AI Implementation: Which Business Process Comes First?

In a meeting about AI, the list of ideas can grow quickly. Finance wants less data entry. Procurement wants faster document processing. Operations wants better demand forecasts.

With limited time and budget, which process should come first? TDI APJ recommends choosing a specific problem with usable data, a responsible owner and measurable results. The busiest process deserves attention, but it may need other changes before AI can help.

Should the most time-consuming task come first?

Repetitive work is an obvious candidate for AI. However, the reason it takes so long deserves a closer look.

For example, finance might compare several spreadsheets because departments use different codes for the same customer. AI could help identify matching records. However, if teams keep creating inconsistent codes, reconciliation work will return every month.

In that case, agreeing on customer records and code creation rules may deliver value sooner. Some subsequent steps may only need straightforward automation.

Before building AI, TDI APJ recommends separating these needs. Which steps require prediction, document interpretation or flexible information processing? Which need a clearer process? This assessment helps a company choose an approach it can maintain.

What makes a process a good candidate?

Does the problem have a clear business impact? First, identify where work creates delays, costs or customer problems. Saying that employees are busy is too broad to define a useful project.

Is the data ready for testing? Besides volume, check completeness, consistency and access rights. Data stored in a system may still need preparation before a team can use it.

How will people use the output? For example, who will use a forecast to change an order plan? If users must re-enter the information elsewhere, the extra work can reduce the benefit.

Who will evaluate the results? Someone with business knowledge needs to review outputs and handle exceptions. Without an accountable owner, even a promising use case can stall after a pilot.

Demand forecasting or document processing: which comes first?

Consider an illustrative distribution business. Its operations team wants better forecasts, but historical sales data does not clearly identify stockouts or promotions.

Low sales could reflect weak demand or a lack of available stock. Without that distinction, the team will struggle to evaluate a forecast properly.

Meanwhile, procurement manually enters documents with reasonably consistent layouts. The team has review samples, an accountable owner and records of current processing times.

In this situation, document processing may offer a more manageable starting point. The scope is easier to limit, and results are easier to compare. Forecasting may still be valuable, but it needs more data preparation.

Conversely, good operational data and costly inventory decisions could justify prioritizing demand forecasting. Therefore, the order of investment should reflect the company’s actual conditions.

How do you decide whether to expand a pilot?

Before testing, record how the work runs today. For document processing, measure time per document, correction rates and review time.

For forecasting, agree on accuracy measures and track how planners use the results. Then include the work needed to review outputs, handle errors and operate the solution.

A tool that speeds up data entry but adds more checking may offer little net benefit. Therefore, test common exceptions as well as easy cases. For example, include incomplete documents and unusual layouts that staff regularly encounter.

Keep the pilot narrow enough to manage. However, make it representative enough to support a decision. The results should help a company expand, adjust or stop the initiative.

Where can TDI APJ help?

Through TDI AI Studio, TDI APJ combines SAP knowledge with AI development capabilities. Assessment starts with the existing process: where data originates, who uses it and which decision needs improvement.

From there, a company can assess available SAP features, integration needs and any custom development. The appropriate scope depends on its systems, usage rights and business requirements.

The first project should produce enough evidence to guide the next investment. Sometimes that means narrowing an ambitious idea to one task that a team can improve and measure.

Discuss your proposed AI use case with TDI APJ, including the process, available data and result you want to improve.

About this article: This article draws on SAP guidance on use-case selection, data and measurement, alongside TDI APJ’s consulting perspective. The scenario is illustrative, not a customer case study. Source: SAP: Effective AI implementation in business (accessed November 2026).