Artificial intelligence has long since arrived in most project-based service companies, but it is being applied in the wrong places. Sales uses AI for proposal drafts, marketing for content, and customer service for initial responses. In controlling and the back office—where project costs, utilization, and margins are actually determined—the reality is quite different.
The reason is rarely a lack of will from management. It lies in the state of the data. Project hours are in one tool, costs in financial accounting, resource planning in Excel sheets, and proposals in yet another system. Anyone wanting to grow must manually consolidate these silos, week after week, project after project.
This fragmentation is not a technical detail; it is a structural problem for the industry. Project service providers sell time and expertise, so their most important management tool is the connection between hours, costs, and utilization. This exact connection is missing in many companies because systems have grown historically and were never built to work together.
The most important points in brief:
- Project hours, costs, and proposals are stored in separate systems, making reconciliation take days instead of minutes
- Forecasts are based on outdated or incomplete data, meaning corrective action comes too late
- Travel expense reporting and proposal calculations tie up specialists who should actually be performing billable project work instead
- Scaling from 20 to 80 employees multiplies the coordination effort in the back office without additional staff being allocated for it
- Mid-sized service providers usually cannot afford a dedicated position for project controlling automation
Each of these gaps has a direct impact on margins. If you don't have a real-time view of utilization, project costs, and forecasts, you are driving by looking in the rearview mirror instead of looking ahead. This is precisely where project controllingdetermines whether AI makes a real economic difference or remains just a pilot project disconnected from daily operations.
It is telling that the very companies that would benefit most from better management have the least time to build it. High-growth project service providers with 50, 100, or 150 employees are caught up in day-to-day business, while the coordination effort in the back office continues to rise with every new project.
It is also striking how unevenly AI budgets are currently distributed. In many companies, marketing and sales receive the first AI tools because the benefits are immediately visible, such as faster proposal drafts or campaigns. Project controlling and the back office are often left out, even though the economic leverage there is at least as significant.
How German companies are using AI in 2025 and 2026
The figures on AI usage in Germany show a clear gap between ambition and back-office reality. According to the Bitkom AI Study 2026, the share of companies actively using AI has risen from 17 to 41 percent—a significant jump in just a few years. However, this progress is distributed extremely unevenly across business departments.
According to the same study, only 17 percent of companies are actively using AI in controlling and accounting. Consequently, the very area where project service providers manage their margins, utilization, and forecasts is lagging significantly behind marketing and sales.
The study cites legal uncertainty (53 percent), a lack of expertise (53 percent), and a lack of personnel resources (51 percent) as the biggest obstacles. For mid-sized project service providers with 5 to 250 employees, this hits the heart of the problem: there is simply no capacity to build and operate AI projects in controlling in-house while ensuring legal compliance.
These three obstacles are closely linked. Without in-house expertise, legal uncertainty is difficult to resolve independently, and without available personnel capacity, even an existing concept remains nothing more than a statement of intent. For management and CFOs, this means that building a proprietary AI solution for controlling is not a realistic option for the vast majority of project-based service providers, regardless of company size.
For day-to-day project business, this has two implications. First, the lag in controlling is not a peripheral issue; it directly affects the areas where project-based service providers lose the most money when management intervention comes too late.
Second, by opting for ready-made AI features integrated into existing software rather than building an in-house AI team, companies automatically bypass the study's biggest hurdles. Legal compliance, expertise, and staffing then become the responsibility of the software provider, not the organization itself. This shifts the barrier to entry from internal workforce planning to choosing the right software.
The concrete impact on management quality is most evident where data is still compiled manually, such as in project business forecasting. If you are reliant on outdated figures here, no AI application in the world can compensate for that in the short term.
The jump from 17 to 41 percent in active AI usage also shows how quickly the landscape is shifting. Companies that are still waiting in 2026 will find it nearly impossible to catch up in a single step. Conversely, those who establish a clean data foundation in project controlling early on will be ready as soon as AI features become available in their software, without having to retroactively adjust their structures.
From data chaos to AI-supported management
In practice, AI in project controlling rarely fails because of the technology. It fails because of the data foundation. A language model cannot calculate a reliable margin if project hours are in one tool, costs in another, and quotes in a third Excel file.
The foundation: consistent data instead of Excel silos
Before a single AI feature can make an economic difference, a shared, structured data foundation is required: project hours, costs, and utilization in one place, in a consistent format, and recorded uniformly for all projects. This is the prerequisite before back-office automation can even take effect.
This foundation cannot be created retroactively by an AI tool. It must be in place beforehand as part of the project management and time-tracking structure. Only when every project hour, every cost item, and every utilization figure is in the same system and the same format can an AI model provide meaningful insights.
We describe what such a shared foundation looks like in practice in our detailed article on Project management software for service providers. Without this foundation, any AI project in controlling remains an isolated experiment.
Level 1: Intelligent suggestions within the product
The first AI level at ZEP builds on this data foundation. In the future, users will query project data in natural language instead of manually building filters and reports. Inputs can be automatically populated from documents, and intelligent suggestions assist with recurring entries.
For management, this means that answers to questions about utilization or project status are available in seconds, rather than waiting for the next reporting cycle at the end of the month.
The effect of this first level is intentionally unspectacular. It does not replace any role in the company; it simply removes friction from daily operations. That is precisely the difference between this and many AI announcements that make big promises without ever reaching day-to-day business.
From outlook to everyday reality: the ZEP Assistant
What is described as a development in this article has been live in ZEP since September 2026. The ZEP Assistant, the integrated chat assistant in ZEP, answers controlling questions in natural language and executes actions directly.
A department head starts the week with a single request: "Give me the weekly update for my team: Who is absent, which vacation requests are waiting for approval, who didn't log their hours last week, and what is the planned utilization?" The assistant compiles the answer from ZEP data. Previously, this would have required four separate reports from four different areas of the software.
Two features distinguish this from generic AI chats: The assistant understands the user's context and permissions, and the data remains within ZEP. No public model training, complete audit trail. Included from ZEP Compact in the standard version.
Level 2: AI agents for back-office automation
The second level goes a step further: autonomous digital assistants that proactively handle recurring tasks in the background. Project controlling, travel expense reporting, quote calculation, and compliance monitoring are typical candidates for this. These are precisely the tasks for which mid-sized project service providers don't hire a dedicated full-time employee, yet they still consume time every week.
AI agents take on these tasks additionally, in the background, without requiring anyone on the team to be reassigned. This is the core of back-office automation as ZEP defines it: providing relief where no one was previously responsible.
For management, this is the real lever when it comes to AI in project controlling. It is rarely about a single function; it is about the sum of tasks that will run in the background without manual intervention. Over the course of a year, this adds up to significant capacity that is currently tied up in recurring administrative work.
Level 3: Standalone AI products on the horizon
In the long term, ZEP is also developing standalone AI products under its own brand. This third level is built on the same principles as the first two: a clean data foundation, DACH-compliant processing, and a modular structure that grows with the company.
For management and the CFO, this sequence is more important than any individual feature. Those who start with level one today are structurally prepared for when level two and later level three become productively available. There is no need for a restart on a different data foundation.
In practice: IT consulting, management consulting, engineering
As different as IT consulting, management consulting, and engineering may be in their day-to-day operations, the challenges in the back office are very similar. All three industries sell project time, all three struggle with distributed systems, and all three usually lack a dedicated position for controlling automation.
An IT consulting firm with 60 employees knows the problem of changing project teams and parallel client projects. With consistent time and cost data, AI in project controlling provides daily updated utilization rates per consultant and project, instead of a manual report at the end of the month. If utilization in a team drops, it becomes visible before the month-end closing shows it.
Management consulting firms often calculate quotes under time pressure and based on experience. An AI agent for quote calculation can draw on historical project data and suggest a reliable cost framework before the quote is even sent. This reduces the risk of under-calculated projects, a classic issue in the consulting business.
Engineering firms often work with long project durations and changing compliance requirements. Automated compliance monitoring in the background reduces the risk of missing deadlines or documentation obligations without needing to create an additional position. Especially for publicly funded projects, such background checks save valuable time during project completion.
The same mechanism applies in all three industries. The AI agents themselves hardly differ; what matters is which project data is available and in what quality. That is why any serious engagement with AI in project controlling begins with the data structure, not a provider's feature catalog.
The article on the performance management system for project service providers shows how these effects interact to impact overall corporate management.. In ZEP Professional these three levels converge on a unified project database, rather than existing as separate, isolated solutions.
{{blog-cta}}
Why now is the right time for project-based service providers
Three current developments are converging to make AI in project controlling a priority for executive management, not just the IT department. The need for back-office automation is growing because open positions in controlling are becoming increasingly difficult to fill. At the same time, there is pressure for tool consolidation: separate systems for time tracking, project management, and finance hinder any AI application before it even begins.
Added to this is the challenge of scaling. Growing from 30 to 100 employees fundamentally changes management requirements, often faster than new controlling capacity can be built.
All three triggers point to the same prerequisite: a common data foundation for project time, costs, and utilization. Without it, any AI investment in controlling remains an isolated solution that creates additional interfaces and maintenance overhead instead of providing relief.
For CEOs and CFOs, it is therefore worth taking a sober look at their own system landscape before discussing specific AI use cases. How many separate tools currently track project time, costs, and utilization? How often is data manually transferred between them? These questions determine how quickly AI in project controlling will actually pay off.
Every additional tool in this chain costs more than just license fees. It costs time for maintenance, interfaces, and reconciliation, and it complicates any future AI application because the data foundation becomes more fragmented rather than more unified. Tool consolidation is therefore not just a cost issue; it is a prerequisite for effective project controlling with AI.
ZEP provides exactly this foundation as a modular platform that grows with your company. Whether you start with simple time tracking using ZEP Clock or jump straight into full project controlling via ZEP Professional: The data structure remains the same when additional AI agents are added later.
Conclusion: The next step
AI in project controlling in 2026 will not be defined by a single feature. It will be defined by the data foundation it is built upon. Those who consolidate project time, costs, and utilization today will be in a position in twelve months to productively deploy AI agents for travel expense reporting, quote calculation, and compliance, while the competition is still reconciling Excel spreadsheets.
The first step is rarely technical. It is organizational: consolidate data in one place, then build automation on top of it. In practical terms, for the coming weeks, this means:
- Audit your system landscape: How many tools currently track project time, costs, and utilization?
- Consolidate your data foundation before selecting individual AI features.
- Test the first AI layer, perhaps via a free trial, instead of waiting for a major project.
FAQs
Do I need my own data science team for AI in project controlling?
No. According to the Bitkom study, this is one of the greatest obstacles: 53 per cent of companies fail due to missing know-how, 51 per cent due to missing personnel capacity. Anyone who relies on integrated AI functions in existing software bypasses this problem, because legal compliance and know-how rest with the software provider.
From what company size does AI in project controlling pay off?
The lever typically becomes tangible from around 20 to 30 employees, when the coordination effort in the back office grows disproportionately to team size. ZEP accordingly targets project service providers with 5 to 250 employees, regardless of growth phase.
What distinguishes an AI agent from a traditional automation rule?
A classic rule works rigidly on an if-then basis. An AI agent evaluates project data, recognises patterns and proactively suggests actions — for example in proposal costing or travel expense processing — instead of merely executing predefined steps.
How are AI in project management and AI in project controlling connected?
Project management delivers the operational raw data such as hours, task status and resource planning. Controlling evaluates this data commercially, for example for margin and forecast. AI can only meaningfully connect both levels when the underlying data basis is consistent.
Which back-office tasks can be automated first?
Travel expense processing and proposal costing are usually the first candidates, because they are strongly rule-based and simultaneously data-intensive. Project controlling and compliance monitoring frequently follow as a second step, once the data basis is running stably.
Is AI in controlling legally deployable in Germany yet?
Legal uncertainty is cited by the Bitkom study at 53 per cent as the greatest obstacle to AI deployment overall. Anyone who relies on DACH-compliant AI functions integrated into established specialist software instead of self-developed solutions reduces this risk significantly compared with building your own.
