Turning Installed Base Complexity into Revenue: How an AI Service Configurator Rewires Aftermarket Sales

At the same time, expectations are rising. Customers want fast, relevant, and proactive service recommendations. Manufacturers want to grow service revenue, smooth demand, and strengthen lifecycle relationships without dramatically increasing headcount or relying on tribal knowledge.

At Spare Parts Business Platform 2026 – Power of 50, Dr. Heiko Dirlenbach and David Ciccioli from Flender explored an actionable example of how to break that deadlock by developing an AI-supported service recommendation configurator integrated into CRM and e-business systems. The initiative shows how Flender, a company with a global installed base of more than 600,000 gear units, is using AI to productize, scale, and connect service expertise directly to order intake.

From installed base blindness to customer self-identification

A recurring challenge for many manufacturers is that they do not know who actually operates a significant portion of their installed base. When sales go primarily through OEMs and intermediaries, end-user data often never reaches the manufacturer. That limits the ability to proactively manage service, plan inventory, and build long-term relationships.

Flender has more than 600,000 gear units in operation worldwide, across a wide variety of industries and applications. A large share of those units were sold via OEMs that do not routinely share end-user information.

Instead of treating this as a pure data-quality project, the organisation reframed it as a service and commercial opportunity to create a self-service interface where end users voluntarily identify themselves and their assets in exchange for tailored service recommendations.

The logic is straightforward:

• The customer enters key data from the nameplate: serial number, product type, size, and year of first shipment.
• The system uses that information to identify the unit, determine its lifecycle phase, and generate a structured recommendation document.
• At the same time, the CRM system captures or updates the customer and asset data, creating leads, tasks, and context for follow-up.

By enabling customers to announce themselves and their equipment in a guided, value-adding flow, the company simultaneously addresses data gaps and creates new commercial touchpoints.

Codifying lifecycle expertise into configurable packages

Behind the AI front end, there’s codifying decades of service experience into a structured lifecycle and package logic.

The installed base spans more than 800 different gear unit types, each operating in diverse industries and applications. That variety makes it difficult to standardise service offerings. Different units and applications require different spare parts, inspection routines, and replacement intervals, and those needs shift across the lifecycle.

To address this, lifecycle curves were mapped for representative gear units, described visually as snakes that illustrate how service needs evolve. These curves were then divided into discrete lifecycle phases. For each phase and unit/application combination, the team defined corresponding service packages, for example:

• Early phase: Seals and buffers for couplings, basic preventive elements.
• Midlife service sweet spot: More comprehensive packages including bearings, geared parts, and on-site repair options.
• Late phase: Recommendations that may include exchange of the entire gear unit instead of repair for smaller or very old units.

All of this was originally encapsulated in a highly sophisticated Excel tool, powered by macros and owned by a few experts. The spreadsheet could generate recommendation documents, but only locally, for a limited number of units at a time, and only when the expert was available.

The transformation lay in taking that logic out of the spreadsheet and embedding it into a digital configurator that could be used self-service by customers and internally by sales, without losing the underlying expertise.

AI as interface, not as oracle

A key insight from the initiative is that AI is used primarily as an interface and orchestration layer, not as the decision engine for technical recommendations.

The core recommendation logic is rule-based and derived from engineering and field experience. For a given unit type, age, and lifecycle profile, the configurator will always propose the same package.

Generative AI comes into play around that logic, for example:

• Interpreting user input and guiding the dialogue in natural language.
• Summarising the interaction for internal records.
• Generating customer-facing text around the recommendations, without changing the underlying technical content.

This division of roles is crucial in a safety and availability-critical context. It enables a natural, human-like interaction while preserving control over what is actually being recommended. Guardrails are set so that the AI cannot suggest operational changes or troubleshooting advice that might endanger equipment or safety. Instead, the AI remains strictly within the domain of explaining and packaging pre-defined, validated recommendations.

The importance of trust: branding, transparency, and guardrails

One recurring concern in deploying AI customer-facing tools is trust and acceptance. The approach taken offers several lessons for industrial companies:

1. Make the AI visible and human, not hidden

Instead of burying AI behind a generic interface, the tool was given a persona and a visual identity, Alfred. This makes it easier for users to understand that they are interacting with an automated agent, not a human, and sets expectations appropriately.

2. State clearly what the agent does and does not do

The opening message to users explicitly states that they are engaging with an AI-based agent and clarifies its specific scope of service recommendations for gear units based on lifecycle and installed base data. It explicitly is not a troubleshooting assistant or a general-purpose chatbot. This reduces the risk of misuse and misaligned expectations.

3. Build strong guardrails and test them consistently

Significant effort went into ensuring that the AI cannot provide unsafe operational advice, expose confidential information, or be manipulated into violating data protection rules. Stress testing included deliberately trying to coerce the agent into giving out sensitive contact details or overruling its constraints. The aim was to validate that the system remains within a tightly defined envelope.

4. Align data protection and governance from the outset

The deployment required a full data protection impact assessment, a generic AI assessment, and agreements with workers’ councils and regional entities. Rather than treating governance as an afterthought, it was built into the project timeline. For industrial companies with strong regulatory and internal compliance requirements, this governance-first approach is a prerequisite for scaling AI-powered customer interfaces.

Designing for dual use: external self-service and internal enablement

An important architectural choice was to design the configurator for both external and internal use.

Externally, customers access Alfred via the company’s website without having to log into a restricted portal. This was a deliberate decision. Keeping the interface open allows end users not yet in the company’s systems to engage, identify their equipment, and receive recommendations. That directly addresses the installed base visibility issue.

Internally, sales and service personnel access the same configurator logic directly through the CRM system. This enables:

• Proactive outreach: Sales can generate recommendation documents for specific customers and attach them to opportunities, using them as a basis for conversations.
• Zero marginal cost usage: Internal use of the logic avoids additional AI processing costs, since it is not mediated through the generative agent. That supports heavy internal usage without escalating operational expenses.

By reusing the same underlying lifecycle and package logic in both channels, the organisation ensures consistency of recommendations while diversifying how they are delivered and monetised.

Integrating with CRM: from “send and forget” to structured follow-up

One of the less glamorous but highly impactful shifts is the move away from “send and forget” behaviour.

Previously, service recommendations sent via email or manual documents often lacked systematic follow-up. Salespeople might or might not remember to check back. There was little visibility into how many recommendations were sent, to whom, and with what outcome.

By embedding the configurator into the CRM system, every generated recommendation document now triggers structured sales artefacts:

• Creation of leads or follow-up tasks when a document is produced.
• Logging of the conversation summary into CRM records.
• Linkage between the recommendation, the customer, and the specific units involved.

This transforms each interaction into part of a managed sales process rather than an isolated event. It becomes possible to track conversion, identify which segments respond best to proactive recommendations, and continuously refine both packaging and timing.

The recommendation document as a commercial door-opener

The output of the configurator is a structured, high-level service recommendation document. It does not include pricing or delivery times. Instead, it focuses on:

• Recommended inspections: What to check, and when.
• Recommended spare parts packages: Which categories of components to stock or replace (for example, bearing and seal packages).
• Lifecycle-specific notes: Such as remaining warranty status for new units or exchange offers for very old units.
• Lubrication guidance and relevant reference links.

For customers, the document provides clarity and a sense of being actively supported throughout the lifecycle. For sales, it functions as a credible, technically grounded advisory piece that opens the door to further dialogue.

The commercial impact goes beyond direct follow-up on the specific units covered. Experience from early use shows that customers sometimes place orders for completely different units after receiving a recommendation report because the document triggered a broader conversation about their installed base and risks.

Managing the inventory and lead time trade-off

An essential consideration is how to balance proactive spare parts recommendations with the realities of inventory cost and production lead times.

On one side, customers expect fast availability for critical parts. On the other, stocking every possible critical component for hundreds of thousands of units is economically unrealistic, especially for complex, in-house manufactured parts like gears.

The approach taken acknowledges this challenge:

• For mass-produced, small standard units where repair is not economical, the focus is on promoting complete unit exchange rather than parts-based service. The proactive configurator is used from a certain unit size upwards, where repair and spare parts become commercially sensible.

• For critical but make-to-order components such as large geared parts, lead times remain a constraint. Here, the goal is not to promise instant availability, but to move the decision point earlier in the lifecycle. By encouraging inspections and stock decisions before failure, the configurator creates time buffers both for the customer’s production and for the manufacturer’s supply chain.

• For bought-in components like bearings and seals, inventory strategies are supported by ERP systems that monitor global stock positions. The configurator can then recommend that customers put specific critical parts on their own stock, especially when they operate multiple identical units. Stocking one set of components for ten identical drives is a rational compromise between cost and risk.

Proactive recommendations are as much about time management and risk reduction as they are about immediate order capture.

Learning by iteration, not by big-bang design

The development of Alfred was explicitly framed as a continuous journey. Several practical insights stand out for other organisations considering similar initiatives:

1. Let usage reveal missing data

Initial assumptions about required data fields often prove incomplete. Testing exposes empty fields and gaps, revealing that the perfect data model only emerges through real usage. It should be treated as an iterative refinement of both the user input forms and the CRM data structures.

2. Do not over-engineer up front

Focus on a narrow, high-value use case of lifecycle-based service recommendations and build around that. Broader use cases, such as troubleshooting or multi-domain support, should be deliberately postponed.

3. Treat the launch as a live experiment

Recognise the need for further testing and gradual rollout. This is aligned with the nature of generative AI systems, which often behave differently with real user populations than in lab conditions.

4. Invest in presentation and customer experience

Spend time refining seemingly minor elements such as customer emails, recognising that acceptance and conversion depend heavily on presentation, not just on underlying logic.

Implications for industrial leaders

For manufacturers focused on aftermarket growth and digital transformation:

• Installed base data problems can be reframed as service design challenges. Instead of trying to clean data in isolation, organisations can create digital services that motivate customers to provide and validate the data themselves in exchange for value.

• AI is most powerful when anchored in strong domain logic. Rather than relying on AI to generate technical recommendations, the most robust solutions combine deterministic, expert-derived rules with AI-driven interfaces and automation.

• Proactive service is as much about process as it is about technology. The real impact often lies in how the agent is embedded in CRM, sales workflows, lifecycle planning, and governance.

The future of aftermarket sales is hybrid. Self-service, AI-supported recommendations will coexist with high-touch human engagement. Tools like Alfred do not replace sales or service teams. They equip them with scalable, consistent, and credible starting points for value-creating conversations.