- Is it really worth evaluating OpenAI Presence for my company?
- What is OpenAI Presence and how does it really work?
- What has OpenAI Presence proven in practice?
- Why everyone's talking about it now: is the Italian market really ready?
- What concrete scenarios open up for Italian SMEs?
- How does Presence compare with the alternatives already available?
- What rules and risks do I need to handle before getting started?
- How do you get started, step by step?
- What mistakes are most common when introducing an AI agent in customer service?
- How do I know if the AI agent is delivering concrete results?
- So, is it worth it for your company?
- How we see it at SHM
- Frequently Asked Questions
- OpenAI Presence is an enterprise platform for voice and chat AI agents, not a simple chatbot: it carries out sequences of actions, accesses external systems, and handles conversations with contextual memory.
- In Italy, the AI market reached 1.8 billion euros in 2025 (+50% vs. 2024) and adoption among businesses rose to 16.4%: the time is right to evaluate solutions like Presence.
- Customer service is the most immediate testing ground: according to Salesforce, by 2027, 50% of support cases in Italy will be handled by AI, up from 25% today.
- Before adopting Presence, it pays to map high-volume workflows, check compatibility with your existing CRM and ERP, and plan a privacy impact assessment (DPIA) for the European market.
Updated on October 25, 2026
OpenAI Presence resolves 75% of support requests without involving a human agent. For Italian companies running a contact center, this is not an experiment: it's a benchmark to compare against your current costs.
Is it really worth evaluating OpenAI Presence for my company?
It depends on the volume of interactions you handle and the maturity of your systems. If you have an active contact center or a steady flow of repetitive requests, Presence deserves a concrete analysis.
| Your situation | Worth evaluating? | why |
|---|---|---|
| Contact center with >500 requests/month | Yes | Enough volume to amortize the integration and measure ROI |
| E-commerce with seasonal peaks | Yes | Agents scale up without hiring temporary staff |
| B2B with qualified inbound leads | Yes | The agent qualifies and schedules, the sales team closes |
| SMEs with <50 requests/month | No, for now | Setup costs outweigh the benefit in the short term |
| Company without a structured CRM | No, for now | Integration requires clean data: fix the foundation first |
What is OpenAI Presence and how does it really work?
Presence is an enterprise platform launched by OpenAI on July 22, 2026 to deploy voice and text AI agents in business processes. It's not a chatbot: it's a system that takes actions, not just gives answers.
A traditional chatbot answers predefined questions. An AI agent, on the other hand, carries out sequences of actions: it accesses the CRM, updates a record, sends a notification to the sales team, all on its own and in a single conversation.
OpenAI positions Presence as a trusted solution, meaning one with governance, an action log, and output controls. This directly addresses the main objection from IT and legal leads: "How do I know what the agent did?"
The platform covers both voice and text (chat) channels. That makes it a good fit for anyone running a traditional contact center who wants to add AI without tearing down existing infrastructure. To dig deeper into AI agents in production, we've analyzed in detail how OpenAI Presence is positioned for enterprise production.
- 1 · CustomerStarts the conversationCalls or writes via chat. The agent replies in natural language, by voice or text.
- 2 · Presence agentUnderstands the intentAnalyzes the request and retrieves context from the CRM or the company knowledge base.
- 3 · Presence agentTakes actionUpdates data, sends notifications, schedules appointments, processes refunds, all without human intervention.
- 4 · Human agentSteps in only when neededThe agent hands off the conversation when the case goes beyond its competence or at the customer's request.
What has OpenAI Presence proven in practice?
OpenAI has already used Presence on its own English-language phone support line, with measurable, public results.
The 75% autonomous resolution figure is significant because it comes from OpenAI's own support line, not from a controlled pilot. The 15-percentage-point improvement in just 10 days, thanks to an optimization loop based on Codex (the AI system that analyzes and improves the agent's code), shows that these systems learn quickly from their own mistakes.
In Italy, Salesforce estimates that by 2027, 50% of customer service cases will be handled by AI, up from 25% today. The same report points to a potential 15% increase in upselling revenue thanks to AI agents, with peaks of 20% in life sciences and biotech.
Why everyone's talking about it now: is the Italian market really ready?
The Italian AI market reached €1.8 billion in 2025, with business adoption tripling in two years. The landscape has changed.
According to the Osservatorio Artificial Intelligence of Politecnico di Milano, the Italian AI market reached €1.8 billion in 2025, growing 50% compared to 2024. 84% of large Italian companies have already purchased generative AI licenses, and job postings requiring AI skills grew 93% in one year.
On business adoption, ISTAT data on businesses and ICT 2025 show that 16.4% of companies with at least 10 employees already use AI, double the 2024 figure. Among large companies (250+ employees) the share rose to 53.1%, while among SMEs it stands at 15.7%.
This means the gap between large companies and SMEs is widening. If you wait, you risk starting behind your larger competitors. AI agents are at the heart of our deep dive on AI and process automation.
What concrete scenarios open up for Italian SMEs?
Three types of companies can benefit right away from a platform like Presence: e-commerce, B2B businesses with consultative sales, and companies with high-volume internal HR or IT processes.
The scenarios below are illustrative: the numbers show the reasoning and are not real client data.
| Company type | Automatable workflow | Expected benefit | Setup complexity |
|---|---|---|---|
| E-commerce (e.g. 2,000 orders/month) | Order status, returns, product availability | −60% tickets to the support team during peaks | Low (Shopify/Magento APIs already available) |
| B2B professional services | Lead qualification, sales FAQs, scheduling | Sales team freed up for high-value deals | Medium (requires CRM integration) |
| Manufacturing company (HR/IT) | Employee onboarding, IT helpdesk, knowledge base | −40% time spent on repetitive internal tasks | Medium (depends on existing data structure) |
| Store network (retail) | Bookings, local promotions, after-sales support | 24/7 coverage with no additional staffing costs | Medium-high (multichannel voice + chat) |
If you run lead generation campaigns, adding a conversational agent can amplify your results. A Presence agent can qualify leads generated by LinkedIn campaigns or Google Ads campaigns in real time, shortening the time between first contact and sales qualification.
Coordinating multiple AI agents, which is useful when workflows are complex, is covered in our article on multi-agent AI orchestration with up to 1,000 agents in parallel.
How does Presence compare with the alternatives already available?
There are already enterprise solutions on the market with defined pricing for Italy. Presence stands out for the quality of its underlying language model, but it is still less transparent on European pricing.
| platform | Strengths | Current limitations | Europe pricing |
|---|---|---|---|
| OpenAI Presence | Top-tier language model, voice + chat, built-in governance | European pricing not yet detailed, partial technical documentation | To be determined |
| Microsoft Copilot Studio | Native Microsoft 365 and Azure integration, clear pricing | Less flexible outside the Microsoft ecosystem | Available |
| Salesforce Einstein | Native CRM, customer data already integrated, enterprise track record | High cost, requires the Salesforce ecosystem | Available |
| Open source solutions (e.g. Rasa) | Maximum flexibility, no vendor lock-in | Requires an in-house technical team, long development times | No license cost |
What rules and risks do I need to handle before getting started?
The GDPR sets precise requirements for handling conversational data. Any deployment in Europe requires a data protection impact assessment before going into production.
The data protection impact assessment, known as a DPIA, is a mandatory document when processing personal data at scale with new technologies. An AI agent that handles customer conversations falls into this category.
The main risk isn't the fine itself, but having the project blocked after deployment has begun. It's better to tackle the DPIA at the design stage, not afterward. The governance of AI agents, meaning what happens when they act unexpectedly, is analyzed in our article on OpenAI and AI agents out of control.
A second risk to consider is vendor lock-in: dependence on a single provider. If conversation data and training models belong to the provider, switching platforms later becomes expensive. Plan a data portability strategy from the start, regardless of the provider you choose. To learn more about the risks tied to AI APIs, read our article on OpenAI agents and the RubyGems cyberattack.
| Risk | Probability | How to mitigate |
|---|---|---|
| GDPR non-compliance | High if not planned for | DPIA before deployment, DPO involved from the start |
| Vendor lock-in | Medium | Contract with data portability clauses, modular architecture |
| Failed integration with legacy systems | Medium | Proof of concept on a single workflow before scaling |
| Insufficient response quality | Low with fine-tuning | Continuous feedback loop, human oversight in the first weeks |
| European pricing not defined | Current | Request a direct quote from OpenAI, compare with alternatives |
How do you get started, step by step?
The safest path is incremental: one workflow at a time, with clear metrics from day one.
- 1 · Management + OperationsMap your high-volume workflowsIdentify processes with more than 200 interactions/month, repeatable logic and structured data. These are your best candidates.
- 2 · IT + CRM ManagerCheck system compatibilityMake sure your CRM, ERP and e-commerce platforms have open APIs. Without integration, the agent can't act.
- 3 · DPO or legal advisorStart the DPIAData protection impact assessment: required by the GDPR. Best done before signing contracts.
- 4 · Technical team + AI partnerProof of concept on one workflowLaunch a pilot on a single process. Measure autonomous resolution rate, customer satisfaction and average handling time.
- 5 · ManagementDecide whether to scaleWith the pilot data in hand, the decision to expand or stop is based on real numbers, not expectations.
What mistakes are most common when introducing an AI agent in customer service?
The most common mistakes aren't technical: they're about governance, expectations and integration with existing processes.
Starting without clean data. An AI agent is only as good as the data it accesses. If your CRM has duplicate records, empty fields or inconsistent categories, the agent will give inaccurate answers. Before any deployment, you need a data audit.
Automating everything at once. The temptation is to replace the entire contact center in one go. The result is almost always a poor customer experience and a project that has to be redone. One workflow at a time is the right way.
Not involving the support team. The agents who handle requests today know the edge cases, the ones outside the norm, better than anyone. Their input during the agent's training phase is valuable and often overlooked.
Forgetting the DPIA. As we said, it's not a bureaucratic formality: it's your guarantee that the project won't be blocked after launch.
Not measuring anything. Without metrics defined before the pilot, you can't know whether the agent is working. Autonomous resolution rate, Customer Satisfaction Score and average handling time are the three minimum indicators to track from day one. We cover the topic of AI agents in marketing in our article on OpenAI AI agents for everyone: what changes in marketing.
How do I know if the AI agent is delivering concrete results?
Three indicators are enough to tell whether the project is heading in the right direction. If all three improve, you scale. If one gets worse, you fix it before expanding.
| Indicator | What it measures | Minimum success threshold | How to measure it |
|---|---|---|---|
| Autonomous resolution rate | % of requests closed without a human agent | >60% after 30 days | Agent platform logs |
| Customer Satisfaction Score (CSAT) | Customer satisfaction after the interaction | Equal to or higher than the human channel | Post-interaction survey |
| Average handling time | Minutes to close a request | −30% compared with the human channel | CRM or ticketing platform |
| Escalation rate | % of requests transferred to a human agent | <25% after the first month | Agent platform logs |
| Upselling revenue | Additional sales generated during the interaction | +10-15% compared with the channel without AI | CRM + sales data |
So, is it worth it for your company?
It's worth it if you handle a high volume of repetitive requests (customer service, lead qualification, HR helpdesk), have an integrated CRM and are willing to invest in a DPIA before launch. The potential to cut operating costs and improve the customer experience is real and measurable.
Somewhere in between: you're in a growth phase and your current volume doesn't yet justify a full deployment. In that case, a proof of concept on a single workflow, with a limited budget and clear metrics, is the right move to gather real data without taking on excessive risk.
Not yet if you have fewer than 200 interactions a month, your data is disorganized or you don't have a structured CRM yet. Before evaluating Presence, it's best to fix the foundation: clean data, documented processes, integrated systems.
How we see it at SHM
For us, AI makes sense when it gives people their time back, not when it replaces them without judgment. OpenAI Presence is a powerful tool, but the value isn't in the technology itself: it's in the workflow it automates and the time it frees up for work that truly requires human judgment.
In the companies we work with, we often see the same pattern: they start out excited about the novelty, underestimate the quality of their existing data and overestimate how fast their teams will adopt it. The result is a pilot that never takes off. Our experience tells us the critical factor isn't the choice of platform, but clarity on the workflow to automate and clean data upstream.
If you're considering how to integrate AI agents into your company, the starting point is a process analysis, not a product demo. Our strategic AI integration service is designed for exactly this: understanding where AI creates real value in your specific context, not in the abstract. If you also want to understand how to get found by AI search tools, our SEO for AI service completes the picture. And if your goal is to build a custom conversational assistant, you can explore our chatbot and virtual assistant development service.
The future of AI agents, including scenarios like those described in our article on invisible AI agents without interfaces, is built today, with well-structured pilot projects and measurable goals.
Frequently Asked Questions
Presence was officially announced on July 22, 2026 as an enterprise platform. Availability in the European market and specific pricing have not yet been detailed by OpenAI. It's best to contact OpenAI directly or a certified partner for up-to-date information on availability in Italy.
A traditional chatbot answers predefined questions by following a fixed decision tree. An AI agent carries out sequences of actions on its own: it accesses external systems like the CRM, updates data, sends notifications and handles conversations with contextual memory. In practice, the difference is that the agent can complete a process, not just answer a question.
Yes, almost certainly. The GDPR requires a data protection impact assessment (DPIA) when processing personal data at scale with new or high-risk technologies. An AI agent that handles customer conversations falls into this category. Involve your DPO or a legal advisor before signing any contract.
A proof of concept on a single, simple workflow, such as order status management in an e-commerce store, can be launched in 4-8 weeks, provided your data is already structured and your CRM has open APIs. More complex workflows or integrations with legacy systems take 3-6 months.
The three main indicators are: autonomous resolution rate (how many requests the agent closes without human intervention), Customer Satisfaction Score (customer satisfaction after the interaction) and average handling time. On top of these comes the potential increase in upselling revenue, which Salesforce estimates at around 15% on average for industries adopting agentic AI in customer service.
According to OpenAI's public documentation, Presence supports API integrations with external systems. Specific compatibility with platforms like Shopify, Salesforce, or other CRMs depends on the setup and requires a case-by-case technical assessment. It's one of the points to verify during the proof of concept phase.
An AI agent isn't a cost saving: it's a choice about where you want the people in your company to spend their time.
Related articles
Discover more articles exploring similar topics, selected to offer you a more complete and stimulating perspective. Each piece of content is carefully chosen to enrich your experience.
