Ce contenu n’est disponible qu’en anglais.

Étude de cas

Taux de réussite de 83,5 % en libre-service :la stratégie de recherche en IA de SAP au cœur de l'action

Voyez comment SAP a utilisé Coveo AI Search pour dévier 1.6M cas et économiser €186M annuellement, supportant 300M+ utilisateurs à travers le monde.

1,64
0
83,5
83,5
30
0
Enterprise software, cloud computing
Headquartered in Walldorf, Baden-Württemberg, Germany
300M+ cloud subscribers
€34B+ (non-IFRS) in FY2024
109,000+ employees across 157+ countries

As a global leader in enterprise applications and business AI, SAP stands at the nexus of business and technology. For over 50 years, organizations have trusted SAP to bring out their best by uniting business-critical operations spanning finance, procurement, HR, supply chain, and customer experience.

For a company serving over 300 million cloud subscribers, every support case matters, and enabling customers to find their own answers instead of seeking assisted support is business critical. But SAP lacked visibility into their digital experience, and had little data to prove self-service’s value. On top of that, a wide variety of customer search styles and content distributed across multiple systems presented a challenging series of obstacles.

“There’s no commercialized search solution in the SAP suite,” said Michelle Lewis-Miller, Head of Digital Experiences and Voice of Customer for SAP Customer Support. “We weren’t hitting our self-service targets the way that we wanted to. We couldn’t measure anything. We couldn’t see what was going on.”

Tasked with leading a search party for search, Lewis-Miller navigated both external vendors and SAP’s own builder mindset. Few companies have ever met SAP’s high expectations for third-party integration. After two or three years of convincing, Lewis-Miller added the Coveo AI-Relevance™ Platform to that esteemed shortlist in 2023.

The Shocking Concur Pilot That Redefined SAP’s Support Strategy

SAP Concur, a travel and expense platform, had already been using Coveo for years to deliver intelligent search across its support experience, long before the rest of SAP made the leap as well.

So, when the hype for generative AI began to trend, Lewis-Miller immediately saw that Concur could be an ideal testing ground for the new technology. SAP joined Coveo’s design partner program to help shape the path that Coveo’s Relevance Generative Answering would take.

“When we piloted Relevance Generative Answering in Concur, it went so well it was shocking,” Lewis-Miller said. “We saw such a huge decrease in case submission that it changed our whole budget strategy for the next year.”

Within six months, SAP Concur saw a +30% drop in customers submitting cases after self-service search attempts — an €8M reduction in annual cost-to-serve. For many customers, this was their first experience with generative AI, and it was an immensely positive one, helping them find what they needed faster than ever before. Internally, agents received the same generative experience for faster knowledge discovery, shortening case resolution times.

As news of the success spread throughout SAP, leaders at all levels approached Lewis‑Miller and her partners in IT about replicating that success for their own teams. The pilot also made SAP confident about greenlighting a much larger rollout.

30
0
8
8

When we piloted Relevance Generative Answering in Concur, it went so well it was shocking.

We saw such a huge decrease in case submission that it changed our whole budget strategy for the next year.”

Michelle Lewis-Miller, VP, Head of Strategy and Transformation for SAP Concur

The Ultimate Stress Test: Scaling Generative Search to SAP for Me

20
0
4,2
0

Malgré le succès de l'intégration de Coveo dans l'expérience de soutien de Concur, le déploiement de la plateforme de recherche en IA sur le portail libre-service de SAP, SAP for Me, serait un défi d'envergure olympique. SAP for Me sert de plaque tournante vitale pour des millions de clients SAP, centralisant les alertes critiques, les mesures et les informations relatives à l'ensemble du portefeuille de produits SAP d'une organisation par le biais d'un point d'accès unique.

"C'est là que la complexité nous a mis des bâtons dans les roues, même avec une organisation informatique aussi incroyablement talentueuse que la nôtre pour travailler sur ce projet. SAP for Me n'est pas le point d'entrée d'un seul produit issu d'un seul corpus de connaissances. Il s'agit de centaines de produits avec entre 14 et 20 corpus de connaissances", a déclaré M. Lewis-Miller.

"Et vous essayez de répondre aux questions d'un public varié : des ingénieurs ayant 30 ans d'expérience dans les produits SAP, à Sally Jane des déserts du Nevada qui porte 10 chapeaux pour son magasin général. Les personas ne pourraient pas être plus différents".

L'intégration de l'IA générative dans le portail mettrait à l'épreuve la capacité de SAP à innover dans les limites de ses normes rigoureuses en matière de confidentialité des données et de sa vaste architecture informatique. Lewis-Miller et son équipe de projet ont été en mesure d'atteindre un équilibre entre la conformité stricte et la convivialité centrée sur le client, grâce à la flexibilité de la plateforme de Coveo.

After a thorough review process, SAP went live with Coveo Relevance Generative Answering on SAP for Me. The experience leverages hybrid search — a combination of lexical and semantic retrieval paired with behavioral machine learning models — that would speed user access to relevant knowledge.
Prefer a shareable version?

We’ve gone from questioning whether third-party search could even fit at SAP to making [Coveo] a core pillar of our intelligent search strategy.

Pretty soon, customers who pop open Joule [SAP’s AI Copilot] will also get the benefit of Coveo in their knowledge search.”

Michelle Lewis-Miller, VP, Head of Strategy and Transformation for SAP Concur

Wait, Is It Broken? Figuring Out How to Measure What Really Matters

When SAP deployed Coveo Relevance Generative Answering on SAP for Me, they anticipated that it would deflect simple questions.

“We expected it to address the low-hanging fruit,” Lewis-Miller said. “But it wasn’t at all like what we saw with Concur. In fact, we initially asked Coveo if something was broken in their backend.”

At first, SAP saw searches per session and its case submit rate go up among queries that received a generated answer — typically signs that the search was less relevant, not more. But by staying calm under pressure and partnering with Coveo to review the analytics, Lewis-Miller and her project team were able to root out the cause: Relevance Generative Answering on SAP for Me was tackling a different and far more complex set of queries than it had on SAP Concur.

“We were deflecting a more valuable subset of cases. It was so unexpected that we didn’t know what we were seeing when we first looked at the data. Instead of breaking it out by ‘I’m going to see if users who get a generated answer are more or less likely to submit a case,’ we learned that you have to look at all the data, generated answer or not,” said Lewis-Miller.

“When we did that — when we changed our perspective and our assumptions — and looked at the overall number, we saw that in reality the number of submissions had gone down, and the relevancy had absolutely gone up.”

1,64
0
83,5
83,5

Since launching in 2023, Coveo’s intelligent search has deflected 1.6 million cases, unlocking an 83.5% self-service success rate. And generative search is now adding millions more to those already staggering numbers.

SAP’s analysts use Coveo analytics to generate insights for executives and also find ways to improve the UX and influence content strategy. An unanticipated source of value was in Coveo-surfaced citations. Lewis‑Miller described how the platform’s hybrid retrieval referenced a nuanced result set to not only respond with a direct answer but also deliver expertly curated source links.

The enhanced Coveo experience proved to be so effective that it even started to win over veteran engineers, who had previously preferred to entirely bypass search by using ID numbers or direct URLs.

“I’d tell them, imagine a world where you can just type in the specific piece of information you’re looking for, and it will pop up,” said Lewis-Miller. “They came around. One SAP-certified engineer with 15 years of experience with our products told me he uses it every time.”

Search That Customers Didn’t Expect — But Now Depend On

With the Coveo platform, SAP simplifies searching across touchpoints like their SAP’s built-in support application, community, customer support, and even internally — across all use cases, a total of 47 sources and 11.2M documents indexed.

“We’ve gone from questioning whether third‑party search could even fit at SAP to making [Coveo] a core pillar of our intelligent search strategy,” said Lewis‑Miller. “Pretty soon, customers who pop open Joule [SAP’s AI Copilot] will also get the benefit of Coveo in their knowledge search.”

SAP’s AI search journey is more than a cost-savings story. It’s a story of cultural change inside one of the world’s most established technology companies, led by champions like Lewis-Miller and her visionary IT colleagues. It’s about proving that better customer experience doesn’t just reduce costs — it builds trust, resilience, and momentum for the future.

As Lewis-Miller noted, the biggest surprise wasn’t just the financial return. It was the customer reaction:

“The most common feedback was: this is cool, this feels like Google, and I don’t have to go to so many places anymore.”

For SAP, search is no longer a challenge to overcome; it’s a strategic enabler. With Coveo, every digital interaction is an opportunity to deliver clarity, value, and confidence at scale.

Timeline

  • Continuing co-innovation
    • Expanding Generative Answering to new use cases.
    • Grounding SAP Joule through Coveo PR-API
  • 2025

    Scaled CRGA to SAP for Me

    • Unified 20 sources & 4.2M documents
    • Supporting 300M+ global cloud users
    • Achieved 83.5% self-service success rate
  • April 2024

    Achieved significant business outcomes:

    • 5% reduction in cases per 1000 search sessions
    • 80% reduction in search queries per customer visit
    • 64% reduction in content gap
    • Achieved €8M in savings
  • March 2024

    Launched Coveo Relevance Generative Answering (CRGA)

    • Pilot started in SAP Concur
    • Fit for production after just 1 month of testing
  • July 2023

    Joined Coveo’s GenAI design partner program

    • Shaped Relevance Generative Answering
    • Participated in early testing and product feedback
  • Early 2023

    Hit performance benchmarks early

    • Met 6-month targets in under 3 months
    • Expanded Coveo to SAP Global: customer community, internal KB, SAP for Me, Concur support portal
  • 2022-2023

    Drove early business impact

    • 20% decline in case volumes

    • Secure index and hybrid search
    • 12 out-of-the-box AI models driving relevance
  • 2018

    SAP Concur became a Coveo customer

    • Focused on self-service, agent support & intranet search

SAP for Me: 1.6M Cases Deflected

SAP for Me: 1.6M Cases Deflected