Business Case for AI in ERP Processes: How to Quantify the Benefit

Why “30% faster” is not a business case – and the decision framework mid-sized and larger companies use to test whether an automation actually changes the bottom line

Short answer: when an AI automation produces a financial effect

This article explains the cause behind that pattern, sets out the decision framework Dreher Consulting uses to test automation projects before the technology decision, and states the limits of the approach.

The finding: the realisation rate has barely moved in three years

In August 2026 the Boston Consulting Group published “Look Past Productivity to Get Real Value from AI”. Its central finding: only 6% of the companies studied achieve a significant economic contribution from AI, measured in reduced costs and increased revenue.

The order of magnitude over time is more interesting than the single figure. Three BCG surveys across three years:

That is where the real observation lies. Investment in AI has multiplied over this period; the share of companies with a demonstrable value contribution has barely moved. A return share that stays constant while deployment rises steeply is not a maturity problem. Maturity problems resolve with time and practice. This pattern does not resolve, because the cause does not sit in the technology but in the decision architecture ahead of it.

The cause: three quantities that are routinely confused

In the same article BCG describes a company that used AI to cut a task from ten days to one. The customer still waited ten days, because the surrounding business process was unchanged. Only once committees, handovers and intermediate steps had been changed did the customer outcome improve.

The pattern is neither new nor specific to AI. It describes the difference between three quantities that business cases routinely treat as one:

Processing time

The time actually spent working on a case. Automation shortens this one reliably.

Lead time

The time from triggering a case to the result reaching the recipient, including waiting, queuing, approval and coordination time. Automation shortens this only when the waiting itself is removed.

Value contribution

The change in a result figure: EBIT, cash, working capital or throughput. This one changes only through a deliberate decision about what happens to the capacity gained.

In most processes, processing time is the smaller part of lead time. Halve the processing time and leave the approval chain untouched, and nothing changes for the customer. That is why a percentage such as “30% faster” says nothing on its own about whether the investment pays.

The second mechanism concerns staff capacity. Make a task 20% faster and you create a 20% time gain – spread across many people, in minutes per day. That gain becomes economically effective only if it is consolidated and given a new use. Otherwise it fills up with the work that has been left undone until now. The effect is genuinely felt by the people doing the job and invisible in the profit and loss account.

The three decisions about released capacity

Released capacity is an option value. It becomes a contribution to the result when one of the following three decisions is taken and assigned to a named person.

In practice the fourth row is the most common. It appears in no business case, because it is not a decision but the absence of one.

Approach: where the proof of value sits in selection and delivery

Dreher Consulting works in four phases. The proof of value is not an additional step but a checkpoint inside the first phase – before requirements are gathered and well before the system decision.

The decisive point is where it sits. An economic checkpoint placed immediately before the technology decision comes too late: by then the process has been described, the specification cut and expectations set inside the business. The decision about the value contribution is made in the process design, not in the system selection. That is the practical meaning of the principle organisation before IT.

Decision criteria: seven items that must exist before approval

An automation project – whether AI-based or part of an ERP implementation – should not be approved until the following seven items exist in writing. If one is missing, the business case is incomplete.

1. Today's economic loss. What does the process cost per year, in euros? With the measurement source stated, not an estimate from a workshop.
2. Target outcome from the recipient's point of view. What changes for the customer, the supplier or the downstream department? Not: what the system will be able to do.
3. Process steps to be eliminated. Which approvals, handovers, committees and intermediate steps actually disappear? Named individually.
4. Capacity released. How many hours per year, in which roles, at which sites?
5. Use of that capacity. One of the three decisions above, with the name of the person accountable.
6. Target figure and measurement point. EBIT, cash, working capital or throughput – with a target value, a baseline and measurement dates 6 and 12 months after go-live.
7. Abort criterion. At what interim result is the project stopped?

A checkpoint like this works only under three conditions: a no must be possible, accountability for the benefit must sit with finance rather than IT, and the follow-up measurement must be scheduled and commissioned before the project starts.

Evidence from projects

The three examples below come from Dreher Consulting engagements. Measured results and projected values are shown separately.

Formycon – biopharmaceuticals, product tracking

L’Osteria – franchise hospitality, finance processes

Anonymised – B2B distribution with heavy logistics

The third example is the most informative, because it makes the realisation gap visible rather than asserting it away. The productivity dip at go-live is the norm, not the exception. A business case that does not plan for it misses its targets on the time axis alone.

Alternatives: when a different route is better

Proof of value before the technology decision is not the right sequence in every situation. The realistic alternative is the technical pilot first.

In practice the combination makes sense: the pilot settles feasibility, the proof of value settles use. It only becomes a problem when the pilot replaces the investment decision.

When this approach does not fit

• Mandatory regulatory projects. Where there is no choice, the proof of value is documentation, not a basis for decision.
• End-of-life systems with no alternative. When maintenance and security support end, the question is not whether but how.
• Projects below the materiality threshold. If the proof costs more than the expected effect, it is bureaucracy.
• Pure security and compliance measures. The benefit is risk avoidance, not an improved result, and is assessed differently.

Dr Harald Dreher Managing Partner and owner of Dreher Consulting GmbH. Has advised mid-sized and larger companies in Europe – with a focus on Germany, Austria and Switzerland – on ERP strategy, ERP selection and digital transformation for more than 30 years. Originator of the SCOReX® and DAA™ methods. Dreher Consulting has worked vendor-neutrally since 1992 and has supported more than 1,200 projects in that time. Profile and further articles →  ·  Arrange a meeting with Dr Dreher →

Further reading

• Independent ERP selection for mid-sized and larger companies
• SCOReX® – structured requirements development through to a tender-ready specification
• Our consulting approach: organisation before IT
• Process optimisation before the system decision
• ERP consulting and requirements specification

Sources

1. Boston Consulting Group: Look Past Productivity to Get Real Value from AI. 31 August 2026. Accessed 1 September 2026.
2. Boston Consulting Group: AI Adoption in 2024: 74% of Companies Struggle to Achieve and Scale Value. 24 October 2024. Survey of 1,000 executives in 59 countries.
3. Boston Consulting Group: The Widening AI Value Gap. 30 September 2025.

The cost potentials BCG cites for AI-enabled transformations – 25–35% in administration and general costs, 20–30% in research and development, 15–35% in sales, marketing and parts of cost of goods sold – are BCG estimates, not general benchmarks. They are deliberately not used as targets in this article.