Four Sources of ROI

Discussions about AI in hospitals often start with applications: discharge summaries, coding, knowledge search, procurement, HR, or IT support.

For hospital leadership, the key question is different:

What changes financially when AI is gradually used across more departments and workflows?

ROI usually comes from several sources:

• additional revenue

• lower costs

• more available staff capacity

• lower risk

When several applications run on the same platform, the fixed cost is spread across more users and processes.

There is no need to define every future application today. For an investment decision, the hospital needs to understand where the financial value comes from and how to calculate it.

1. Additional revenue

Much of the information relevant to coding and billing already exists in discharge summaries, operative reports, findings, materials records, or medication histories. It is not always fully taken into account.

AI can search for this information, bring it together, and flag discrepancies. For example:

• potential additional payments

• relevant secondary diagnoses and complications

• missing information for GOÄ or elective-service billing

• possible grounds for Medical Service appeals

• documentation gaps before a case is closed

AI does not change codes or generate unreviewed claims. Clinical coding specialists and medical controlling remain responsible for the final assessment.

For ROI purposes, only revenue that is additional to the current process and actually collected should be counted.

2. Lower costs

AI may affect several cost categories:

• external writing or translation services

• overtime

• repeated manual reviews

• duplicated work across departments

• multiple AI products with separate contracts and integrations

The starting values can often be found in invoices, contracts, overtime records, and internal workload data. Only the change that actually occurs after implementation belongs in the ROI calculation.

3. More available staff capacity

Many AI applications reduce the time spent on recurring tasks:

• searching for information

• summarizing documents

• preparing reports and correspondence

• reviewing files

• structuring workflows

• answering questions based on internal policies

The time saved does not automatically reduce payroll. It creates capacity.

In a hospital with vacancies, overtime, and backlogs, that capacity can reduce the need for external support, speed up processing, increase the number of completed tasks, or make it possible to address work that is currently left undone.

Assume that 500 employees use AI regularly and save an average of eight minutes per working day:

500 users × 8 minutes × 220 working days
= approximately 14,700 hours per year

Only the share that is actually used in the operation should be included in the calculation.

At 30 percent, that would be:

14,700 hours × 30 percent
= approximately 4,400 hours

At an assigned value of €50 per hour:

4,400 hours × €50
= €220,000 per year

Each hospital can replace these four assumptions with its own figures:

• How many people use the platform regularly?

• How much time do they save?

• What share of that time is actually used?

• What value does the hospital assign to that capacity?

AI can also make it possible to perform reviews and research that currently happen only on a sample basis, or not at all. More patient records can be reviewed, more contracts can be checked, and more documentation can be assessed before completion.

This can increase both the volume and the quality of work performed.

4. Lower risk

AI can flag missing documentation, missed deadlines, conflicting information, or incomplete workflows.

Possible effects include:

• fewer recoupments

• fewer missed deadlines

• less rework

• fewer breaches of internal policies

• fewer errors caused by incomplete information

Risk should only be assigned a euro value when historical cases, losses, or recoupments are available. Without that basis, the risk effect should be shown separately from the financial ROI.

How the ROI is calculated

The formula is:

ROI = (benefit – cost) ÷ cost

Benefits include:

additional net revenue

• avoided costs

• economically used capacity

• avoided risk

Costs include:

licenses

• infrastructure and operations

• implementation and integration

• internal project work

• administration

• professional review effort

One-time costs should be spread across the selected evaluation period. For many investment decisions, a three-year view is appropriate.

The same effect must not be counted twice. If reduced overtime is already included as a cost saving, the same hours cannot also be fully counted as additional capacity.

An example calculation

Assume an AI platform costs €250,000 per year, including operations, implementation, integration, and internal effort.

The annual benefit might look like this:

This results in:

€480,000 benefit – €250,000 cost
= €230,000 net annual benefit

The ROI is:

€230,000 ÷ €250,000
= 92 percent

These figures are an example, not an industry benchmark.

If the capacity effect is reduced from €220,000 to €110,000, total benefit falls to €370,000. ROI then falls to 48 percent.

This type of scenario analysis shows which assumptions drive the business case.

A closer look at revenue capture

Revenue capture can be tested using the hospital’s own data.

Assume that €150,000 in annual cost is allocated to this process and 15,000 cases are reviewed:

€150,000 ÷ 15,000 cases
= €10 per case

The question then becomes:

Does the AI-supported process generate more than €10 in additional net value per reviewed case compared with the current process?

Not every AI flag results in revenue.

The sequence is:

identified → reviewed → documented → billed → accepted → collected

If AI finds something that medical controlling would have found anyway, there is no additional revenue.

A retrospective test using completed cases shows which additional findings are clinically and technically valid. A limited live deployment can then show what is actually accepted, billed, and collected.

This gives the hospital figures based on its own cases and processes.

Why additional applications improve ROI

Once the platform is in place, further applications can use existing components:

• user management

• roles and permissions

• language models

• internal data sources

• security controls

• integrations

• operating processes

Every new use case still needs an accountable business owner. Depending on the application, additional integration or configuration may be required.

The technical and organizational foundation is already in place.

This can make smaller applications financially worthwhile. An assistant for internal policies, document summaries, or IT service support would rarely justify a separate AI procurement on its own. On an existing platform, most of the additional cost relates only to the specific workflow.

As more applications are added, fixed costs are spread across more users and tasks.

A shared platform can also prevent the hospital from introducing separate products for coding, discharge summaries, knowledge search, and contract review, each with its own contracts, user accounts, and security assessments.

How a hospital can calculate its own ROI

A first calculation requires these values:

The calculation is:

Total benefit = net revenue + avoided costs + capacity + avoided risk

ROI = (total benefit – total cost) ÷ total cost

For the investment decision, the hospital should be able to answer:

• Which revenue and cost effects can already be tested?

• How many employees will actually use the platform?

• How much of the time saved will be used in day-to-day operations?

• Does the calculation still hold under conservative assumptions?

• How does ROI change as further applications are added?

Hospital leadership does not need to know today which applications the organization will use three years from now.

It needs to determine whether additional revenue, lower costs, more capacity, and lower risk together exceed the cost of the platform.

How basebox supports this approach

basebox provides hospitals with an AI platform that can run in their own infrastructure or in a customer-isolated environment in a German data center.

Users, language models, internal data sources, and applications operate in one shared environment.

Applications with a direct revenue or cost impact can first be tested using the hospital’s own data. Additional applications can then be added to the same platform over time.