# SAP report: Only 6% of US CX leaders say systems are connected for AI buyers

By MarketScale Newsroom · Published 2026-11-19 · Marketing Tech on MarketScale
Canonical: https://www.marketscale.com/industries/marketing-tech/sap-finds-just-6-of-us-companies-have-systems-connected-for-ai-buyers

> 93% of US CX decision-makers feel ready for AI-assisted buyers. 6% have connected systems. The fix runs through pricing, inventory and ERP data.

## Key points

- AI assistants work from whatever price, stock and service data they can access, so disconnected data raises the risk of incomplete or stale answers. In SAP’s survey, 6% of US CX decision-makers (3% in Germany) said all relevant systems are connected end to end.
- A sharper question for the next stack review: does the price an outside AI agent quotes come from the same system that produces the invoice?

Ninety-three percent of US customer experience decision-makers told SAP they feel prepared for customers who shop with AI. Six percent said all their relevant systems are actually connected end to end. In Germany, the connected share was 3%.

Those figures come from SAP's AI in Customer Experience Report 2027, subtitled The Autonomy Gap, which SAP Customer Experience CMO Jessica Keehn introduced on the SAP News Center on Nov. 12. Four days later, MarTech's Mike Pastore read the findings as a to-do list for marketing operations. MarTech rounds the confidence number to "more than 90%" of CX leaders; SAP's own write-up puts it at 93% for US decision-makers.

The survey drew 4,800 respondents across the US and Germany: 4,000 consumers, 400 CX practitioners and 400 CX decision-makers. SAP sells the CRM and ERP software that would close the gap it measured, which is worth keeping in mind, and the readiness numbers are self-reported.

Even with that caveat, the split is too wide to ignore. When an AI assistant compares vendors on a buyer's behalf, it works from whatever price, stock and service data it can find. That raises a practical question for the marketing ops leader who runs the stack: whose systems is the agent actually reading, and are they wired to each other?

## The shortlist forms before the website loads

Keehn describes a customer journey that has split in two. Discovery, research, comparison and shortlisting increasingly happen inside AI tools. The purchase, fulfillment, service and return still belong to the business. Consumers and business buyers both do their early research this way, and business buyers do it almost universally.

Block Field

SAP has a name for the person moving between those halves: the hybrid AI buyer, neither fully AI-driven nor fully human-directed, and expecting the business to be ready whichever side they arrive from. By the time that buyer reaches a brand, Keehn writes, the shortlist has already been made.

B2B deals still slow down in the second half. SAP notes that negotiated prices, contract terms, approvals and implementation plans still require human work, so an AI can find the right supplier in minutes while closing the deal takes much longer. The first cut, though, happens before a sales rep knows the account exists.

The mechanism is blunt. If an AI engine can't parse an offering, it won't recommend it, Pastore writes, which means specs, pricing, availability and multimodal assets such as photos, video and voice have to be structured and accurate enough for an agent to read. SAP Customer Experience Chief Product Officer Balaji Balasubramanian made the same point: offerings the AI engine doesn't know about don't surface at all. Pastore adds that brands need to spell out non-price strengths such as expertise, service commitments and reliability, or an AI comparison will flatten them into a commodity.

SAP's post goes a step further. It argues that discoverability is a moving target, because models keep re-evaluating what they surface, and that a one-time catalog audit or SEO project no longer holds. SAP pitches its Merchandising Assistant and Catalog Optimization Agent for continuous content review. Whatever tool does the work, the underlying claim suggests catalog quality needs to be watched like an ongoing metric. A one-time cleanup won't hold.

SAP also names the signals it would watch. Search patterns reveal unmet demand, taxonomy analysis shows where products drop out of discovery, and content quality scoring flags descriptions, attributes and translations that are letting products down. That gives a catalog team three concrete places to look before buying anything, since each signal maps to data most commerce teams already hold.

Marcus Holm, chief revenue officer at Exa, which sells web search built for AI agents and recently passed $100 million in annual recurring revenue, said on the GTMnow podcast that, per Cloudflare, more than half of web traffic is now agents and bots. That's one vendor's reading of another company's data. It still points the same way as SAP's survey.

Holm added that agent query volume has passed human volume, and that agents want a clean experience with no SEO and no ads. If that holds, the persuasion layers marketers have spent years buying would be precisely what an agent filters out, which leaves the structured product data underneath doing the work.

## Why readiness runs through the ERP

Marketing teams aren't short on AI. MarTech reports that, according to SAP's findings, about three-quarters of marketing teams are already putting AI to work on content, campaigns and lead generation. Pastore contends that gains on individual tasks won't fix a journey that is broken. For marketing ops, he prescribes integration: link the tools customers interact with to inventory, pricing, billing and ERP systems on the back end, so AI agents never commit the business to promises it can't fulfill.

SAP's report found only 6% of US organizations have all systems connected across the end-to-end journey, which suggests many AI agents could be working from incomplete price and availability data.

Handoffs are where the gap shows up for buyers. Nearly 80% of CX decision-makers in the survey said customers have to repeat information across channels and run into stalled handoffs. When a buyer moves from an outside AI assistant to a brand's website or into a sales workflow, the intent and context they built up often get lost on the way. For marketing ops, Pastore frames this as a mapping job: chart each transition and make the data carry across.

Block Field

The same post lists what "connected" means in practice: real-time inventory signals informing commercial decisions, one customer identity across service, commerce and fulfillment, and operational data reaching AI at the speed of customer behavior instead of in a nightly batch. That last item is the easiest one to test. For companies whose price or stock feeds still update overnight, an agent asked at 3 p.m. could be quoting the morning's numbers.

SAP argues the advantage compounds. A competitor with connected systems, Keehn writes, doesn't just execute faster; its AI improves faster because it learns from complete data rather than partial signals, so the distance between connected and disconnected brands grows in both halves of the journey. That is the seller's framing of its own category, but the logic is checkable: a model trained on overnight snapshots can only be as current as the snapshot.

## Buyers asked for accuracy, budgets went elsewhere

The survey also shows CX leaders and customers pulling in different directions. CX leaders' top AI priority is personalized recommendations. Consumers put accurate pricing and availability first, followed by discounts applied correctly and problems resolved.

Autonomy has limits too. Even if AI could complete purchases flawlessly, 87% of US consumers and 83% of German consumers want final approval before buying. Pastore's conclusion is to put AI effort into smooth research, comparison and assisted preparation, and to stop pushing for fully autonomous checkout.

Paul Meinshausen, head of agentic at MoEngage and co-founder of Aampe, offers a useful frame for why that split makes sense. In a Conversations with MarTech interview, he distinguished shopping from buying and cited marketing and psychology research showing people spend 90% of their time in the shopping process: searching, evaluating and forming impressions before checkout. Read alongside SAP's approval numbers, that suggests the bulk of the opportunity for AI sits in the long shopping phase, while the short buying moment stays with the customer.

Meinshausen also explained why accuracy gets the attention. Satisfaction with early agent deployments still trails expectations, he said, and the moments people remember are failures: a wrong notification reaching the CEO, an ad that misfires on social media. That is why so many marketing conversations have turned to governance and guardrails for agents. A wrong price quoted by an AI assistant is that kind of moment.

The integration argument has a reasonable counterpoint. Writing in MarTech, Shiv Gupta, principal at Quantum Sight, warns that AI recommendations are neither fully objective nor safe from manipulation. What an agent suggests comes down to how good the available information is, how the agent was built and what the buyer cares about most. If a company's data is clean and connected, an agent is more likely to describe its offer correctly. That still doesn't guarantee the company gets picked.

## The returns desk joins the acquisition funnel

AI support is currently at its weakest after the sale. MarTech's account of the report says AI has trouble with post-purchase tasks today, and only about 25% of shoppers rate it as helpful when handling returns or exchanges. Pastore anticipates that agents will weigh a brand's operational track record more and more when making recommendations: fulfillment accuracy, inventory reliability and service resolution.

Gupta takes that to its commercial end. An agent recommending an insurer, airline or telecom provider may weigh complaints, service responsiveness, billing practices and fulfillment reliability alongside price and features. A brand with a standout ad campaign and a weak service record could rank below a less familiar competitor that delivers more consistently.

He describes the result as a feedback loop: better experiences produce stronger evidence of performance, which can raise an agent's confidence in recommending the brand and bring in new customers. In that model, the service a company gives existing customers helps decide whether new ones ever consider it, and marketing's job grows to include arguing for operational fixes. Gupta says marketing doesn't need to own operations. It needs a louder voice in operational decisions.

Keehn lands in the same place from the vendor side. In an AI-mediated market, she writes, brand shifts from persuasion toward proof, reliability, service quality and outcome certainty, and checkout, fulfillment, returns and exchanges become the moments where a brand can differentiate. A software seller and an independent consultant arriving at the same conclusion doesn't prove it, but it does show the argument isn't confined to one sales pitch.

The departments affected first are the ones that rarely appear in a campaign brief: fulfillment, billing and customer service. For CMOs at companies where those teams report elsewhere, the SAP data suggests their numbers belong in marketing reviews as inputs that shape who gets recommended.

## Questions for the next stack review

The reporting supports three practical questions a marketing ops lead can take to IT and operations now. Where does an outside AI agent get our price, availability and spec data, and how old is that data when it arrives? When a buyer moves from an AI assistant to our site or a rep, what context comes with them? And which service measures, such as fulfillment accuracy and return resolution, would an agent find about us, and who is responsible for improving them?

Gupta adds a sorting step that keeps the answers useful: separate the problems better communication can address from the ones that need a real change in the customer experience. A thin product description is a content fix. A slow claims process is not, and no campaign will rewrite the record an agent reads.

Block Field

SAP dated this edition 2027, and the number to watch in the next one is the 6%. If confidence stays in the 90s while the connected share barely moves, the gap SAP measured will have widened at exactly the point where buyers start their search.

## Sources

- [Marketing ops can turn AI agent readiness from a blind spot to a strength](https://martech.org/marketing-ops-can-turn-ai-agent-readiness-from-a-blind-spot-to-a-strength/) (MarTech)
- [The Customer Journey Is Fractured, and That's Progress | SAP News Center](https://news.sap.com/2026/11/hybrid-ai-buyer-customer-experience-research-data-journey-progress/)
- [AI agents make CX marketing's problem](https://martech.org/ai-agents-make-cx-marketings-problem/) (MarTech)
- [How AI Agents Help Improve Your Customer Experience](https://www.youtube.com/watch?v=eaj6ta0VLGs)
- [Inside Exa's Run to $100M ARR: Building Google for AI Agents](https://gtmnow.com/gtm-211-search-for-ai-agents-exa-marcus-holm/)

Tags: SAP, AI in Customer Experience Report 2027, AI shopping agents, Marketing operations, ERP integration, Answer engine optimization

---
Source: MarketScale, https://www.marketscale.com/industries/marketing-tech/sap-finds-just-6-of-us-companies-have-systems-connected-for-ai-buyers. Published for AI indexing and citation; cite the canonical URL. Site guide for agents: https://www.marketscale.com/llms.txt
