to the person closest to the process. What matters is that someone is accountable.
The mature position is trust: the model, the dashboard and the auditor use the same data, with the same definitions and the same access rules. That is what makes AI scale beyond a demo.
What to do in the next 90 days
- List the ten critical data elements the business depends on.
- Assign an owner and a source of truth to each one.
- Write one quality rule per element and measure it for 30 days.
- Fix the top duplicate or missing-data problem at the point of entry.
- Create a 50-case evaluation set for the first AI workflow.
How to measure progress
- Ownership coverage: the share of critical elements with an owner and a definition.
- Quality score: completeness, validity, uniqueness and timeliness.
- Freshness: the share of datasets updated within the promised window.
- Time to data: how long it takes to answer a new business question.
- Incidents: quality failures that reached a decision or a customer.
AI does not create a data problem. It exposes one. The businesses that fix the foundation will get more value from the same models.
Where this fits in our work
If you want help applying this to your business, start with Operations Engine. We publish scope and starting prices on the pricing page. For the next step in this series, read AI adoption for SMEs: 2027, 2028, 2029, 2030.
Sources
- McKinsey, "The State of AI: Global Survey 2026", 2026.
- DataGuidance, "Bangladesh: Personal Data Protection Act 2026", accessed 16 November 2026.
- NIST, AI Risk Management Framework, accessed 16 November 2026.
- NIST Privacy Framework, accessed 16 November 2026.
Responses (0)
No comments yet. Add the first one below.
Leave a response
Your email address will not be published. Required fields are marked *.