How to Measure the ROI of AI in Your Company's Budget

How to Measure the ROI of AI in Your Company's Budget

You measure the ROI of AI the way you measure any other budget line: a baseline year to compare against, a named cost category, a deadline with a milestone, gross savings shown next to what you paid to achieve them, and a quality metric next to the cost metric. The clearest public example came on October 5, 2026, when the German telecom group Deutsche Telekom told investors in Bonn it expects AI and automation to cut its indirect costs by about €2.5 billion by 2030 versus 2023.

Many companies still measure AI by the number of pilots and subscriptions they have. Saudi Arabia has declared 2026 the Year of Artificial Intelligence, so how you measure what you spend on it has become this year's question. For measurement tools, see how to measure the real ROI of AI.

What Deutsche Telekom announced, in numbers

• 2027 target: gross savings of about €1.1 billion from AI and automation outside the U.S., compared with 2023. The target set at its 2024 Capital Markets Day was about €800 million, so the company raised it.
• 2030 target: about €2.5 billion lower indirect costs compared with 2023.
• Revenue: about €250 million of AI-related revenue from business with companies outside the U.S. expected in 2026, rising to about €800 million in 2030.
• Customer service: the Frag Magenta chatbot handled about 2.6 million customer service calls in the first half of 2026. In its U.S. operations, customer service calls fell 55%, and AI agents now handle 40% of customer contacts.
• Network: an agent called RAN Guardian cut the response time to mobile-network strain events from several hours to about one minute.
• Quality: complaints fell 30% when AI detects connection setup problems early.
• People: more than 100,000 employees have been trained on AI tools under the AI for All initiative.

These are figures a company publishes about itself at an investor event, and the 2027 and 2030 targets are forecasts, not results. But the way they are written is worth copying.

Four elements that make a target accountable

• A baseline year: every figure is compared with 2023. A saving means nothing if you do not know where you started, and a business that does not record its costs today cannot prove any saving tomorrow.
• A named cost category: the 2030 target is on indirect costs, not on total cost. The company did not break down what the category includes, but naming it alone stops AI savings from being mixed up with every other cost reduction.
• A deadline and a milestone: 2027, then 2030. The milestone is what let the company raise its target after measuring what it had achieved.
• Gross, not net: the 2027 target is described as gross savings, before what is spent on tools and development. In your business you need both numbers side by side: what you saved, and what you paid to save it.

There is a fifth element that is easy to miss: a quality metric next to the cost metric. A 30% drop in complaints says the saving did not come at the customer's expense. A saving that raises complaints is not a saving. It is a deferred cost.

Where the savings come from: repetitive work, not showcase projects

The examples the company showed are not exotic projects. They are everyday tasks repeated thousands of times:

• Repeat customer questions: an order status, a bill, a known fault. That is what Frag Magenta handles, and it is the same kind of question that reaches your customer service on WhatsApp and the phone.
• Catching a problem before the customer reports it: RAN Guardian does not answer customers. It watches the network and acts within a minute. In your business the equivalent is an alert on stock about to run out, an invoice past its due date, or an order running late.
• Everyday employee tools: the company says it uses ChatGPT Enterprise, Microsoft Copilot and its own internal assistant, AskT. Broad training is what turns a paid subscription into actual use.

How to write a target like this at your own scale

A 30-person business will not talk about billions, but the same shape works at any size. Start with one page:

• Pick three repetitive processes, no more: answering customer enquiries, entering supplier invoices, preparing the weekly sales report, for example. The process that repeats daily is where savings show up.
• Record the baseline this month, before any tool: how often the process runs, how long it takes, who does it, and how many errors or complaints it produces. Without that number, every later saving is an opinion rather than a figure.
• Give each process a cost metric and a quality metric: time per transaction with the error rate, or conversations closed without staff with the number of complaints.
• Set a checkpoint at 90 days and a target at one year: the near checkpoint tells you whether to adjust the target or change the tool.
• Calculate the net: subtract subscription costs, human review time and training time from the saving.
• Decide in advance where the saving goes: Deutsche Telekom said it will reinvest the additional 2027 savings in accelerating its digital transformation and expanding fibre in Germany. In your business it might mean moving staff hours from data entry to customer follow-up.

We covered how to measure return before and after a project in our article on measuring the real return on AI, and worked through the return on a customer service assistant step by step in our article on AI assistant ROI.

What does not apply to your business

• Scale: a company of this size spreads the cost of tools and teams across millions of customers. A small business needs the saving in one process to cover the tool's cost within months, not years.
• Data: an agent that acts within a minute needs data that reaches it as events happen. Early alerts in your business first need orders, stock and invoices in one system, not in scattered files.
• An announced number is not a result: €2.5 billion is a 2030 target. Do not measure a vendor's promises against it. Ask the vendor for the baseline and the metric they will be held to.

The Origami view

We read Deutsche Telekom's announcement as a shift from asking whether to use AI to asking how much it saves and where. That is the question every manager in the Kingdom should ask before signing a new subscription, especially in a year full of offers.

In the projects we build, we start in the same order: measure the process as it runs today, connect the tool to the system where the process actually happens, then measure cost and quality together after go-live. AI that cannot see your order and invoice data stays a writing assistant, not a tool that lowers costs.

Conclusion

• Deutsche Telekom is targeting about €1.1 billion in gross savings outside the U.S. in 2027 and about €2.5 billion in lower indirect costs in 2030, both against 2023.
• The value of the announcement is its shape: a baseline year, a cost category, a deadline with a milestone, and a quality metric next to cost.
• The savings shown came from repetitive work: customer questions, catching problems early, and everyday employee tools after broad training.
• Start with three processes, record the baseline this month, calculate the saving net, and decide in advance where it goes.

Sources

• Deutsche Telekom: Deutsche Telekom boosts growth, efficiency and quality through the use of AI (October 5, 2026): savings and revenue targets, customer service, network and training figures
• EuropaWire: Deutsche Telekom sets 2030 AI growth and efficiency targets: the tools used inside the company
• Investing.com: Deutsche Telekom raises cost-saving targets at AI Day event: the previous €800 million target and the reinvestment of savings
• Saudi Press Agency: Saudi Arabia declares 2026 the Year of Artificial Intelligence