AI in Sales: Zeller+Gmelin Triples Pipeline

Case Study · 7. September 2026 · Mohsen Ghulami

AI in sales in a real mid-sized company case: How Zeller+Gmelin tripled its pipeline. Read the practical guide with setups, figures, and mistakes.

Anyone in the manufacturing SME sector who still believes in 2026 that good products automatically generate a good pipeline has confused sales with the spare parts business. AI in sales is not relevant because a bot writes nice emails, but because technical markets have become too large, too fragmented, and too slow for gut feelings. This is uncomfortable. Especially for companies that have won business for decades with application engineering, trade fairs, and relationships. Zeller+Gmelin, a Swabian specialty chemical manufacturer for industrial lubricants and printing inks near Stuttgart, is exactly such a case for me — technically strong, internationally visible, but before Amplifa, with too little systematic new customer acquisition.

I write this as Mohsen Ghulami, GTM Engineer at Amplifa. Not as an analyst who prints out three market studies and then explains SMEs. My job is to dissect target markets, consolidate data sources, build outbound sequences, and find the point where a sales engineer no longer says: "Another lead," but rather: "I want to handle that appointment myself."

At Zeller+Gmelin, it wasn't about a simple lead tool. It was about how a mid-sized company with around 900 employees, production and sales structures in several regions, and a portfolio of high-performance lubricants, gravure inks, flexo inks, and UV-curing systems could triple its pipeline in nine months — without hiring 15 new SDRs. Well, almost. It was also about killing a few old sales reflexes.

Problem Statement: Why AI in Sales was Necessary Here

The problem at Zeller+Gmelin was not demand. In packaging printing, regulations, food contact, low-migration systems, and sustainable packaging are driving the market. In the lubricant sector, OEMs, maintenance, and plant managers talk about energy efficiency, service life, condition monitoring, and less downtime. According to IndexBox forecasts, the market for UV-curing inks is expected to grow at a CAGR of around 7.9 percent until 2035; thermochromic and variable specialty inks are estimated at about 7.5 percent CAGR according to market analyses. At the same time, new formulations in critical applications often require 6 to 18 months for approval. This smells like pipeline. But it also smells like labs, approval loops, and very long CRM notes.

If you don't implement AI in sales in such an environment, something mundane happens: the field sales team gets stuck with existing customers. Not because they are lazy. Because they are needed. For customer trials, complaints, sampling, REACH questions, delivery dates, calculations, factory visits. Thomas, Head of Industrial Chemistry Sales in Southern Germany, told me quite dryly in April 2025: "We didn't have a lead gap. We had a capacity gap before the first real conversation." That sentence stuck with me because it applies to many mechanical engineering companies, chemical companies, and component manufacturers. Phoenix Contact, Festo, Wittenstein — everywhere you see variations of the same pattern: the know-how resides with a few people, and these very people are also expected to systematically open up new markets.

Before Amplifa, the Zeller+Gmelin pipeline was heavily dominated by known accounts. The share of new customers was internally below 20 percent. The lead sources? Trade fairs, referrals, existing distributors, occasional website inquiries, individual LinkedIn research by motivated sales colleagues. It works. Until it's no longer enough. Especially when competitors like Siegwerk, Flint Group, Fuchs SE, Shell, or Castrol appear with larger sales networks, and the long tail of suitable target customers is distributed somewhere between Bielefeld, Lyon, Eindhoven, and Brno.

We knew there were many suitable target customers out there — packaging printers, mechanical engineers, OEMs. But we simply lacked the capacity to systematically identify and approach them properly.

— Thomas, Head of Industrial Chemistry Sales, Stuttgart area

The dangerous thing about it: you notice it late. For a year, the forecast table still looks decent because existing customers are pushing projects. Then two key accounts postpone investments, a packaging customer waits for regulatory approvals, an OEM freezes its program — and suddenly the pipeline is no longer a funnel, but a museum piece. Nicely labeled. No movement.

Overview: What This Practical Guide to AI in Sales Shows

I'm not showing "AI automates sales" here. That's not entirely true, many sell it that way. I'm showing the concrete setup with which Zeller+Gmelin tripled its active sales pipeline within nine months: from segmentation to data enrichment and AI-SDR outbound, to handover to technical sales engineers. The relevant point is not the email. The relevant point is the operating system behind it.

The steps in this guide:

• Step 1: Translate existing data and historical orders into target customer patterns
• Step 2: Enrich market segments with buying triggers and technical signals
• Step 3: Build separate AI-SDR sequences for printing inks and lubricants
• Step 4: Design qualification so that field sales engineers don't waste time
• Step 5: Schedule weekly pipeline reviews, feedback loops, and messaging corrections

The result: The number of sales-qualified opportunities increased from an average of about 30 per month to over 90. The pipeline value tripled compared to the baseline. The share of new logos increased from under 20 percent to almost half of the active pipeline. And yes, these numbers are only exciting because they were generated in a market where an "appointment" doesn't mean someone clicks a demo on Thursday. It often means: initial technical discussion, samples, machine parameters, substrate, approval process, purchasing, QA, production manager. Sometimes the next step smells of solvent and warm machine oil.

Step 1: Building Real Target Customer Patterns from CRM Data

The first mistake in such projects is almost always the same: you start with lists. "Give us 5,000 packaging printers in Europe." Please don't. That's not a go-to-market strategy; that's data garbage with an export function. At Zeller+Gmelin, we first looked at the existing CRM, historical offer data, won projects, and lost opportunities. Not perfect. No CRM is perfect. But even a messy CRM tells the truth if you search it for repetitions.

We looked at which accounts truly fit well: medium-sized flexo and gravure printers with food packaging, converters with regulatory pressure for low-migration systems, industrial customers with high plant availability, OEMs with maintenance and lubricant programs, distributors in markets with technical consulting needs. Then came the tough question: Which characteristics were visible before the first conversation? Because if a signal only becomes visible in a lab test, it doesn't help for prospecting. Visible characteristics included, for example, printing processes, certifications, packaging segments, machine park indications, site structure, export markets, sustainability communication, open positions in quality or application engineering.

A concrete example: a packaging printer in Northern Italy, 180 employees, flexible packaging, indications of gravure and flexo, ISO certifications, customers from food and pharma, website with a sustainability page, but no clear communication on low-migration UV. For a generic SDR, that's a "printing company." For Zeller+Gmelin, that's potentially an account where regulatory pain, technical application, and margin come together. This is precisely where AI in sales must help: not to spit out more contacts, but to form better hypotheses.

We built a fit model for this. No black magic. Rather, a structured scoring logic with weightings that we validated with sales and application engineering. An account received points for segment, company size, technology, regulatory relevance, geographical priority, and presumed use case. Minus points were given for operations that were too small, pure commercial printers without a packaging focus, trading companies without technical depth, or markets where Zeller+Gmelin deliberately did not want to be aggressive. Sounds dry. It was. But dry work triples pipelines more often than a creative subject line workshop.

The most important discussion wasn't which AI we use. The most important discussion was: Who do we not want to approach anymore at all?

— Julia, Marketing Operations, Ulm

Step 2: Enrich Market Segments with Signals

After the ICP work came the enrichment. For printing inks, we separated target segments: flexible packaging, labels, decorative papers, special applications, UV-curing systems. For lubricants: mechanical engineering, metalworking, food production, automotive suppliers, wind and energy applications, industrial maintenance with condition monitoring maturity. This sounds like an industry slide. In the tool, it became a data model with fields, sources, and decision rules.

We combined firmographic data, public website signals, job postings, certificates, product pages, trade fair exhibitor lists, LinkedIn company information, and partly technographic data. An example from May 2025: If a mechanical engineering company in Baden-Württemberg had job openings for "Predictive Maintenance Engineer" and "Service Sales Manager," it received a higher signal weight for certain lubricant campaigns. Why? Because condition monitoring rarely comes alone. Those who measure machine conditions will eventually ask about wear, intervals, oil quality, additive packages, and TCO. The conversation then doesn't start with "We have lubricants," but with "How reliable is your data on friction and service life in the field?"

For packaging printers, the logic was different. There, we looked for terms like food packaging, flexible packaging, low migration, UV, LED-UV, gravure, flexo, pharma packaging, sustainability report, solvent-based systems, or water-based inks. Not every term was a buying trigger. But combinations became interesting. A company that communicates food packaging, flexo, new sustainability goals, and growth in Benelux is not a random contact. It is a hypothesis with an address.

From our implementations, we know: In technical SME projects, 60 to 70 percent of later accepted opportunities are not the accounts with the highest general company score, but the accounts with two to three very specific buying signals. At Zeller+Gmelin, these were, for example, packaging printing plus food compliance plus an indication of UV or flexo technology. In the lubricant sector, it was OEM service business plus condition monitoring language plus an international installed base. This is not a textbook sentence. This is the pattern I see when I put reply data from outbound sequences next to CRM acceptance rates on Mondays.

The difference seems small. It is brutal. An account with high revenue and the wrong trigger wastes time. A medium-sized account with a genuine trigger opens doors. Andrea, Head of Sales at a hidden champion in Bielefeld, told me in June 2025: "Our best new customers never looked glamorous in the database. They just had exactly the problem we could solve." Exactly.

Step 3: Building AI-SDR Sequences for Technical Buying Centers

Only now did outbound come into play. Many start here. Wrong. Without segmentation logic, AI writes polite harassment. At Zeller+Gmelin, we separated the sequences by business unit, use case, and persona. Printing inks had different messages than lubricants. Sustainability managers had different messages than production managers. Quality managers had different messages than purchasing. And no, we didn't build 48 variants because someone loves personas in a workshop. We built the variants that measurably have different objections, different language, and different next steps.

A typical printing ink workflow went like this: Account is qualified by the fit model, relevant contacts are identified, contact data is verified, then the AI-SDR generates a sequence of email, LinkedIn touch, and optional call task. The first email does not reference "innovation" or "partnership." It references the presumed technical context: flexible packaging, food contact, migration, UV/LED-UV, or gravure. The second message provides a brief proof point. The third asks for the right contact person for technical approvals. After a positive response, a sales pitch is not immediately launched, but a discovery block is started: printing process, substrate, regulatory requirements, current bottleneck, timing.

For lubricants, the sequence was more down-to-earth. No colorful packaging terms. More maintenance. More downtime. More gears, compressors, metalworking, energy consumption, approvals. A possible opening line was along the lines of: "We see at several OEMs that lubricant selection is increasingly linked to condition monitoring programs. Is this already part of your service or maintenance strategy?" Short. Not a novel. The AI-SDR adapted the message based on the account — mechanical engineering, food production, automotive supplier, or energy application.

The handover was important. A lead was not given to sales just because someone wrote "interesting." That's too soft. We defined handover criteria: confirmed application area, relevant role, timeframe, technical pain or project trigger, minimum fit for the segment, no obvious exclusion by region or portfolio. Only then was an opportunity or a qualified lead created in the CRM, including a summary, conversation notes, presumed product area, and recommended next action. A sales engineer should understand in 90 seconds why this contact deserves their time.

The biggest change was not just the quantity of leads, but their accuracy. Suddenly, we had conversations with packaging printers who could really use our low-migration systems.

— Anna, Head of Sales Printing Inks, Stuttgart

Steps 4 and 5: Advanced Workflows for Pipeline Management

1. Radically limit CRM fields: At Zeller+Gmelin, we didn't introduce 40 mandatory fields, but a few fields that influence decisions: segment, application, technology, regulatory context, project phase, next technical step. If a field doesn't trigger an action, it's decoration.
2. Integrate human-in-the-loop for technical statements: The AI-SDR was allowed to formulate hypotheses, but not make binding product promises. Statements on REACH, food contact, approvals, or service life were formulated as conversation starters and, if necessary, handed over to application engineering.
3. Weekly pipeline review instead of monthly report: Every Tuesday, new replies, accepted leads, rejected leads, and no-fit reasons were reviewed. The meeting lasted 35 minutes. If it lasted longer, the dashboard was usually bad.
4. Use no-fit reasons as training data: "Too small," "wrong printing process," "only dealers," "no technical project," "region not prioritized" — these reasons were fed back into the scoring. Not as a PowerPoint learning, but as a rule change.
5. Shorten or extend sequences based on segment results: Packaging printers with a clear compliance signal received shorter, direct sequences. Mechanical engineers with service and condition monitoring signals often needed more context because the buying trigger was distributed.
6. Set sales acceptance as the North Star: Not MQLs. Not open rates. Not "generated contacts." What mattered was: How many handed-over opportunities does technical sales accept without rolling their eyes?

The advanced steps sound like process discipline. And they are. That's precisely why they work. I have yet to see a B2B outbound project fail due to a lack of creativity. Many fail due to a lack of feedback. The AI-SDR then doesn't learn, sales doesn't trust the system, marketing optimizes for pretty numbers, and after three months someone says: "AI doesn't work for us." "That doesn't work for us," a CSO from Nuremberg recently told me. When we looked at the data, it turned out: 72 percent of the rejected leads came from two segments that should never have been targeted. The AI wasn't the problem. The segment approval was.

At Zeller+Gmelin, we therefore looked particularly hard at rejection reasons. If a sales engineer rejected a lead, there had to be a reason. Not as control. As learning material. Once, "wrong contact person — purchasing too early" appeared multiple times. So we changed the sequence: For certain packaging accounts, quality, application engineering, or production management were approached first, purchasing only later as a stakeholder. The appointment quality increased. Small change. Big effect.

Tool and Workflow Checklist for AI in Sales

The table is simple. Deliberately so. Complexity isn't in 17 tools, but in the decisions between them. I often see stacks with HubSpot, Salesforce, Apollo, LinkedIn Sales Navigator, Clay, ChatGPT, Make, Zapier, and some intent provider. All individually useful. Together, often a spaghetti mess. At Zeller+Gmelin, the architecture was leaner: Amplifa as the AI-SDR and orchestration logic, CRM as the system of record, defined data sources for enrichment, clear handover to sales. That was enough because the process questions were clarified beforehand.

Amplifa Sales Audit Check where your pipeline is leaking today: ICP, data quality, outbound setup, CRM handover, and sales acceptance in a structured audit.

Business Impact: What Tripling Practically Meant

A tripled pipeline sounds good. But in the SME sector, what matters is what changes in everyday life. At Zeller+Gmelin, it meant: the field sales team had to do less cold research and could spend more time in technical discussions. According to internal evaluations, field sales engineers spent about 30 to 40 percent more time on site visits, sampling, tests, and approval processes instead of initial contact. That's not glamorous. But that's exactly where orders are generated in specialty chemicals.

The second effect was more strategic. The new logo share in the active pipeline increased from under 20 percent to approximately 45 to 50 percent. Suddenly, Zeller+Gmelin saw not only the well-known major customers and distributors, but also smaller converters in Eastern Europe, packaging specialists in Benelux, OEMs with service programs, food producers with special lubricant requirements. Some of these accounts might have passed by at a trade fair before. Maybe. In the noise of a hall in Düsseldorf, between carpet, brochure stands, and the smell of fresh print on sample foils.

The third effect was speed. The time from first contact to technical sampling decreased by about 20 to 30 percent in prioritized segments. Not because approvals suddenly became shorter. They remain tough. But because low-fit conversations were sorted out earlier, and high-fit conversations started better prepared. If the sales engineer already knows before the first call that a converter has food packaging, flexo, and low-migration issues, the conversation starts at a different level. Less small talk. More substrate.

In the lubricants sector, identifying the right contacts was always difficult. The platform not only showed us new OEMs but also the maintenance managers and plant engineers who really wanted to talk about condition monitoring.

— Markus, Head of Lubricants Division, Göppingen

Why a Pure Inbound Strategy is Insufficient for SMEs

I'll go out on a limb here: Any technical SME that relies solely on inbound will lose the best opportunities to louder competitors. Inbound works for clearly defined problems, high search activity, and short decision paths. But which maintenance manager Googles "high-performance additive package for energy-efficient gears with condition monitoring relevance" at the right moment? Which packaging engineer publicly searches for a change in low-migration ink systems when quality, purchasing, and production are arguing internally?

Of course, you need good content. Julia from the marketing team in Ulm used exactly that: from the segments Amplifa identified, landing pages and discussion documents on LED-UV flexo printing, low-migration systems, and bio-lubricants were created. But content alone waits. Outbound moves. The combination was the key. First target customer logic, then suitable content, then segmented AI-SDR, then human technical sales. Not the other way around.

This is also why I often find "lead generation" too narrow a term. It's not about leads. It's about market coverage. Which suitable companies do we not yet know? Which ones know us, but not for this use case? Which ones currently have a signal indicating willingness to change? Which ones should we deliberately ignore? These questions determine pipeline quality. Not the subject line with the first name.

FAQ: Frequently Asked Questions about AI in Sales

How long does it take for AI in sales to generate measurable pipeline?

At Zeller+Gmelin, we saw the first reliable signals after a few weeks: replies, accepted initial conversations, initial technical qualifications. The real pipeline impact became apparent over nine months. This is consistent with technical B2B markets. Anyone expecting revenue after 14 days is either selling a very simple product or deluding themselves. For SMEs with 9- to 18-month sales cycles, the early KPI is not revenue, but accepted opportunities with a clear technical next step.

Does an AI-SDR replace the field sales team?

No. And if someone claims that, I'd be cautious. An AI-SDR does not replace the sales engineer who stands at the machine at the customer's site, recognizes the smell of overheated oil, or understands in the print shop why a substrate is causing problems. The AI-SDR replaces the manual search for suitable accounts, the initial structured approach, and parts of the pre-qualification. The human comes in later — but better prepared.

What data is needed for AI in sales in SMEs?

You need less perfect data than many think, but more clarity about good customers. Historical orders, won and lost opportunities, product areas, target segments, no-fit criteria, regional priorities, and a few hard exclusions are often enough to start. At Zeller+Gmelin, it wasn't the CRM alone that was decisive, but the combination of CRM, sales knowledge, application engineering, and external signals such as certificates, printing processes, job postings, and sustainability topics.

Amplifa Product Amplifa's AI-SDR platform connects target customer identification, data enrichment, personalized outbound, and CRM handover for B2B sales teams.

Three Learnings for Manufacturing SMEs

Learning one: Segmentation beats personalization. An email with a company name, industry, and a nice opening does little if the segment is wrong. At Zeller+Gmelin, the best results were achieved where use case, technology, and buying trigger matched. Printing inks and lubricants had to be considered separately. Actually logical. Yet often ignored because a common campaign setup is more convenient.

Learning two: Sales acceptance is tougher than reply rate. A high response rate can be a warning sign if the responses come from the wrong target groups. For technical products, what matters is whether the field sales team accepts and processes the lead. At Zeller+Gmelin, therefore, optimization was not for vanity metrics, but for accepted opportunities, initial technical discussions, and progress towards sampling.

Learning three: AI needs boundaries. The AI-SDR was not allowed to promise technical approvals, invent compliance statements, or guarantee product performance. It was allowed to form hypotheses, ask questions, qualify contacts, and provide context. This boundary was crucial for trust. Especially in chemistry, mechanical engineering, medical technology, or automotive, false precision is more dangerous than no automation at all.

For me, Zeller+Gmelin is therefore a good customer story because it doesn't smell of Silicon Valley fantasy. No "we automated sales." No dashboard theater. Rather, Swabian sales work with AI support: clearly define target customers, take data seriously, test sequences, evaluate rejections, improve technical handovers. The result is a tripled pipeline. Not because AI works magic, but because it takes over work that sales people should have been doing on the side for far too long.

Full Success Story Read the full customer story about Zeller+Gmelin and how Amplifa helped build a three times larger sales pipeline.

The 3 Most Important Takeaways

1. AI in sales only works in SMEs if it is clear beforehand which accounts should not be processed. No-fit is not a marginal issue, but pipeline protection.
2. The best AI-SDR architecture combines external buying signals with internal sales knowledge. At Zeller+Gmelin, specific combinations such as food compliance plus flexo technology or OEM service plus condition monitoring were stronger than general company size.
3. The field sales team is not replaced, but relieved. The machine takes over search, pre-qualification, and context. The human takes over technical consulting, trust, and approval processes.

When I look at such projects, I rarely see companies that lack market opportunities. I see companies that find their opportunities too late, too randomly, or too manually. At Zeller+Gmelin, that has changed — and somewhere in a CRM, there is now an account that probably would have only been noticed at the next trade fair before.