Reps spend 40% of their day deciding what to do next. Call now or wait? Send a quote or schedule a demo? Disqualify or keep warm? Without real-time guidance, teams drift into reactive mode — checking Slack, answering email, avoiding hard choices. Velocity X's AI Brain solves this by reading ten signals from every deal and outputting a ranked action list: "call now 0.94 confidence", "send quote 0.87", "wait 7 days then nudge 0.76". The rep glances at the dashboard, picks the 0.94 call, and moves.
What "Next Best Action" Actually Means
Next-best-action is not "what should happen eventually". It's "what is the highest-ROI activity for this deal, right now, given what we know". The algorithm reads current deal state, compares it to historical win patterns, and outputs a sorted action list with confidence scores. An action is specific: not "follow up" but "call at 2pm after they've had time to review the proposal" or "disqualify if they don't respond to the email in 3 days".
The Ten Input Signals
Brain reads deal state across these dimensions:
- Deal value. Pipeline weighted by revenue. High-value deals get different urgency curves than small ones.
- Days since last contact. Recency. 14 days since a call is different from 2 days.
- Current deal stage. Prospect vs Qualified vs Negotiating vs Closed Lost. Stage position strongly predicts next action.
- No-show count. How many times has the prospect missed a meeting? 0 is different from 3.
- Response velocity. How fast does this prospect reply to emails/calls? Hours vs days.
- AI sentiment from notes. Brain runs a mini sentiment classifier on rep notes ("they seemed excited" vs "a lot of budget headwinds"). Captured in a 0–1 score.
- Proposal sent date. Did we send them something to look at? How long ago?
- Meeting cadence. How often have we met? Once, twice, regularly?
- Competitor signals. If the rep mentioned competitor names in notes, Brain flags competitive risk.
- Team velocity. What's the average conversion rate for deals at this stage, for this team, in this month?
The Ranking Model
Brain doesn't hardcode "if days since contact > 7 then call". Instead, it runs a contextual ranker: a prompt-based decision tree that reads all ten signals, weights them against historical outcomes (trained on your team's deal-win patterns), and outputs a ranked action list with confidence scores.
Example: Deal A is $50k, Qualified stage, 9 days since contact, prospect replies in 2 hours, sentiment 0.8 (positive), proposal sent 5 days ago. Brain ranks: "Call now 0.94", "Send quote 0.84", "Wait 3 days 0.71". Example: Deal B is $8k, Prospect stage, 2 days since contact, replies in 24 hours, sentiment 0.4 (lukewarm), no proposal yet. Brain ranks: "Send discovery email 0.89", "Call 0.76", "Wait 5 days 0.62".
The ranker uses Claude Sonnet as the inference engine — it's fast enough to evaluate a deal in 1-2 seconds, and capable enough to handle the nuance ("if they mentioned budget concerns but replied enthusiastically, lean toward a call rather than waiting").
Which Model, Which Threshold
Velocity X uses Claude Sonnet for the default next-best-action ranking. Sonnet runs in 1-2 seconds per deal, costs ~$0.0008 per evaluation, and produces actionable confidence scores that correlate strongly with rep outcomes (reps who follow Brain's top-ranked action close 23% more deals).
For high-value deals (configurable threshold, default $100k+), Brain auto-upgrades to Claude Opus. Slower (3-5 seconds), more expensive (~$0.01 per eval), materially more nuanced at detecting deal-specific risks. The cost is justified: an Opus recommendation can prevent a $200k deal sliding into the lost column.
Confidence thresholds are configurable per team. Default: top action must be 0.80+ to auto-suggest. Below 0.80, Brain says "inconclusive — your call" and the rep decides. Reps can override any recommendation; Brain learns from overrides.
The UI Surface
On the rep dashboard, each deal card shows a Brain badge in the top-right: green (0.85+), yellow (0.70-0.84), grey (below 0.70). Tap the badge to see the ranked action list and the reason (why is "call" ranked first? "High value + 9 days since contact + positive sentiment + prospect replies fast"). If the rep disagrees, they tap "dismiss" — Brain logs the feedback and adjusts future recommendations for this deal.
Admins see aggregated stats: "Team average follow-up latency dropped from 3.2 days to 1.1 days after Brain launched." "Reps who follow Brain's top 3 actions close 23% more deals." "Most-overridden action last month: 'Wait 5 days' (reps prefer to call earlier than the model suggests)".
How It Compares
Salesforce Einstein prescribes actions but doesn't explain them — reps distrust black-box recommendations. Pipedrive Smart uses rule-based scoring (if X then do Y), which misses the 20% of deals that break the rule. Mapview.ai focuses on lead scoring for outbound, not deal progression. Brain combines explanation (here's why), personalization (I've seen your team's patterns), and nuance (confidence scores instead of binary yes/no) in a single dashboard glance.
Frequently Asked Questions
Does Brain recommend different actions for different reps on the same deal?
No — the deal context is objective. What differs is the rep's track record. If Rep A closes 35% of Qualified deals but Rep B closes 15%, Brain might recommend "call now" for Rep A and "nurture 7 days" for Rep B on the same deal, because the probability of closing differs. This is surfaced in admin dashboards.
What if my team's win patterns are weird (e.g., we win more by waiting than by calling)?
Brain learns from your data first. The model reads your team's historical deals and re-weights the ranking. If your team genuinely closes more deals by waiting, Brain will recommend waiting. No ideology, only data.
Can Brain see competitor mentions in deal notes?
Yes. If a rep writes "prospect mentioned they're also talking to HubSpot", Brain flags competitive risk and may lean toward a faster action (call sooner) to move things forward before the prospect decides elsewhere.
Does Next Best Action cost extra?
No. It's bundled in Velocity X's platform pricing. The per-deal inference cost is rounding error (~$0.0008 per Sonnet eval, ~$0.01 per Opus eval). For a 20-person team evaluating 500 deals/month, the cost is under $100/month.
Can I turn off recommendations for certain deal stages?
Yes. In team settings, you can configure which stages Brain should rank (e.g., "skip Prospect stage, only rank Qualified+"). This is useful if your first-touch workflow is entirely scripted and you want Brain only on later-stage decisions.
What if a deal hasn't been touched in 30 days — does Brain still recommend?
Yes. Stale deals get higher urgency signals. Brain will likely recommend "call to reconnect or disqualify" if 30+ days have passed. This surfaces old deals that reps forget about.
The Bottom Line
Next-best-action turns deal state (value, age, sentiment, history) into a ranked action list in 1-2 seconds. Reps glance at the badge, see the top recommendation and the reasoning, and either act or override — either way, they're deciding faster. The model learns from overrides, team win patterns, and historical outcomes. No hardcoded rules, no "call if > 7 days" nonsense — just contextual intelligence. Bundled in Velocity X, no separate AI subscription. See the follow-up email feature in our companion post.