Our articles and insights on AI in sales operations

Almost every customer visit includes the same moment: someone from inside sales shows me an order and says something along the lines of, “We always need to be careful with this customer — they write it differently.” Most of the time, the issue is not typos or missing information. It is units of measure. “50 pcs”, “50 EA”, “50 ST”. They all mean the same thing, but they can be processed differently in the ERP system — with different levels of speed and accuracy.

It is not the big, visible problem. It is the small, recurring one. That is exactly why we have integrated unit conversion and validation directly into AI order processing.

Four areas where unit handling regularly causes challenges

A customer in the electronics industry may write “CTN” in one order, “Cartons” in another, or simply “CA”. For an inside sales employee, this creates another moment of hesitation. For our AI order processing, it is clearly the same unit that is maintained in the ERP system and is automatically standardized accordingly.

Things become more challenging when conversions between volume and weight are required. For example, a customer may order 800 liters of oil, while the ERP system uses kilograms as the primary unit of measure. There is no universal conversion between liters and kilograms because the relationship depends on the specific product. That is why AI order processing uses product-specific density information from your master data to reliably convert the order into the correct unit. This is not a simple calculation trick, but applied product knowledge — automated, without anyone in inside sales needing to know or manually look up density values.

Another common scenario occurs when the same customer orders 20 pieces of a connector that is delivered in cartons of four pieces each. In practice, this can easily lead to mistakes, such as assuming the order refers to 20 cartons. Instead, our AI order processing calculates the correct quantity automatically: 5 cartons.

And where a conversion genuinely does not make sense — for example, when a piece quantity is ordered for a product that is only managed by weight — the system does not silently continue processing. Instead, it flags the case for review. Inside sales managers no longer need to rely on every team member knowing every exception and special case by heart.

A feature you notice most when a problem disappears

That is exactly what makes unit validation so valuable. It does not stand out because of a new window or an additional click. You notice it because a specific type of error simply stops happening. No incorrect deliveries because a unit of measure was overlooked. No follow-up questions about whether “ST” really means “pieces”.

This level of simplicity and reliability is still rarely found in inside sales operations today.

“We already automate with RPA — so why do we also need AI agents?” This is a question we regularly hear from IT leaders. And it is understandable. RPA has proven its value for clearly structured, rule-based processes. In sales operations, however, this approach reaches its limits when processes vary and require decisions. This is precisely where the difference between RPA and AI agents becomes clear.

RPA and AI agents do not automate the same thing

Both technologies are grouped under the broader term “automation” and are therefore often compared with one another. Yet they are based on fundamentally different principles. RPA automates execution. A bot follows predefined rules and carries out defined steps. AI agents, by contrast, operate at a different level. They evaluate information, understand context and select the appropriate course of action within defined boundaries. The difference is therefore not simply a matter of RPA being an older or less intelligent technology, but of two fundamentally different approaches to automation.

The difference becomes clear when processes vary

RPA works reliably as long as inputs, workflows and outcomes are predictable. In sales operations, this is often not the case. Customer inquiries are phrased differently, information may be missing, documents vary in format and exceptions occur regularly. An RPA bot can only handle these situations if they have previously been defined as rules. Otherwise, the process stops or has to be taken over manually.

The issue, therefore, is not that RPA is poor at automation. The decisive question is whether the process itself can be fully translated into fixed rules.

When processes require decisions

This is where AI agents come in. They can process natural language and unstructured information, reconcile different sources of information and select the next action based on the specific context. When they encounter uncertainty, they can also hand a case over to an employee instead of simply stopping the process.

This makes them particularly well suited to processes in which not every possible situation can be defined in advance.

Three questions to distinguish RPA from AI agents

When choosing the right automation technology, the key consideration is therefore the nature of the process itself:

1. How much does the process vary? Can the system handle unknown or unexpected inputs?

2. Does the process require decisions? Does the system follow fixed rules, or does it need to evaluate information in context?

3. How often does the process change? Does the workflow remain stable, or do the underlying rules need to be adjusted regularly?

Anyone who answers these three questions honestly can quickly assess whether RPA or an AI agent is the right choice for their sales operations.

Where AI agents create value in inside sales operations

At enmit, we use AI agents where processes are characterized by high variability, manual decision-making and numerous exceptions. In sales operations, our solutions automate processes such as order entry, quote creation and email routing.

On average, our customers achieve:

• 90% time savings on routine processes

• ROI in under six months

• Less than 1% error rate in automated processing

Our conclusion: Don’t just copy rules. Understand context.

RPA and AI agents are not competing versions of the same technology. RPA is strongest where processes are stable, rule-based and predictable. AI agents become valuable where processes require contextual understanding, decision-making and the ability to handle varying inputs.

The key question, therefore, is not which technology is more modern, but which type of automation best fits the process.

A sales support employee receives a customer inquiry by email: ten line items, different product numbers, and a handwritten bill of materials attached as a photo. By the time the quote has been entered into the ERP system, they have spent twenty minutes searching for products, checking prices, and manually transferring data.

The problem is rarely a lack of knowledge. Instead, the issue is that this knowledge is not structured or connected to the relevant systems. For the person responsible for inside sales, the case is easy to understand because they know where to look – in the ERP, the CRM, or the customer’s most recent email. For an AI agent, things become much more difficult when exactly these connections are missing.

What can a generic AI agent really do – and what can’t it do?

A generic AI agent can draft quotes, format content, and turn predefined requirements into a linguistically polished document. That is where its real strength lies.

But without connections to the relevant systems and a clearly defined process behind it, that is not enough to create a reliable quote. The agent lacks access to validated product data, current prices, and the context stored across the ERP, CRM, and email. The result: fabricated or incorrect product numbers, unvalidated information, and no adherence to the actual sales process. In the end, the quote still has to be manually entered into the ERP – meaning the time savings the AI was supposed to deliver simply do not materialize.

The challenge begins when a process depends heavily on company-specific rules, data, and systems – because these are not automatically part of a generic model. Typical scenarios include:

1. Relevant knowledge may already exist digitally – but it is not structured or accessible to an AI agent. Special agreements, individual customer requirements, or employees’ experience-based knowledge may be stored in emails, Excel spreadsheets, or distributed across multiple systems. In this case, a generic agent often does not even know how to handle this information, even if it theoretically exists.

2. The data is available, but many rules interact with one another. An ERP system may contain standard prices, special pricing conditions, volume discounts, small-quantity surcharges, and delivery terms. A generic agent may well be able to find this information – but the question is whether it can reliably derive the correct decision from it. Is the special price still valid? Does it take precedence over the volume discount? What happens when two pieces of information do not clearly match?

3. There is no clearly defined quote-creation process – including integration with ERP, CRM, and email. Without this integration, an agent can generate text, but it cannot provide a reliable basis for the quote itself. The result is a quote containing incorrect products or unvalidated information – and one that ultimately still has to be manually entered into the ERP.

The more such rules, exceptions and missing system integrations interact, the less sufficient it becomes to simply connect an AI agent to the available data. The challenge is no longer understanding the information, but reliably applying the underlying business logic and processes.

That's why enmit goes a step further

At enmit, we connect the data available in the ERP with the specific business logic of each company – including special pricing conditions, small-quantity surcharges, delivery prices, framework agreements, and the rules governing how this information interacts. enmit is directly connected to the ERP and email, follows a clearly defined quote-creation process, and creates the quote directly in the ERP – instead of merely generating text that then has to be manually entered.

If information is missing or a quote is ambiguous, the system does not simply make an automatic decision. Instead, it recognizes the uncertainty and specifically escalates the case to a human for review.

Benefits our customers gain from this:

• Faster processing: more quotes handled in the same amount of time, without additional headcount – faster responses increase close rates and revenue from existing inquiries

• Centralized knowledge: special terms and business rules live in one place, instead of on sticky notes, in spreadsheets, or in individual employees' heads

• More time for active sales: less manual double-checking means more room for customer relationships and closing deals

"We probably need to hire two more people for inside sales." We hear this sentence constantly and it sounds plausible. More orders, more inquiries, more work, so more staff. But what if there's another way? A recent study by the Institute for SME Research Bonn (IfM Bonn) suggests exactly that.

Are new hires a solution despite the labor shortage?

Mid-sized companies are structurally disadvantaged in the competition for skilled workers: lower wages, less attractive locations, limited training opportunities compared to large corporations. According to the IfM study, this means the Mittelstand suffers disproportionately from unfilled positions — a gap that will only widen further due to demographic change.

Anyone who responds to this exclusively with new hires is competing in a labor market that simply doesn't hold the right talent in reserve. The real lever lies elsewhere: in the tasks the existing workforce is already handling today.

The solution: substitutive use of AI

The study distinguishes between substitutive and complementary AI use. Substitutive means: AI takes over individual, clearly defined tasks — without eliminating jobs. According to IfM Bonn, this is currently the dominant form of AI use in the Mittelstand, and it works not by displacing people, but by relieving them.

Across the case studies examined — in tax consulting, skilled trades, and manufacturing — a recurring pattern emerged: repetitive tasks such as document processing, quote creation, or customer communication are taken over by AI. The time freed up flows into value-creating work.

How enmit brings AI to inside sales

This is exactly what we implement at enmit with our AI software for inside sales. Our goal: giving inside sales teams more time back for active selling – by specifically automating administrative routine processes.

• AI order processing — orders from all common formats such as email, PDF, Excel, XML, or CSV are captured automatically and transferred directly into the ERP.

• AI quote creation — quotes are prepared with AI support instead of being assembled manually.

• AI email routing — customer inquiries are automatically classified and routed to the right person on the team, or directly to our other two AI agents.

None of these applications replace inside sales teams. They give employees back the time that is currently lost to manual forms, data entry, and repetitive administrative work.

Why creating capacity matters more than growing headcount

More people don't create more capacity — less time spent per task does. The IfM study describes this as a short-term effect: substitutive AI use can ease unfilled-position problems, without a single new hire.

Importantly, the study finds that it's precisely low-threshold, easily integrated applications that make the difference in the Mittelstand — without a large IT department, without months-long implementation projects. This is exactly the standard enmit was built for: our AI software for inside sales integrates directly into existing processes and ERP systems, rather than replacing them or requiring large-scale IT projects.

Conclusion: Don't scale up headcount, free up capacity

Three questions help identify the biggest opportunities:

• Where is the team currently spending the most time on repetitive tasks?
  This is where the greatest short-term potential lies.

• Which activities still require human judgment, such as customer consulting or price negotiations?
  The time created through automation should be invested in these higher-value activities.

• How much growth can the existing team support before additional hiring becomes necessary?

Conclusion: Create capacity instead of adding headcount

The skills shortage in inside sales can't be solved by recruiting it away — the labor market simply doesn't allow for that. But it can be eased by freeing existing teams from exactly the tasks that consume their time the most today. AI software for inside sales like enmit already makes that achievable for mid-sized companies today.

Not more people. More room to work for the people already there.

AI is fundamentally changing software. But what does that actually mean for the ERP systems on which mid-sized companies have built their entire process landscape? Those who currently focus primarily on chatbots, copilots, and new AI features are missing the bigger picture. The real change is not happening at the surface level, but deeper down. It concerns the fundamental question of how companies will design their processes in the future, how they will connect them, and how they will organize them into coherent process chains.

Why AI will not simply replace the ERP system

The discussion is often framed as an either-or choice. Either the ERP system remains the central system environment, or a loose ecosystem of AI tools emerges around it and gradually displaces the ERP system. Both views fall short. What is actually emerging is a new division of responsibilities between the ERP system and specialized solutions. And that division is changing where the real value is created within the organization.

What is really changing right now

Two developments are shaping the transformation we are currently seeing:

• More is being built internally — but no longer as large, monolithic custom developments maintained over years. Instead, companies are creating fast, modular process components with embedded AI, low-code workflow automation, and lightweight integrations that specifically close operational gaps.

• The ERP is becoming less of an “all-in-one” system and more of an anchor within a broader ecosystem. The reason: AI is significantly reducing integration costs. And when connecting systems becomes easier and more affordable, specialized tools with real business value become increasingly worthwhile.

In short: more tools, more flexibility — but also greater architectural complexity and more components that need to be carefully orchestrated.

Why ERP systems are not losing importance in the age of AI

Precisely because more is happening outside the ERP system, ERP systems can make better use of their core strengths than ever before. As the system of record, they remain the authoritative data foundation on which all other systems rely. At the same time, they remain indispensable as the central backbone for core processes wherever workflows need to be regulated, auditable, and reproducible. AI can accelerate many processes. It does not, however, eliminate the need for stable and verifiable process chains.

Why generic solutions are not enough on top of the ERP system

As workflows become more complex, require precise output, and demand specialized expertise, a quick and generic implementation reaches its limits. A generic copilot can summarize texts or suggest formulations. It cannot, however, fully review an incoming request, classify it correctly within the ERP system, reconcile it with master data, and process it automatically without requiring someone to intervene manually at the end.

This is precisely where new categories of software are emerging. These solutions are closely connected to the ERP system while going significantly deeper from a functional perspective than a generic assistant. They use the existing data foundation to create robust, process-reliable automation. In areas with high volumes, clear patterns, and little tolerance for errors, such as sales operations, this is what makes the difference between a useful feature and a process that the business can actually rely on.

Three decisions for the ERP strategy of the coming years

Companies planning their ERP and AI strategy for the years ahead will need to address three key questions.

• Which ERP processes can genuinely be automated with AI — especially those with high volumes, clear patterns, and measurable manual effort?

• Where is a generic AI tool sufficient, and where is a specialized solution required because processes are complex, business-critical, or highly sensitive to errors?

• What architecture brings both together — a stable ERP at the core, complemented by specialized solutions with clean integrations?

Conclusion: The foundation matters, but the real difference is built on top

The ERP system is not disappearing in the age of AI. On the contrary, its role as a reliable foundation is becoming more important. But that foundation alone is no longer enough to create a competitive advantage. The difference is increasingly made where specialized software builds on the ERP system and turns stable data into faster, more reliable processes.

A stable foundation. Decisive value on top. That is where things get interesting.

Many AI projects in inside sales don't fail because of the technology, but because of how they're introduced. The most common mistake: leadership decides top-down which tool to roll out, then announces it to the entire team. What's missing is proof that it actually works — and that's not something you can mandate, only something people experience. The more effective approach thinks the other way around: bottom-up instead of top-down.

Why an announcement isn't enough

A kick-off presentation can explain, but it can't prove. Sales reps who've worked with the same spreadsheets and ERP screens for years need more than a promise that a new tool will improve their day-to-day. They need proof — ideally by experiencing themselves that it works. Without that proof, something quietly becomes the norm in many companies: the tool exists, but nobody uses it. Employees go back to familiar processes the moment no one's watching. Not out of laziness, but because uncertainty outweighs a one-time announcement.

Bottom-up instead of top-down

The more effective approach doesn't start with a tool — it starts with a conversation. Before any use case is even defined, it's worth talking to the inside sales teams themselves. Where do they spend most of their time on manual, repetitive work? Which workflows follow patterns enough to automate? Where do they see the biggest potential? Nobody can answer these questions more reliably than the people who run the process every day. This flips the usual order: leadership doesn't decide what gets automated — the team provides the answer that shapes the use case.

The proof of value

That conversation produces a concrete use case for the team — automated order processing, for example. That's exactly what the proof of value tests. It shows up not in an opinion, but in a result: does the automation hold up reliably in real day-to-day work, or not?

What makes a pilot a convincing proof of value

Not every use case is equally suited to a pilot. Three criteria determine whether a proof of value in inside sales is meaningful:

• A use case with a high volume of recurring, rule-based tasks — such as order processing or quote creation

• Clear success criteria defined before the start, such as time saved or fewer errors on a specific task

• A timeframe of weeks, not months

If a pilot meets these three criteria, it delivers a solid result within a short time — the basis for the next decision: scale it, or not.

Building blocks for successful scaling

What makes scaling successful comes down to a single criterion: it must not create a major additional burden at any level.

• For leadership, that means: no new strategic decision, just extending a system that's already proven

• For IT, that means: no months-long integration project

• For inside sales employees, that means: no new experiment, just a solution that's already working within their own company

Where these three building blocks hold true, scaling happens almost on its own.