This is the second in our four-part blog series sharing insights and practical lessons from the Job Boards Connect Annual Conference 2026.
Where is AI already making a useful difference inside a job board business?
At the Job Boards Connect Annual Conference in London on 8 October 2026, Paul Farnsworth and Luiza Avetisyan shared practical answers – from qualifying sales leads and accelerating development to improving candidate search and supporting customer teams.
Hosted by Alex Chukovski and sponsored by Madgex, AI: What’s Delivering Results Today? explored the work behind those applications, the challenges encountered and the lessons other operators can apply.
Both speakers were candid about the gap between an impressive demonstration and a reliable business process. Their case studies showed useful results alongside the need for testing, clear boundaries and human judgement.
Paul Farnsworth: Making AI part of the working day
Paul Farnsworth, President of Dice, began by separating AI into three areas: AI within products, AI as a tool for employees, and agentic AI that can carry out tasks through a sequence of actions.
Dice is using AI across all three. In its products, that includes search and matching support, plus tools that help recruiters build more effective Boolean searches. Internally, Paul described a substantial change in software development, with engineers increasingly managing agents and delivery processes.
The main case study focused on a specific sales task: qualifying inbound leads.
Before a lead reaches a salesperson, someone needs to establish whether the business is genuine, whether the contact is associated with it and whether the opportunity already exists in the CRM. Dice built an agent called Chase to handle that research.
Chase checks external information, consults Salesforce and records its findings. It also posts activity into Microsoft Teams, where employees can ask questions about leads and the work completed.
That accessibility was a deliberate design choice. Paul wanted the agent to feel like another team member, available through tools people already use, rather than requiring a separate portal.
By the conference, Chase had qualified hundreds of leads. Paul said it significantly reduced the research work required from the sales team, while acknowledging that the data could still be inconsistent.
Useful automation still needs boundaries
Dice’s experience also demonstrated why giving an agent access to business systems requires care.
Chase has credentials and access controls, with rules governing what it can retrieve and answer. Its activity is recorded so the team can review quality and understand what it has done.
It was not contacting customers directly. People still checked the quality of the leads before acting on them.
“You still need humans in the loop,” Paul said.
He also shared a revealing failure: when asked about the previous week’s leads, Chase returned no results because it was working with an incorrect understanding of the date.
The lesson was practical. An agent can handle complex research and still stumble over something basic. Reliability needs to be tested across the whole workflow, including the assumptions behind its answers.
For Dice, the value lies in reducing work while building a framework the business can understand, manage and extend.
Connecting job data to the tools candidates already use
Paul also explained Dice’s decision to make its data accessible through external AI tools.
Using Model Context Protocol, or MCP, connections, candidates can search Dice jobs from an AI assistant such as Claude. Rather than trying to build every part of the conversational experience itself, Dice is exploring how to participate in the environments its audience already uses.
Paul reported around 7,000 users and approximately 315,000 interactions with the connection at the time of the conference.
Access for recruiters brings additional considerations around authentication and candidate information. Paul described returning links to Dice profile pages rather than passing candidate details into the language model.
That approach raised a wider question for operators: which parts of the experience should a job board build, and where should it connect its data and services to other platforms?
Luiza Avetisyan: AI across a recruitment ecosystem
Luiza Avetisyan, Co-Founder and CEO of staff.am, described how the Armenian business has expanded from a job marketplace into an ecosystem covering recruitment, skills development, market intelligence and media.
“Job boards solve access, ecosystems solve hiring,” she said.
That wider model provides several places to apply AI. Luiza separated them into customer-facing features and internal workflows.
For candidates, staff.am had just launched a paid AI assistant offering salary information, career guidance and help with applications. Early registrations and purchases were encouraging, although the product was only a day into launch.
One important boundary was built into its application support: it could clarify and improve how a candidate presented their experience, but should not invent qualifications or make the application untrue.
The business had also added candidate signals to its search, including interests, followed companies and skills gained over time. The aim was to improve relevance beyond what a job title or CV alone could reveal.
Faster content and development with review retained
Inside the business, Luiza shared several examples of AI reducing manual work.
For staff.am’s media division, an AI workflow prepares articles from supplied information, including headings and formatting. An agent also scans company websites, news sources and social media to suggest relevant stories.
The team reviews the content before publication.
In engineering, requests submitted through Slack feed into an AI-assisted development process, with code review and escalation to the engineering manager where needed. Luiza reported a 58% reduction in turnaround time while retaining the review process.
Customer teams use an agent to identify potential cross-selling and upselling opportunities across the ecosystem. For example, repeated advertising for the same role might prompt a conversation about salary competitiveness or employer branding.
Those patterns provide a starting point for a customer conversation; they do not establish the cause of a hiring problem on their own.
AI also supports checks across billing and operational systems, helping the team identify inconsistencies.
What did not work – and why that matters
Luiza was equally open about applications that had fallen short.
Call sentiment analysis sometimes labelled a customer conversation as negative when a human listening to it found no problem. The team identified challenges with Armenian-language interpretation and began exploring a local-language model.
It was a clear example of why an apparently useful score needs checking against the underlying conversation, especially before drawing conclusions about an employee’s performance.
AI-generated designs also struggled to follow the strict templates required for regular social posts and adverts. After testing different tools, the business retained its designer for that work.
Adoption brought another lesson. An earlier AI matching and analysis feature attracted less use than expected. Conversations with clients revealed that releasing the feature was only part of the job.
“Launching a feature is not sufficient,” Luiza said. “You need to do some education, especially when it comes to AI.”
By contrast, conversational job search had been well received. Luiza reported that only around 15% of users switched from the default AI search to the standard search option.
Choosing the next useful application
Across both case studies, the clearest opportunities came from identifiable tasks: researching a lead, preparing an article, developing a feature or helping a candidate find relevant work.
The speakers also showed that different applications need different measures of success. Time saved, better relevance, user adoption and reliable outputs each tell part of the story.
For job board leaders, a useful next step is to choose a workflow with a clear problem, define what improvement would look like and test the output with the people who will use it. Keep review where it matters, investigate failures and allow the results to guide the next investment.
Also, check out Alex Chukovski’s LI post about the conference