Building AI Agents for Smart Prospecting

Goava Discover is a B2B sales intelligence platform that helps Nordic companies identify and prioritise high-fit accounts. Sales development reps (SDRs) use it to find, score, and route prospects into their CRM.

Then competitors started shipping AI automation. Goava had to either move the platform toward AI-native workflows or watch customers evaluate tools that already had.

The answer was a new layer on top of Discover: the Prospecting Agent, an agent that automates account discovery, with a signal-based trigger mode that surfaces real-time buying signals inside the same system.

I looked at three things going in: where users were already spending time (and dropping off) in prospecting flows, how SDRs actually decided whether an account was worth pursuing, and what AI-native competitors were offering that Goava's customers were comparing against.

1. Research time went to dead ends: SDRs spent a large share of their time on research that didn't lead to qualified leads. Automating it would help retention and help SDRs find the right accounts at the right moment.
2. ICP isn't fixed: Companies didn't have one ideal customer profile. Different products, territories, and campaigns needed different targeting logic, so agents had to support multiple concurrent ICPs.
3. Signals were missed: The data for buying signals already existed inside the platform, but nothing surfaced it at the right moment.
4. Usage data revealed sensible defaults: Research showed that configuring an agent meant setting too many parameters. Analyzing how often users already prospected let us pre-fill agent settings with realistic defaults, which cut setup effort and cognitive load.

1. Users set up an agent from Discover's agent settings, choosing an existing ICP profile to score against. AI generates a few of these during onboarding, so most users start from one instead of building their own.
2. They select a segment, the pool of companies the agent evaluates. Onboarding creates an auto-segment, and tighter, more relevant segments produce better results.
3. The agent evaluates every company in that segment against the chosen ICP and sorts them by Goava Score.
4. Qualifying companies land on a positive outcome list (auto-exported to the CRM if one's connected); rejected companies go to a negative outcome list, keeping the positive list clean.
5. Users set an AI credit budget (a cap per prospect or per day/week/month), and the agent pauses automatically once it's spent.
6. Multiple agents can run in parallel, each pointed at a different segment or ICP.
7. If the user has opted in, a daily email digest summarizes the agent's qualified leads.

Under the hood: The agent evaluates each company by combining state-of-the-art LLMs with Goava's own data sources, including real-time news and event feeds, so it isn't limited to static firmographic fields. That lets it reason about context, like a recent leadership change, instead of only matching fields against a profile.

Getting the results: Qualified prospects show up in the configured Discover list and CRM. Paired with Research Agent Automations, prospects arrive with contacts and a research note already attached, so an SDR gets more than a company name and can act on it right away.

AI agents use a lot of resources, and a single misconfigured agent could burn through a customer's entire AI credit allocation. That means surprise charges and lost trust. The job was to give users control without burying them in configuration.

Defaults from real usage. Making users configure everything from zero adds a lot of friction, so we analyzed past usage in Discover to set grounded defaults. Users can adjust them later, but they start from a realistic baseline.

One switch to start. Because both locks have a sensible default, most users never touch the configuration. Turning an agent on is a switch inside the dashboard, with no setup form.

The Prospecting Agent launched as a limited pilot. Within the first 7 days, the agent accounted for 90% of all contacts exported to the CRM, and lifted total export volume by almost 7× without any additional SDR effort. More exported contacts means more pipeline, and the team got it without adding headcount.

On company qualification, the agent now drives 18% of all companies qualified in Discover, up from a fraction of that before launch. In the CEO's post-launch analysis, agent-sourced qualification went from a sliver of the total to nearly a fifth in a single week.

Before a broader rollout was announced, 12 companies had reached out on their own asking for access. Among pilot customers, Goava saw a 27% upsell rate, and the sales team made the agent central to their pitch, which helped bring in 47 new deals.

These numbers cover the first 7 days of a limited pilot. Longer-term retention effects will become clearer as the rollout expands over the next two quarters.

What went well: Basing defaults on real usage data was one of the best calls in the project. It cut setup time and configuration errors, and it pushed us into an early, useful conversation with the data team.

What I'd do differently: The schedule-based vs. signal-based trigger split still asks users to learn two different mental models for the same agent. I explored unifying them into a single trigger model, but the two run on different mechanics under the hood, one on a timer and one on continuous monitoring. Collapsing them would have meant rebuilding both, and the UX gain didn't justify the cost. I kept the split on purpose, as the right tradeoff for now.

What I learned: Designing for AI systems means designing for failure states and cost far more than traditional product work does. Treating AI credits as a design constraint, the way you'd treat screen real estate, was new to me, and I'm glad I learned it. I also had to design for where AI is today. State-of-the-art LLMs are powerful but not fully deterministic, so reliability was a design problem as much as an engineering one. That meant consistent scoring, guarding against false or overconfident signals, and showing the user when the agent was confident versus just plausible. This shaped as much of the interface as the workflow did.

What's next: The next step is coordination between agents: multiple Prospecting Agents, each scoped to a different ICP or segment, informing each other's behavior in real time. Tighter integration with Research Agent Automations would let qualification and enrichment happen in one step.