HUDRA, the AI that finds your prospects, runs the conversations and converts 52% of interested leads into meetings

HUDRA is a multi-agent system developed by Qubitech that automates B2B prospecting end to end. Deployed from February to May 2026, over four months, at a 40-person B2B SaaS company, it took over the entire top of the sales funnel. Research agents working in parallel build up to 100,000 qualified prospect records in a few days. The system then creates the campaigns, runs each email conversation through to completion and books the meetings directly in the team's calendars.

Across 64,000 emails sent at 800 per business day, the campaign generated an 8% reply rate and a 3% interest rate on delivered emails. 52% of interested prospects were converted into meetings, that is nearly 1,000 qualified meetings spread across a team of six sales reps. Around 95% of conversations were handled without human intervention.

HUDRA is language-model agnostic and works with LLMs such as Claude or Mistral, in the cloud or hosted locally to guarantee data sovereignty. Its intelligence rests on a Qdrant vector database enriched by several tens of thousands of conversations. Setup takes a few days and full autonomy is reached after two to three weeks of supervision. The cost per qualified meeting is cut by more than 50 times compared with human prospecting.

What is a multi-agent B2B prospecting system?

A multi-agent B2B prospecting system is an artificial intelligence made up of specialised agents that automate the entire prospecting cycle. The agents identify and qualify prospects, write personalised messages, run the email conversations through to the booked meeting and sync the slots with the sales team's calendars.

HUDRA, developed by Qubitech, applies this principle at scale. Unlike an email sequencer, it does not stop at sending. It acts as a prospect search engine, a campaign generator, an email client and a prospecting CRM. As soon as a prospect replies, the thread is understood and carried through to the meeting, with the level of context of a human exchange.

What was the client's situation?

The client is a 40-person B2B SaaS company. It sells a management and invoicing solution aimed at small and mid-sized businesses. Its addressable market potentially covers every SME in its target markets. Its sales team has six people, already busy with product demos and closing.

Outbound prospecting relied on four disconnected tools. A database was used to find contacts, a sequencer handled sending, a CRM tracked follow-up and a calendar was used to schedule meetings. The sales reps spent most of their time on low-value tasks such as finding targets, gathering contact details, writing each message individually and chasing follow-ups. For lack of time, a large share of warm prospects was lost between forgotten follow-ups and conversations abandoned before the meeting.

What problem needed to be solved?

The underlying problem was not a lack of tools but the impossibility of personalising at scale. Mass emailing platforms send generic messages whose reply rate collapses and which damage the brand image. Email providers also flag these near-identical sends as spam, which degrades deliverability.

Conversely, manual personalisation produces excellent results but does not scale. A sales rep realistically writes 40 to 50 genuinely personalised emails a day. On a fully loaded salary, each message costs CHF 7 to 8. Simply building a file of qualified prospects takes weeks of work. No human organisation can produce tens of thousands of qualified records in a few days. The challenge was therefore to multiply prospecting volume without giving up any of the quality and personalisation of a human exchange.

How does HUDRA work, step by step?

The full process unfolds in seven steps.

1. Defining the target. The user specifies the target sector and region. This is the only human input required upstream.
2. Research and qualification in parallel. Several agents simultaneously identify matching companies, gather their contact details and collect information on each company and its relevant contacts. The system thus builds up to 100,000 qualified records in a few days, the equivalent of several years of manual research. For markets subject to compliance requirements, HUDRA draws on the client's existing databases or CRM, already built in line with regulations.
3. Campaign creation. The user specifies the offer, the positioning and, if needed, the desired structure of the messages. A writing agent produces messages tailored to each recipient's profile. Several campaigns can run in parallel.
4. Scheduling and sending. The user selects the contact list, the time window and the pace, here 800 emails per business day. HUDRA includes its own email client with detailed per-prospect status tracking and connects to existing inboxes such as Gmail or Outlook.
5. Conversations run end to end. As soon as a prospect replies, the thread is handled autonomously. Messages are understood and answered accordingly, just as a permanently available sales rep would do.
6. Meeting booking. When an exchange converges, a scheduling agent proposes slots based on the availability of the connected calendars and books the meeting. The team is notified as soon as it is confirmed.
7. Tracking and reporting. A daily report is generated. The team receives a summary each morning and can check the dashboard at any time.

Which edge cases does the system handle on its own?

• A request for additional information receives an answer drawn from the knowledge base.
• A general question receives a suitable answer, within the defined limits.
• When a prospect is on holiday, follow-ups are paused and then resumed on their return.
• When a prospect asks to be contacted again in a few months, the follow-up is scheduled for the right time.
• When an email indicates that the person has left the company, HUDRA identifies the new contact and continues the exchange with them.
• When a prospect no longer wishes to be contacted, they are removed immediately from all campaigns.
• When a meeting is cancelled, new slots are proposed automatically.

How does human supervision work?

At the start, every AI reply is drafted and validated by a human. Qubitech recommends this supervised phase for two to three weeks. The system learns the team's style and corrections, then switches to autopilot. The validations feed a vector knowledge base that refines the exchanges over time, on a foundation of several tens of thousands of conversations. This knowledge is stored in the vector database and not in the model. It is therefore retained even if the LLM is changed.

Even on autopilot, HUDRA applies a confidence threshold. If an agent is not certain of its answer beyond that threshold, set for example at 90%, it hands over to a human rather than risk an unsuitable reply. Sensitive information defined upstream is never disclosed. The system relies on its knowledge base rather than inventing an answer, which eliminates hallucinations. On average, around 95% of conversations are handled without any human intervention, and in some configurations all of them.

Which technologies does the system use?

• Language-model agnostic. HUDRA works with any LLM, in the cloud or hosted locally for data sovereignty. Models such as Claude or Mistral, the latter particularly strong in French, are supported, among others.
• Qdrant vector database. The intelligence accumulated over several tens of thousands of conversations is stored in a Qdrant vector database, independent of the model, and survives a change of LLM.
• Predictive scoring. The system estimates each prospect's conversion probability by matching them against similar companies and profiles from its history.
• Channels. The email and WhatsApp channels are operational. SMS and LinkedIn can be added as needed.
• Integrations. HUDRA connects to existing calendars and inboxes such as Gmail or Outlook and centralises management end to end, which removes the need to stack a third-party CRM.

What results did HUDRA achieve in four months?

The deployment ran from February to May 2026, at 800 emails per business day, that is 16,000 emails a month and 64,000 in total. The campaign achieved an 8% reply rate and a 3% interest rate on delivered emails. B2B cold emailing benchmarks put a good interest rate between 3 and 5%. 52% of interested prospects were then converted into meetings. The deployment thus generated nearly 1,000 qualified meetings in four months, that is around 250 a month, spread across the calendars of the team's six sales reps, which represents about two meetings per day per rep. Around 95% of conversations were handled without any human intervention.

Normalised to a base of 1,000 delivered emails, the system generated 80 replies, 30 interested prospects and around 16 qualified meetings. Campaigns run on already-known contacts, imported from the client's CRM or from past relationships, generally achieve better results.

What caution applies at higher volume?

These rates call for cautious reading when scaling up. The first campaigns naturally target the most receptive segments of the audience. As the scope widens, reply and interest rates tend mechanically to decline. Qubitech recommends applying a caution factor to any projection beyond the tested scope and managing the scale-up in stages, measuring the rates at each step. Realistic sizing is better than an unkeepable promise.

How much does automated prospecting cost compared with a sales rep?

A sales rep costs around CHF 8,000 a month on a fully loaded salary, plus CHF 250 to 400 a month in tools for the contact database, the email sequencer and the CRM. At 40 to 50 genuinely personalised messages a day, each message costs CHF 7 to 8, before even counting the time spent building the file. Over four months, they would have produced 3,000 to 4,000 personalised emails. At the same conversion rates, that represents about fifty meetings, at a cost of several hundred francs per meeting.

With HUDRA, the marginal cost of a personalised message drops from several francs to a few centimes, without any loss of context. On this deployment, the cost per qualified meeting was cut by more than 50 times. It comes to a few dozen francs, against several hundred with human prospecting. In terms of volume, HUDRA handled in four months what a sales rep would produce in six years.