Not every workflow deserves a digital project. The strongest candidates are usually repetitive, data-driven processes with substantial manual effort, multiple handoffs, and measurable error costs. The best starting point is where better execution directly improves cycle time, quality, delivery reliability, or management control—not where a newly promoted tool happens to attract attention.
Why Should Digitalization Begin Before Software Selection?
Many digital transformation projects begin with a product demonstration. A vendor presents a workflow platform, AI assistant, or industry application, and the company then tries to find a business process that fits the product. This sequence often produces a technically functional solution with limited operational value.
Adding another digital channel does not improve a process by itself. Replacing a paper form with an online form has little impact when approvals still move through email, employees manually transfer data into the ERP system, and missing details are resolved by phone. The media break has merely moved to a different point in the workflow.
Current figures show why German SMEs need to make selective investment decisions. According to the KfW Digitalization Report for SMEs 2025, only 30 percent of companies had recently completed digitalization projects. A 2026 Bitkom survey also found that 51 percent of German companies with at least 20 employees were struggling to manage digital transformation. These findings do not argue against investment. They indicate that companies need a stronger process-based method for deciding where investment will produce operational value.
Assess where AI can create real value
The KrambergAI AI Readiness Assessment helps companies identify suitable AI use cases, evaluate process readiness and define realistic next steps for structured implementation.
Structured assessment · Practical prioritization · Made in Germany
A viable project therefore begins with observation of the business: Where are employees waiting for information? Where is the same data entered more than once? Where do questions, rework, scheduling delays, or approval loops consume capacity? Once the operational constraint is understood, the company can decide whether the appropriate response is a workflow, integration, mobile application, knowledge system, or AI-assisted process.
This distinction is particularly important for the German Mittelstand. Many companies operate with mature ERP systems, specialized industry software, long-standing customer relationships, and processes that have evolved over decades. The goal is rarely to replace everything. It is usually to remove friction from specific parts of the value chain while preserving systems and practices that continue to work.
How Can a Company Identify a Process Worth Digitizing?
A suitable process does not need to be completely standardized. It should, however, occur often enough to reveal a repeatable operating pattern. The most attractive candidates usually have a dominant standard path and a manageable number of exceptions requiring professional judgment.
Several criteria have proven useful in practice:
- The process occurs regularly and consumes meaningful employee capacity.
- Information must be searched for, copied, or entered repeatedly.
- Employees cannot easily determine the current processing status.
- Handoffs and follow-up questions extend cycle time.
- Missing information, errors, or delays create downstream cost.
- Roles, operating rules, and expected outputs can be described.
- Required data already exists or can be captured at the source.
- Business results can be measured before and after implementation.
The most valuable opportunities combine several of these characteristics. A small administrative step may be attractive when it occurs continuously across a large number of transactions. Conversely, a low-volume approval process may justify investment when one incorrect decision can lead to major rework, warranty exposure, downtime, or contractual risk.
Time savings should not be the only criterion. For manufacturers, field service providers, logistics companies, construction businesses, and skilled trades, the main benefit may be more complete documentation, lower error rates, faster invoicing, more dependable scheduling, improved traceability, or reduced dependence on a small number of experienced specialists.
Companies should also consider the point at which the benefit occurs. A solution may save time in one department while adding work elsewhere. For example, customer service may capture inquiries faster, but dispatch may still need to reconstruct the technical details. A useful digital project improves the end-to-end process rather than optimizing one activity in isolation.
Which Processes Commonly Produce the Greatest Value for SMEs?
The strongest opportunities are often found at the boundaries between departments, systems, and job roles. These are the points where information is most likely to be incomplete, delayed, or disconnected from the next activity.
Incoming customer inquiries are a common starting point. Requests may arrive through phone calls, email, web forms, messaging applications, or individual account managers. Information about the service location, equipment, technical issue, desired date, drawings, or photographs is often incomplete. A digital customer interface can collect the relevant details, request missing information, and route the case to sales, dispatch, or engineering.
Quote preparation is another valuable candidate. Scope descriptions, measurements, product data, labor rates, technical documents, previous projects, and pricing experience may be distributed across the ERP system, email inboxes, shared drives, and individual employees. A digital solution can assemble the available context, retrieve comparable projects, and prepare a reviewable draft without transferring commercial responsibility to the software.
Field service and installation companies frequently benefit from digital scheduling and mobile documentation. Photographs, readings, equipment identifiers, materials, working hours, customer confirmations, and follow-up requirements should be attached to the correct job while the technician is still on site. When employees reconstruct this information later from handwritten notes or personal messages, invoicing and follow-up planning are delayed.
Other common candidates include invoice review, purchase requests, goods receipt, inspection records, complaint handling, maintenance planning, employee onboarding, contract administration, quality notifications, internal support requests, and knowledge retrieval. The greatest value generally comes from connecting the information flow across these activities rather than digitizing a single document.
Industry context remains important. A recurring process in an equipment manufacturer may involve bills of materials, engineering changes, quality deviations, and production release. A technical service provider may focus on dispatch, service history, required certifications, and customer acceptance. A construction or trade business may prioritize measurements, work preparation, material availability, jobsite documentation, and progress billing.
Which Processes Should a Company Avoid Digitizing First?
Not every frustrating process is a suitable starting point. Some workflows are expensive mainly because responsibilities change frequently, services are not consistently defined, or decisions must be renegotiated for each case. Software can represent these organizational problems, but it cannot resolve them without management decisions.
Poor initial candidates include processes that occur rarely, depend on chronically missing information, or require a different treatment for nearly every case. A high number of manual steps is not sufficient evidence by itself. When those steps involve negotiation, professional assessment, risk evaluation, or customer relationship management, digital tools should normally prepare information and support employees rather than make the final decision.
A process without an accountable owner is also a weak candidate. Someone in the business must decide which inputs are mandatory, how exceptions are handled, who can approve results, and when rules should be changed. The IT team or an external provider can operate the technology, but it should not own decisions that belong to sales, operations, finance, quality, or engineering.
Companies should also avoid building an extensive custom solution for a workflow that will soon be replaced by a planned ERP upgrade, new industry platform, merger, or operating-model change. A limited interim solution may still be justified, but only when its temporary value exceeds the later migration and retirement effort.
Another warning sign is the absence of usable cases for testing. A workshop can produce an elegant process diagram that does not reflect daily work. Before committing to implementation, the company should examine real orders, service tickets, reports, approvals, and exception scenarios. These cases reveal whether the process is sufficiently repeatable and whether the required data can be obtained.
How Can Digitalization Potential Be Evaluated Systematically?
A practical benefit-and-feasibility matrix is usually sufficient for the first stage. Each process is evaluated according to its expected contribution to the business rather than the attractiveness of a particular technology.
The benefit dimension includes employee effort, waiting time, error cost, customer impact, delivery reliability, revenue contribution, compliance requirements, and dependence on specific specialists. The feasibility dimension includes data availability, process stability, interface options, exception volume, ownership, security requirements, and implementation effort.
A process with high potential and manageable implementation conditions is a strong pilot candidate. A process with high potential and difficult technical conditions may be strategically important, but it requires preparation. A process that is easy to implement but produces little operational benefit may be suitable for a technical experiment rather than a business priority. Low-value, high-complexity candidates should normally be rejected.
The unit of analysis should also remain narrow enough to manage. “Digitize sales” or “automate customer service” is too broad. Better candidates include “qualify incoming inquiries,” “assemble documents for quote preparation,” “check a field service report for missing information,” or “identify incomplete order data before dispatch.”
A useful assessment also distinguishes between local and end-to-end effects. A department may benefit from faster processing while another department receives lower-quality information. The evaluation should therefore follow the workflow from its triggering event to the output that creates value for the next employee, customer, or system.
How Do Common Process Types Compare?
| Process type | Typical starting point | Possible digital approach | Expected operational value | Common implementation risk |
|---|---|---|---|---|
| Incoming customer inquiries | Requests arrive through multiple channels and often lack essential details | Central intake, required fields, automated routing, and follow-up requests | Faster response, fewer calls, and better handoffs | The company adds another form without integrating existing channels |
| Quote preparation | Relevant information is spread across ERP, email, documents, and employee experience | Context assembly, retrieval of comparable projects, and reusable calculation or text components | Shorter response time, lower search effort, and more consistent proposals | Drafts are accepted without technical or commercial review |
| Scheduling and dispatch | Availability, skills, materials, location, and urgency are coordinated manually | Rule-supported scheduling, status tracking, and mobile updates | Less coordination, stronger capacity use, and improved delivery reliability | The design does not support urgent changes and operational exceptions |
| Service and inspection reports | Notes, photographs, and readings are transferred after the visit | Mobile data capture assigned directly to the job or asset | Faster invoicing, more complete records, and less rework | Employees must enter the same information in more than one system |
| Invoice and document review | Employees manually compare, assign, and approve documents | Document extraction, rule checks, matching, and approval workflow | Shorter processing time, auditable approvals, and fewer data-entry errors | Supplier, order, or master data is unreliable |
| Engineering or specialist decisions | Each case requires experience, context, and professional judgment | Knowledge assistance, case summaries, and decision preparation | Faster research, better documentation, and easier access to experience | A recommendation is treated as an authorized final decision |
The comparison illustrates an important distinction. Digitalization can capture information, coordinate work, apply rules, or provide professional knowledge. These capabilities should not be treated as interchangeable. A process can be digitally supported without being fully automated, and an AI component may be useful within a workflow without becoming the system that controls the workflow.
What Does an Economically Viable Pilot Look Like?
A strong pilot does not attempt to rebuild an entire department. It addresses a bounded workflow with real users, production data, and a measurable output.
A technical service company, for example, might begin with the qualification of incoming service requests. The objective would not be to redesign the complete customer service organization. The initial scope could centralize requests, capture equipment, site, urgency, and problem details, and ask customers or internal staff for missing information before dispatch. The existing service management or ticketing system would remain the system of record.
Before implementation, the company documents the current operating situation. How often does dispatch request additional information? How long does a case wait before it can be scheduled? How frequently is it assigned to the wrong technical group? Which documents are normally missing? Which employees act as informal information hubs?
The pilot then processes actual cases. Employees record successful outcomes as well as exceptions, workarounds, incorrect classifications, and additional effort. The company should not hide difficult cases merely to demonstrate a high success rate. Exceptions provide essential information about operating rules, required human approvals, and the limits of the chosen technology.
At the end of the pilot, management decides whether the workflow should be adjusted, expanded to additional request categories, integrated more deeply, or discontinued. A successful pilot produces an operational process rather than a demonstration. It includes ownership, access controls, interfaces, exception handling, support responsibilities, monitoring, and a method for updating the underlying rules.
The solution should also have a plausible route to ongoing operation. A prototype that depends on manual intervention from the project team is not yet a production process. Companies need to know who will maintain integrations, review AI behavior, update templates, respond to user issues, and approve changes.
What Commonly Goes Wrong in Digitalization Projects?
One frequent mistake is reproducing the existing process in software without questioning it. Every spreadsheet column, signature, internal copy, and approval stage is rebuilt even when some steps exist only for historical reasons. The result is a digital version of the same inefficiency.
Another problem is designing only for the standard case. During a workshop, the workflow may appear straightforward. In daily operations, however, employees encounter incomplete orders, special contract terms, missing master data, urgent scheduling changes, technical deviations, and customer-specific requirements. When the system cannot support these cases, employees return to email, phone calls, and personal spreadsheets.
Projects also fail when no baseline exists. Participants may report that the new workflow feels faster, but the organization cannot demonstrate whether handling time, waiting time, error frequency, throughput, or cost has improved. Without baseline data, management cannot distinguish real value from a positive user impression.
Lack of ownership creates another recurring problem. IT can maintain applications and interfaces, but it cannot decide which business information is mandatory, which exceptions are acceptable, or who has the authority to approve a commercial or technical result. These decisions need an accountable process owner.
Projects often become too broad as well. An initiative that attempts to replace CRM, ERP, document management, customer service, sales processes, and knowledge management at the same time will accumulate dependencies. A smaller workflow with meaningful financial relevance usually creates usable experience and reusable components faster.
A final problem is adopting AI before the surrounding process is designed. A language model may produce a convincing summary or document draft, but the organization still needs to determine where the output is stored, who reviews it, which sources are authorized, how errors are corrected, and what happens when the model cannot complete the task.
What Roles Do Data, Interfaces, and Ownership Play?
Digitalization can operate sustainably only when the company knows which system owns each category of information. Customer and order data may belong in CRM or ERP, technical documentation in a document management system, field activity in an industry application, and support cases in a ticketing platform. A new portal should connect these systems rather than create another competing source of data.
Master data is particularly important. Automated routing, review, scheduling, and reporting depend on usable records for customers, sites, products, assets, employees, qualifications, and responsibilities. Automation does not repair poor master data. It distributes the same defects more quickly into downstream workflows.
The 2026 Bitkom survey illustrates the issue: 61 percent of the surveyed German companies reported that they were making little or no use of the potential contained in their existing data. Process digitization therefore does not always require collecting more data. Many organizations first need to make existing information accessible, assign it to the correct business object, and provide it within the relevant workflow.
Interfaces determine whether this information can move without repeated manual entry. APIs are preferable where available, but they are not the only option. File transfers, event-based integration, database views, and controlled robotic process automation can serve as intermediate approaches. The selected method should reflect transaction volume, reliability requirements, security, and the expected lifetime of the integration.
Each workflow also requires a business owner. This person or role defines mandatory inputs, decision rights, exception paths, approval requirements, and change procedures. Without ownership, the digital solution will gradually diverge from actual operating practice.
When Is Digital Support Enough, and When Is Automation Appropriate?
Digitization, workflow management, rules-based automation, and AI represent different levels of intervention.
Basic digitization captures information electronically and makes it available. A mobile inspection record can replace paper without automating professional decisions.
A workflow then coordinates handoffs, responsibilities, deadlines, and approvals. The case moves to the appropriate role, and its status becomes available to authorized users.
Rules-based automation is effective when decisions can be reliably described using structured data. Examples include routing, completeness checks, deadline calculations, matching records, or selecting a predefined processing path.
AI becomes relevant when the workflow must interpret unstructured information. It can process emails, reports, images, free-form descriptions, transcripts, and large document collections. It can summarize context, retrieve similar cases, identify missing content, and prepare a recommendation or draft.
The control logic should still remain within the workflow and the assigned business roles. AI should not silently determine who receives a case, which source is authoritative, or whether a high-impact transaction is approved unless the organization has explicitly designed and governed that behavior.
Full automation is most appropriate when input quality is dependable, exceptions can be detected, and the consequences of an error are limited. Human approval is generally appropriate for prices, contracts, engineering commitments, employee data, safety-related decisions, or other outcomes with substantial financial or legal impact.
How Does AI Change the Selection of Suitable Processes?
AI expands the range of workflows that can receive digital support. Traditional workflow systems generally depend on structured inputs. Modern AI can interpret email, free-form text, images, reports, call notes, and extensive technical documents. This makes processes accessible that were previously difficult to automate because their inputs varied too widely.
Technical inquiry processing is one example. AI can extract relevant information from an email and its attachments, identify the product or asset involved, flag missing data, summarize the request, and retrieve comparable cases from company knowledge. It should not automatically commit the business to a price, technical solution, or delivery date.
Documentation offers another practical use case. Field service reports, quality notifications, jobsite photographs, and inspection records can be organized, summarized, and checked for missing content. A Company Brain can then provide approved work instructions, previous decisions, project experience, and regulatory requirements within the relevant operational context.
AI can also lower the effort required to document processes themselves. Interviews, work instructions, ticket histories, and example cases can be used to prepare an initial view of the actual workflow. Employees still need to validate the result because informal workarounds and professional judgment are rarely represented completely in source documents.
Technology adoption alone does not establish that workflows are well designed. Eurostat reports that 71 percent of EU SMEs reached at least a basic level of digital intensity in 2025. This means they used several defined digital technologies. It does not establish that their systems are integrated, their information flows end to end, or their most important processes have been economically optimized.
How Should the Business Impact Be Measured?
Before implementation, the company should define the operational change it expects. Appropriate measures depend on the workflow. Common examples include active handling time, waiting time, follow-up questions, rework, error frequency, on-time completion, throughput, cost per case, and time to invoice.
Sales processes may require response time, quote cycle time, conversion, backlog, and gross margin. Field service may focus on first-time-fix rate, documentation completeness, time on site, repeat visits, and billing delay. Manufacturing and logistics may prioritize scrap, inventory variance, downtime, throughput, and delivery performance.
The company should measure the end-to-end result rather than only the digital activity. Faster inquiry capture creates little value when the case waits longer in dispatch. Automatic document extraction is not beneficial when employees spend more time correcting the extracted data than they previously spent entering it.
Adoption is part of the economic assessment. A technically functional system produces no value when employees avoid it. Low usage, frequent manual corrections, duplicate spreadsheets, or continued reliance on email indicate that the process design, user experience, data, or implementation approach requires attention.
Quality should also be evaluated, especially for AI-assisted work. A faster draft is not an improvement when reviewers must perform extensive corrections or when important facts are omitted. The measurement approach should therefore include both efficiency and the reliability of the final business output.
How Can Individual Pilots Become a Scalable Digital Operating Model?
After a successful pilot, the company should avoid purchasing another isolated tool immediately. It should first identify which components can be reused. Examples include identity management, permissions, integrations, document processing, notifications, audit logs, master data, monitoring, and access to company knowledge.
Structured inquiry intake may later support quote preparation, scheduling, customer status updates, and service reporting. A mobile field service record can provide data for invoicing, maintenance planning, quality management, inventory consumption, and organizational knowledge. The value increases when information is reused across the value chain rather than processed only once.
For most SMEs, a layered architecture is more practical than a complete replacement program. Existing ERP, CRM, document management, and industry applications can remain in place while they continue to perform their core functions. New components can address gaps, connect information, and support selected activities.
Governance should scale with the portfolio. The organization needs shared principles for data ownership, access, human approvals, vendor evaluation, security, documentation, and operational support. Otherwise, successful pilots can create a fragmented collection of applications that becomes expensive to maintain.
KrambergAI GmbH helps German SMEs identify suitable workflows, define bounded pilot projects, and integrate digital and AI-supported solutions into existing operations. More information: https://krambergai.com/
Frequently Asked Questions
Which Processes Should an SME Digitize First?
The best candidates occur frequently, follow repeatable rules, and currently consume substantial searching, data entry, or coordination time. Common starting points include inquiry qualification, quote preparation, order handoffs, scheduling, field service reports, invoice review, and inspection records. Before work begins, the company should define the expected business benefit, the process owner, and the baseline.
Is Digitization Worthwhile for Low-Volume Processes?
Yes, when each case is expensive, time-sensitive, or operationally risky. A rare engineering approval may be more valuable to improve than thousands of routine emails if one mistake creates major downstream cost. With low volumes, the solution should usually assist specialists rather than pursue full automation, and it should reuse existing systems instead of creating an elaborate custom platform.
Must a Process Be Fully Standardized Before Digitization?
No. However, the normal path, major variants, decision rights, and handoffs must be understood. A digital project can improve the process, but it should not be expected to resolve every organizational issue at once. In practice, it is often enough to structure the standard case, expose exceptions, and require human approval for consequential decisions.
How Much Process Documentation Is Needed Before Starting?
A large process-modeling repository is unnecessary. For an initial pilot, document the trigger, inputs, main activities, handoffs, decisions, exceptions, systems, and output. Real examples and direct observation of day-to-day work matter most because official procedures often describe the intended process, while employees may rely on informal workarounds to complete actual cases.
When Is a Workflow Better Than an AI Solution?
A rules-based workflow is usually better when inputs are structured and decisions follow fixed conditions. AI becomes useful when the process must interpret emails, images, free-form text, or inconsistent documents, summarize context, or prepare recommendations. Many effective solutions combine both: the workflow controls routing and approvals, while AI handles selected knowledge-intensive tasks.
What Role Does the ERP System Play?
The ERP system often remains the system of record for orders, materials, labor, invoices, and master data. New applications should consume those records and return approved results instead of creating parallel spreadsheets. When interfaces are limited, a temporary bridge may be acceptable, but ownership of each data element must remain defined throughout the operating model.
How Can a Company Prevent New Media Breaks?
Trace the entire information path: where a data point originates, who adds to it, which system stores it, and who needs it next. Another portal or assistant adds little value when employees must retype or copy the same content. Strong solutions capture information at the source and make it available automatically at the next operational step.
How Long Should a Pilot Run?
A pilot should run long enough to encounter real variants and exceptions, but remain bounded enough that decisions are not postponed. Scope matters more than a rigid duration. The pilot is complete when the team has processed enough genuine cases to evaluate value, risk, adoption, integration effort, and whether the solution should be adjusted, scaled, or stopped.
Which Metrics Should Be Collected Before the Project?
Useful measures include handling time, waiting time, handoffs, follow-up questions, rework, errors, on-time completion, throughput, and cost per case. Depending on the workflow, quote conversion, first-time-fix rate, complaint frequency, or inventory variance may also matter. Each measure should connect to the intended business outcome; an oversized dashboard only creates more administrative work.
When Should a Digitalization Project Be Stopped?
Stop when the workflow is rarely used, the data remains unsuitable, exceptions dominate the normal path, or integration and operating costs exceed the likely benefit. Persistent low adoption after process changes and training is another warning. Ending a pilot is a sound management decision when it happens early, uses evidence from real cases, and preserves the lessons learned.
Sources for the Statistics
KfW Research – KfW Digitalization Report for SMEs 2025
https://www.kfw.de/%C3%9Cber-die-KfW/Newsroom/Aktuelles/News-Details_891136.html
Bitkom e. V. – Digitalization of the Economy: Almost Every Company Is Addressing AI
https://www.bitkom.org/Presse/Presseinformation/Digitalisierung-der-Wirtschaft-Unternehmen-beschaeftigen-sich-mit-KI
Eurostat – Digitalisation in Europe, 2026 Edition
https://ec.europa.eu/eurostat/web/interactive-publications/digitalisation-2026
Further Reading
Fraunhofer IAO – Business Process Optimization and Digitization
https://www.digital.iao.fraunhofer.de/de/leistungen/Prozessoptimierung/OptimierungundDigitalisierungvonGeschaeftsprozessen.html
OECD – Digitalisation of SMEs
https://www.oecd.org/en/topics/digitalisation-of-smes.html
Eurofound – SME Digitalisation in the EU: Trends, Policies and Impacts
https://www.eurofound.europa.eu/en/publications/all/sme-digitalisation-eu-trends-policies-and-impacts
All Articles AI Employees and Automation
