Hydra logoHydra · Hydra Support

Configuration

Lead qualification scoring

Turn on qualification scoring for a sales/SDR bot and Hydra scores every conversation against a rubric you control: Budget, Timeline, Authority, Use-case fit, or whatever criteria you define. Each conversation (and the lead it creates) gets a 0-100 score, a qualified/pending/disqualified decision, and a per-criterion breakdown.

Lead qualification scoring

Lead qualification scoring lets a bot tell you how promising each conversation is. When you enable it on a bot, Hydra runs a quick background AI pass after every conversation turn, scores the conversation against a rubric you control, and records the result on the conversation, and on the lead that conversation creates.

It's built for sales and SDR-style bots, where knowing which conversations are worth a follow-up actually matters. It is off by default on every bot. You opt in per bot. Leaving it off on a pure support bot is the right call: scoring adds a small per-turn AI cost, and there's no reason to spend it where you're not chasing leads.

Turning it on

Open app.hydra-help.com → Bots → pick a bot → Qualification section.

  1. Toggle Enabled.
  2. The first time you enable it, the rubric is seeded with a default BANT rubric: Budget, Timeline, Authority, Use-case fit. You can fully edit it, or leave the default if it fits.
  3. Set your thresholds (see below).
  4. Click Save.

Nothing scores until you do this. A brand-new bot, or a bot you've never opened the Qualification section on, isn't scoring anything.

The rubric

The rubric is a list of criteria. Each criterion has four parts:

  • Label: the human name, e.g. Budget.
  • Field key: a short machine name like budget. This is the stable identifier the per-criterion result is filed under.
  • Weight: how much this criterion counts relative to the others. Weights are relative: a criterion with weight 30 counts more than one with weight 10. They don't have to add up to 100, or to any particular number.
  • Scoring question: the prompt the AI uses to judge this criterion from the conversation. This is the part that matters most. Write a clear, specific question (e.g. "Has the visitor indicated they have budget allocated or a price range in mind?") rather than a vague label.

Editing the rubric:

  • + Add criterion: add a new row.
  • ✕: remove a criterion.
  • Reset to BANT: restore the default four-criterion rubric.

Edit labels, keys, weights, and scoring questions in place, then Save.

Thresholds

Two numbers decide what happens with each conversation's score:

  • Qualify at or above (default 70): a composite score at or above this is qualified.
  • Disqualify at or below (default 30): a composite score at or below this is disqualified.

Anything in the band between the two thresholds is pending. Keep the disqualify threshold below the qualify threshold. The gap between them is the pending band, and if you invert them the decision logic stops making sense.

How scoring works

After each conversation turn, the background pass does three things:

  1. Scores each criterion 0-100 based only on evidence in the conversation. This is conservative by design: no evidence means a low score. The AI won't invent buying signals that aren't there. A casual product question with no buying intent scores low, and that's intended behavior, not a bug.
  2. Computes one composite score 0-100 as the weighted average of the criteria, using your weights.
  3. Makes a decision by comparing the composite against your thresholds: qualified, disqualified, or pending.

Each conversation ends up with:

  • A composite score (0-100).
  • A decision (qualified / pending / disqualified).
  • A one-line rationale explaining the call.
  • A per-criterion breakdown: each criterion's score plus a short assessment of what in the conversation drove it.

Where results show up

  • On the conversation: visible in the conversation detail and via the API.
  • On the lead: when a conversation creates a lead, the originating conversation's qualification result travels with that lead, surfaced through Hydra's MCP/API leads endpoint. So a tool reading your leads sees not just the contact but how that conversation scored.

Qualified leads are tracked, so you can build follow-up around them. A dedicated SDR funnel view is on the roadmap for a later release. For now, read qualification off the conversation or the lead via the API.

Tips and gotchas

  • It's opt-in and off by default. If you're not seeing scores, confirm the Qualification toggle is enabled on that specific bot and that you saved.
  • Scoring is intentionally conservative. Low scores on low-signal chats are the system working as designed, not a miss. Don't tune your way around this by inflating prompts. Let weak conversations score weak.
  • Weights are relative. Bump the weight on the criteria you care about most; don't worry about making them sum to a target.
  • Write specific scoring questions. Vague prompts produce vague scores. A concrete, answerable question for each criterion is the single biggest lever on score quality.
  • Mind the threshold ordering. Disqualify-at-or-below should always sit under qualify-at-or-above, with the pending band in between.