Plan · Enterprise AI agent examples
AI agents for sales: what they do and what companies report
A sales agent works the steps around the seller: it answers inbound leads within minutes, asks qualification questions, writes the answers to the CRM, researches accounts before meetings and drafts proposal groundwork. Sellers keep the relationship and the deal, pricing included, and the agent hands them qualified leads and prepared material.
In this guide: 83 pages
Key takeaways
- Siemens Smart Infrastructure has published a sales example with a full funnel: two Agentforce agents answer about 2,500 inbound leads a month across 132 countries, with response time cut from days to minutes and a 6% qualification rate, as Salesforce reports.
- Siemens publishes its funnel, an 11% engagement rate and a 6% qualification rate, which a sales leader can check against a baseline. Most other sales stories publish hours saved or a productivity percentage without a method.
- An agent that emails prospects needs consent rules, opt-outs and disclosure that the prospect is talking to an AI. An agent that writes to the CRM needs a list of the fields it may change and a log per lead.
- Adoption decides the result for seller-facing agents. Microsoft’s own Sales Agent reaches up to 25 systems of record, and its Inside Track team describes adoption only as "a decent level".
What do AI agents do in sales?
Two kinds of work. Prospect-facing agents answer inbound leads, follow up and qualify them, then route the good ones to a seller. Seller-facing agents research accounts, draft proposal groundwork and let sellers update systems from one chat.
Microsoft’s own deployment shows the seller-facing kind at scale: one chat entry routes each request to the agent behind the right system, so a seller can create a CRM opportunity or submit an investment request without leaving the conversation. It runs on Copilot Studio and uses MCP to reach up to 25 systems of record.
Sales tasks by what the agent is allowed to do. "Assists" means it drafts or researches and a seller acts; "acts" means it changes something or contacts a prospect itself.
| Task | Level | What it touches, and the control we add | Seen in |
|---|---|---|---|
| Answer inbound leads within minutes and follow up | Acts | Email to the prospect; consent and opt-out checked first | Siemens (engagement agent) |
| Qualify on budget, authority, need and timing | Acts | A conversation with the prospect; answers saved to Sales Cloud | Siemens (qualification agent) |
| Route qualified leads with a summary, close the rest politely | Acts | Lead status and owner in the CRM | Siemens |
| Draft proposal groundwork from internal knowledge | Assists | Product and past proposal content; a seller edits | Fujitsu |
| One chat front door to create opportunities and deal books | Acts | Up to 25 systems of record, through the underlying agents | Microsoft (Sales Agent) |
| Account research before meetings | Assists | CRM, email and public financials | Lumen (2024 source) |
| Find priority prospects and prepare briefs | Assists | CRM and lead data | The Adecco Group (Agentforce Coworker) |
The inbound lead is where agents act on their own most often, because the work is a fixed sequence and the alternative is a lead nobody answers for days. None of the published examples describes an agent setting prices, discounts or contract terms, and we keep those with the seller.
What have companies published about sales agents?
Four companies publish something specific, and only Siemens gives a funnel. The others give one productivity figure, a capability claim or no figure at all, and each comes from the vendor that sells the product.
Publicly reported sales agents, as published by each source and opened on 2 November 2026. None of these companies is a QueryNow client.
| Company | What the agent does | Published figure | Source and date | How to read it |
|---|---|---|---|---|
| Siemens Smart Infrastructure | An engagement agent emails the prospect within minutes of an action and follows up twice; a qualification agent asks qualifying questions, saves the answers to Sales Cloud and routes qualified leads with a summary | About 2,500 in-scope inbound leads a month answered; 132 countries covered; response from days to minutes; 11% engagement; 6% qualification; 80% of interactions rated Good or Very Good | Salesforce customer story, 14 September 2026; press release, 15 September 2026 | Vendor claim with funnel figures; Siemens’ head of sales effectiveness is quoted |
| Fujitsu | An orchestrator and specialist agents pull knowledge from internal sources, draft the groundwork of sales proposals and brief new hires | Proposal creation 67% more productive; no baseline published | Microsoft customer story, 1 March 2025 | Built on what is now Microsoft Foundry Agent Service; one figure without a method |
| Microsoft (own sellers) | Sales Agent: one chat front door that routes to underlying agents across up to 25 systems of record | No result figures; adoption described as "a decent level"; baseline: about 70% of seller time went to work away from customers | Microsoft Inside Track, 24 September 2026 | Vendor self-use on Copilot Studio with MCP |
| Lumen Technologies | Copilot for Sales researches accounts and summarizes context before meetings, with Salesforce and Outlook | Research in "15 minutes instead of four hours"; a revenue value the company calculated from four hours a week | Microsoft customer story, 21 May 2024 | Older source; a capability claim and a calculation |
Siemens is the example to model a business case on, because it publishes the steps of the funnel. The Adecco Group is rolling Agentforce Coworker out to about 27,000 employees in more than 40 countries after pilots in the UK and France, and has published no outcome figures for it yet.
What controls does a sales agent need?
A sales agent touches two things a company guards closely: its prospects’ inboxes and its CRM. Each control below protects one of the two, and the handoff protects the relationship.
- Consent and contact rules. The agent emails only people who acted, such as a download or a form, within the consent they gave, honours an opt-out at once and stops after the follow-ups you set. Siemens’ engagement agent follows up twice.
- Disclosure. Prospects are told they are talking to an AI and can reach a person. The EU AI Act’s Article 50 duty for AI that interacts with people has applied since 2 August 2026.
- CRM write scope. A qualification agent writes named fields on the lead record, such as the answers and the status, logs each write, and leaves opportunities, prices and forecasts to sellers. A seller-facing agent like Microsoft’s should act with the seller’s own rights.
- Handoff. A qualified lead reaches a named seller with a summary, and the seller can read what the agent said. Siemens routes qualified leads with a summary and closes the rest politely.
- Quality review. Sample conversations every week for tone, wrong product claims and missed opt-outs, and track engagement and qualification rates by segment.
Two failures are common in lead agents. The agent qualifies on the questions it was given and passes leads a seller would reject, which shows up as falling seller acceptance of routed leads. And the follow-up sequence keeps running after the prospect replied somewhere else, which reads as spam. Track seller acceptance, and stop the sequence on any reply in any channel.
Where should you start with a sales agent?
Start with inbound leads nobody answers within a day, because the baseline is easy to measure and a fast reply has visible value. Outbound prospecting by an agent carries more consent and reputation risk, so leave it for later.
- Measure today’s lead response time and the share of inbound leads never contacted. Those two numbers are the baseline.
- Write the qualification questions and the routing rules with sales leadership, including what makes a lead not worth a seller’s time.
- Launch in one region or product line with disclosure, opt-out and a follow-up limit. Read every conversation for the first two weeks.
- Compare engagement, qualification and won deals from agent-handled leads with the baseline over a full sales cycle.
- Add seller-side help, such as account research or proposal groundwork, once the inbound flow holds.
How do Microsoft, AWS and Google implement it?
Where the published sales agents ran, and what each cloud offers for the next one, with statuses as the vendor pages showed on 2 November 2026.
Capability, Microsoft, AWS, Google
| Capability | Microsoft | AWS | |
|---|---|---|---|
| Where the examples ran | Fujitsu on Foundry Agent Service; Microsoft’s Sales Agent on Copilot Studio with MCP; Lumen on Copilot for Sales | No sales example with figures in our research set | No sales example with figures in our research set |
| Build the agent | Copilot Studio (low code) or Microsoft Foundry Agent Service (pro code) | Amazon Bedrock AgentCore Runtime and Gateway (GA 13 October 2025) | Gemini Enterprise Agent Platform with ADK and Agent Runtime |
| Connect the CRM | Power Platform connectors, or an MCP server for the CRM | AgentCore Gateway, with a one-click integration named for Salesforce | Gemini Enterprise connectors; Agent Registry holds MCP servers and tools |
| Identity for the agent | Microsoft Entra Agent ID, created for every new Copilot Studio agent since July 2026 | AgentCore Identity with on-behalf-of token exchange | Agent Identity (SPIFFE-based); API in preview since 18 June 2026 |
Salesforce Agentforce ran the Siemens and Adecco examples inside Salesforce, where the lead data already lives.
Frequently asked questions
Can an AI agent qualify sales leads on its own?
Siemens runs two agents that do. One engages each inbound lead within minutes; the other asks budget, authority, need and timing questions, saves the answers to Sales Cloud and routes qualified leads to a seller. Salesforce reports a 6% qualification rate on about 2,500 leads a month.
Should an AI agent do outbound prospecting?
Carefully, if at all at first. Outbound email to people who never acted raises consent questions under GDPR and similar laws, and a wrong claim from an agent reaches a prospect directly. Start with inbound leads and seller-side research, where the published results are.
How do you measure a sales agent?
Use the funnel: response time, engagement rate, qualification rate and, over a full sales cycle, pipeline and won deals from agent-handled leads against a baseline. Hours saved for sellers matter less than whether more leads became revenue.
What does a sales agent need from the CRM before launch?
Clean lead sources, a defined lead status model and an owner for each region or product, because the agent routes by those fields. If half the leads have no source or the territories are out of date, fix that first, or the agent will route quickly to the wrong seller.
Does a sales agent replace sales development reps?
In the published examples it takes the leads nobody had time for. Siemens describes full coverage of in-scope leads across 132 countries, and qualified leads still go to a seller, so the people work on fewer, better leads.
More in Enterprise AI agent examples
Sources
Each source was read on 2 November 2026. The date shown is the date on the page, or the page’s ms.date for Microsoft Learn; an undated page shows the date it was read.
- Siemens takes lead response from days to minutes with Agentforce
- Siemens and Salesforce deepen AI partnership to redefine industrial sales and service
- Fujitsu is revolutionizing sales efficiency with Azure AI Agent Service
- Driving seller adoption of Sales Agent through role-based change management at Microsoft
- Lumen’s strategic leap: how Copilot is redefining productivity and employee engagement
- Adecco Group announces Salesforce’s Agentforce Coworker rollout across 40 countries
- AI Act: regulatory framework for AI
- Amazon Bedrock AgentCore Gateway
- Amazon Bedrock AgentCore is now generally available
- What’s new in Copilot Studio
- Gemini Enterprise Agent Platform release notes
- On-behalf-of token exchange (AgentCore Identity)
- Agent Identity overview
- What is Microsoft Foundry Agent Service?
- Use connectors in Copilot Studio
- Introduction to connectors and data stores
Planning an AI program?
Talk to the team that builds it and runs it after launch.
- Built to SOC 2, HIPAA and GDPR standards
- EU AI Act-aligned delivery
- Microsoft Solutions Partner