> ## Content Index > Fetch the complete content index at: https://www.frontierprofile.com/llms.txt > Use this file to discover other available public pages before exploring further. # Ines Salhi, Chief Product Officer, Payfit - URL: https://www.frontierprofile.com/ines-salhi-chief-product-officer-payfit/ - Published: 2026-10-06T18:59:23.000Z - Updated: 2026-10-06T18:59:23.000Z - Description: Ines Salhi discusses how Payfit combines AI agents with human expertise to transform payroll and HR administration for European businesses. - Author: Ewan MacLeod - Tags: Payroll, HR Technology, AI Agents, Enterprise AI, SaaS [Ines Salhi](https://www.linkedin.com/in/ines-salhi-b330845a/?ref=frontierprofile.com) is Chief Product Officer at [Payfit](https://payfit.com/?ref=frontierprofile.com), a payroll and HR service that combines platform technology, human expertise and AI agents to help businesses across France, Spain and the UK manage payroll without losing visibility or control. Trained as an engineer at CentraleSupélec in Paris, Ines previously founded a transport route optimisation company before joining Payfit in 2018\. In this interview, she explains how Payfit AI interprets customer information from spreadsheets and photos, prepares variable payroll entries and answers HR questions — all while working within strict access rights and without using customer data for training. My questions are in bold - over to you Ines: --- **Who are you and what is your background?** I'm Ines Salhi, Chief Product Officer at Payfit. I was born in Paris and trained as an engineer at CentraleSupélec in Paris. Before joining Payfit, I founded a software company focused on transport route optimisation. That experience taught me how much distance there can be between the problem you think a customer has and the one they actually experience. It led me towards product work, where understanding the real problem is essential. I joined Payfit in 2018 and have seen its product evolve from several perspectives, first building the technology directly and now leading the teams shaping what comes next. AI initially entered my work through colleagues and conversations at home. My husband was deeply engaged with it and, once I started exploring it myself, its relevance to many parts of my work became impossible to ignore. **What is your job title and what are your general responsibilities?** I'm Chief Product Officer at Payfit. I lead product management and product design, alongside teams working on product compliance and the implementation of payroll rules in our technology. My responsibility is to shape the product roadmap around what customers need and where we believe we can exceed their expectations. That means staying close to their experience while deciding what we should build next. As Payfit becomes a payroll and HR service supported by AI and dedicated experts, my focus is on making that service exceptional in everyday use. **Can you give us an overview of your business?** Payfit is a payroll and HR service designed to take work off a company's shoulders while keeping it informed and in control. Customers provide the information needed for payroll, then choose which steps they want to handle themselves and which they want Payfit AI or their dedicated team to take care of. That can include entering payroll variables, checking information and drawing attention to payslips that need a closer look. The service combines a payroll platform, human expertise and an AI agent that increasingly helps with operational work. It also includes a broader platform for managing employees throughout their time with a company. We are expanding that vision into areas such as IT equipment and employee benefits. Payfit serves businesses in France, Spain and the UK. Its public figures report more than 22,000 customers, over 250,000 payslips managed each month, more than 700 employees and €433 million raised. **What problem do you solve with AI?** We want payroll to fit the way each customer works, rather than requiring every customer to adapt to a rigid process. Payroll has to be accurate and completed every month, but people should not have to spend hours preparing information, moving between systems and resolving small questions. AI helps us make the service much more adaptable. A customer might have bonus or leave information in a spreadsheet, another file or even a photo. Payfit AI can interpret what it receives, prepare the information in the right place, ask questions when something is unclear and request approval before an action is completed. Our technology and dedicated experts remain part of the same experience. The aim is for customers to get the outcome they need, with the level of involvement that suits them. **What does the AI actually do, and how is it built?** One concrete example is the preparation of variable payroll information. A customer can upload a file containing bonuses, leave or overtime. Payfit AI reads the information, identifies anything that needs clarification and prepares the relevant entries in the payroll software. The customer can review and validate what happens. It can also help a customer understand information that would otherwise require a series of reports or manual checks. For example, an administrator can ask which employees have received the smallest pay increases over a given period. Payfit AI can bring together the relevant information so the administrator can examine the situation and decide what to do. It does not make that decision for them. We do not build our own foundation models. Our expertise lies in connecting models to payroll technology, customer context and the actions they need to perform. We use models from external providers and choose between them according to their performance and cost. Model access runs through AWS Bedrock. **Who are your target customers, and what is your revenue model?** We serve businesses with roughly 1 to 1,000 employees in France, Spain and the UK. They range from very small companies running their first payroll to larger organisations with more complex needs. Payfit operates on a recurring subscription model. The price depends on factors such as the number of employees and the level of service selected. Customers get the most value when they want payroll and HR work handled reliably without losing visibility or control. Payfit AI is included in the service rather than charged for separately. **What is the evidence it works, and what have you learned?** Payfit AI was adopted unusually quickly by our customers, even though we did relatively little to promote it at launch. Customers frequently mention it when they explain what they value about Payfit, and we see it being used across many different tasks. We have also observed a decline in requests to customer support that are primarily asking for information or clarification. That decline has continued as we have released more capable versions of Payfit AI. It is a useful signal, although I would be careful about attributing every change in support volume to one product. One lesson was that we initially underestimated the work involved in evaluating an agent at scale. A good demonstration is one thing; consistently high quality across many payroll and HR questions is another. We have increased the time and resources devoted to regular evaluation and proactive improvements. **How do you handle the data question?** Payfit AI works within the access rights of the person using it. If an administrator can see certain information in Payfit, the agent can use that information to help them. It cannot draw on another customer's data or access information that the user is not entitled to see. Its expertise also comes from documentation, including our customer help content and internal documentation of the processes followed by our experts. Customer data is not used to train the models. We use AWS Bedrock to access external models, with our setup designed around European data processing. The distinction is straightforward: an agent may use authorised customer information to complete a request, but that information does not become training material. **What is new at the company, and what is coming next?** Payfit AI is moving from answering questions towards taking on more of the operational work involved in preparing payroll. It can help process payroll information and highlight payslips that deserve a closer review because a change may need to be checked. That helps customers focus their attention where it matters. We are also broadening the service around the employee journey. This includes IT equipment and access, benefits such as health insurance, and better ways to manage people who are part of a company's workforce but do not receive payroll through Payfit, such as freelancers. The ambition is a more complete People Operating System in which payroll, HR and related services work together. **If you had a magic wand, what would you change about how AI gets bought and deployed?** I would make it much easier to connect AI securely to the systems a company already uses. Those connections take time and investment. Access has to be granted carefully, and the work usually has to happen step by step. There are good reasons for those safeguards. But the effort involved in giving an agent the right access is one of the main things slowing down useful deployment today. If that work became simpler while remaining secure, companies could reach meaningful results much faster. **What is your message for large enterprises evaluating AI?** Start by using it yourself. Leaders should spend substantial time testing the tools and speaking with peers who are further along. That gives them a clearer sense of what is already possible and where the real difficulties lie. Then give people across the company access to the tools. The first stage is adoption: more people need to understand what AI can do and use it in their daily work. At first, that means accepting the cost of access and learning before expecting major gains. The broader transformation becomes possible when that understanding spreads through the organisation. **Who should people follow in AI?** I tend to follow the work of teams rather than a fixed list of individual voices. I would recommend the research and product updates from [Anthropic](https://www.linkedin.com/company/anthropicresearch/?ref=frontierprofile.com), [OpenAI](https://www.linkedin.com/company/openai/?ref=frontierprofile.com) and [Mistral](https://www.linkedin.com/company/mistralai/?ref=frontierprofile.com). Each offers a useful view of how the technology is advancing and where it may be applied. I also learn a great deal through direct conversations with people at other French technology companies. Comparing what is working in practice, including what has proved difficult, is often as valuable as reading an announcement. **What AI tools do you personally use day to day?** I use [ChatGPT](https://chatgpt.com/?ref=frontierprofile.com) every day. It is the tool we chose for our product and technology teams at Payfit, and I use it to manage emails and tasks, draft and revise content, review websites, and find information across financial systems and product analytics. I also use [Dust](https://dust.tt/?ref=frontierprofile.com) for some agents that bring together information from internal documentation. These tools are most useful to me when they make information easier to find and help me move from a question to a useful action. **What is the best new AI product or service you have seen recently?** Google Search with AI Overviews has been especially striking to me. Getting a useful summary directly in the search results saves time on ordinary questions and makes the value of AI immediately tangible. It is a change people can experience without learning an entirely new tool. **What trends will define AI over the next year?** We will delegate a growing share of work to agents. Software development is already showing what this can look like, but the next step is connecting agents to more sophisticated processes that involve several systems and teams. That is where I expect substantial gains in both quality and productivity. The other major change is the customer experience. Agents will become more capable of taking action and anticipating what a customer needs. We already see the beginnings of this at Payfit, and I expect it to become much more visible over the next year. --- Many thanks to [**Ines**](https://www.linkedin.com/in/ines-salhi-b330845a/?ref=frontierprofile.com) for taking the time to share her insights with **Frontier Profile**. You can learn more about [**Payfit**](https://payfit.com/?ref=frontierprofile.com) on their website.