Loading tutorials…
Loading tutorials…
Freddy AI is the headline feature on Freshsales Pro+ — and the one most teams turn on, get garbage scores, and quietly stop checking. Get the data hygiene + training inputs right and Freddy becomes the highest-leverage feature in the portal. Here is the discipline that earns rep trust.
Who this is forRevOps leads, sales managers, and founders on Freshsales Pro / Enterprise. Free / Growth tiers have no Freddy AI. If reps say 'I don't trust the Freddy score' or you have never opened the Freddy Insights tab, this tutorial is for you.
What you'll need
Step 1
Freddy AI is a machine learning model. Bad input data = bad predictions. Audit data hygiene before turning anything on.
Freddy needs: 6+ months of historical Contact and Deal data, clear Lifecycle Stage values, a stamped "Became Customer Date" for positive examples, a stamped "Lost Reason" for negative examples, and consistent Lead Source tracking.
Run a quick audit: Contacts list → filter "Lifecycle Stage = Customer" → count. If under 200, Freddy will not have enough positive examples to learn from — wait or use Freshworks' generic model.
Check Lead Source field completeness: Contacts list → filter "Lead Source is empty" → if more than 30% of contacts are missing source, fix data collection at form/import layer before turning on Freddy lead scoring.
Check Lifecycle Stage timestamps: filter "Became MQL Date is empty AND Lifecycle Stage is MQL." If many rows return, you have stage values without stamped dates — Freddy needs the timestamps to learn velocity patterns.
Spend 4-6 hours on data hygiene before enabling Freddy. Garbage in, garbage out is doubly true for ML.
Step 2
Freddy AI Copilot is the in-product assistant — email drafting, deal insights, next-best-action. Free to enable on Pro+.
Admin Settings → Freddy AI → Freddy AI Copilot → toggle ON.
Configure which modules Copilot acts on: Contacts (suggested next-best-action), Deals (deal insights, summary, suggested follow-ups), Conversations (email reply drafts).
Set user-level access: Admin Settings → User Management → Roles → for each role, toggle "Use Freddy AI Copilot." Decide who gets AI assist (typically all sales roles) vs read-only viewers (typically no).
On each Deal record, you will now see a Freddy AI panel showing: deal health (Healthy / Stuck / At Risk), suggested next action, AI-generated deal summary.
On each Contact record, Freddy suggests next best action (call vs email vs nurture) and best time to contact based on past response patterns.
Step 3
Admin Settings → Freddy AI → Predictive Contact Scoring. The single highest-ROI Freddy feature when configured properly.
Admin Settings → Freddy AI → Predictive Contact Scoring → enable.
Freddy auto-trains on your historical data: who converted to Customer vs who churned at MQL vs who never engaged. The model is built in 24-48 hours after enabling.
Once trained, Freddy assigns each contact a score 0-100 + a category (Cold / Warm / Hot). The score appears on the contact record and is available as a filter on list views, sequences, and workflows.
Tune signal weights: Admin Settings → Freddy AI → Predictive Contact Scoring → Signals tab. Adjust which signals Freddy weights heavier: Email engagement, Website visits (if Freshmarketer connected), Form submissions, Phone activity, Job title match, Industry match.
Set Hot/Warm/Cold thresholds: default 75+ = Hot, 50-74 = Warm, <50 = Cold. Calibrate after 30 days of data — most teams find defaults are too generous and tighten Hot to 80+.
Step 4
Deal Insights surface stalled, healthy, and at-risk deals automatically. The most under-used Freddy feature.
Admin Settings → Freddy AI → Deal Insights → enable per pipeline.
Freddy analyzes each open deal and tags it: Healthy (active, on track), Stuck (no activity in expected window), At Risk (negative sentiment in recent emails or activity drop-off), Trending Positive (engagement increasing).
On the Deals list view, add the "Deal Insight" column. Sort by Insight to surface At Risk + Stuck deals to the top of weekly pipeline review.
Set up a workflow (Admin Settings → Automations → Workflows) that fires when Deal Insight = At Risk: send notification to deal owner + manager with "Freddy flagged this deal at risk — review by EOW."
Pin a Dashboard widget: Reports → Dashboards → "Sales Pipeline" → add widget "Deals by Insight." Visible to leadership for at-a-glance pipeline health.
Step 5
Freddy AI Writer drafts personalized email openers, follow-ups, and replies. Saves reps 5-10 hours/week if used properly.
Admin Settings → Freddy AI → Freddy AI Writer → enable.
In any email compose window (within Freshsales or Gmail/Outlook extension), click the Freddy AI icon → pick action: "Draft cold outreach," "Draft follow-up," "Summarize thread," "Rewrite tone."
Train Freddy on your voice: Admin Settings → Freddy AI → Freddy AI Writer → Voice Settings → upload 10-20 examples of your best-performing past emails. Freddy learns to mimic phrasing, structure, and CTAs.
Set guardrails: forbid Freddy from claiming things you have not approved (pricing, SLAs, integrations). Add "Banned Phrases" in Voice Settings.
Reps should treat Freddy drafts as a starting point, not a finished email. Sequence templates remain the source of truth for repeatable plays; Freddy drafts the situational one-offs.
Step 6
Reports → Freddy AI Performance. Track score accuracy, prediction confidence, and rep adoption.
Reports → Freddy AI Performance (or Admin Settings → Freddy AI → Performance dashboard).
Key metrics: Score Accuracy (of contacts Freddy scored Hot 90 days ago, what % became Customer?), Prediction Confidence (Freddy reports its own confidence — values under 60% mean it does not have enough data), Rep Adoption (what % of reps use Freddy Copilot suggestions weekly).
If Score Accuracy < 50% after 60 days: data hygiene problem. Audit Lifecycle Stage timestamps, Lead Source completeness, and activity logging.
If Prediction Confidence consistently < 60%: not enough training data. Wait another 60 days or add more historical data via import.
If Rep Adoption < 30%: reps do not trust the scores or the UI is hidden. Surface Freddy score on the Highlight Panel; do a 30-min team training showing how the scores have predicted recent wins.
Step 7
Enterprise tier adds Freddy AI Forecast — automated commit/best-case/upside predictions per AE per quarter.
Admin Settings → Freddy AI → Forecast Insights → enable (Enterprise only).
Freddy analyzes each open deal's probability + age + activity + sentiment and aggregates into a per-AE quarterly forecast.
Compare Freddy's forecast vs AE-submitted forecast weekly. Gaps over 25% mean either Freddy is wrong (audit signals) or the AE is sandbagging/overcalling (coaching opportunity).
For leadership: pin the Freddy Forecast dashboard widget to the Sales Leadership dashboard. Weekly board update becomes "Freddy says $X, AEs commit $Y, here is the delta and why."
Forecast Insights is the feature that justifies the jump from Pro to Enterprise for most teams >25 reps. For teams under 25 reps, Pro's deal insights + manual forecasts are sufficient.
Common mistakes
Enabling Freddy AI without 6 months of historical data
What goes wrong: Freddy has nothing to learn from. Scores are essentially random. Reps see Hot scores on dead contacts and Cold scores on real opportunities. Within 30 days, reps mute Freddy and stop checking. Cost: $0 saved time, full price for the Pro tier upgrade.
How to avoid: Wait until you have 6+ months of data and 200+ Customer-stage contacts before enabling Predictive Scoring. Use Deal Insights (which works on less data) in the interim.
Not stamping Lifecycle Stage timestamps
What goes wrong: Freddy cannot learn velocity patterns (Lead → MQL → SQL → Customer). It treats all contacts at the same stage as equivalent, regardless of how long they have been there. Scoring accuracy drops 30-40%. Cost: half the value of Pro tier ($300-500/mo unused).
How to avoid: Build workflows that stamp "Became MQL Date," "Became SQL Date," "Became Customer Date" on every stage transition. Backfill historical timestamps from activity logs if possible.
Trusting Freddy Lead Scoring with empty Lead Source data
What goes wrong: If Lead Source is empty for 40% of contacts, Freddy cannot distinguish a paid-search lead from an outbound lead from a referral. Score accuracy is ~50% (coin flip). Reps chase the wrong contacts. Cost: 20-30% of SDR time wasted on cold leads scored as Hot.
How to avoid: Fix Lead Source at the form/import layer. Make Lead Source required on every contact create. Audit existing contacts and backfill from UTM data or origin records.
Never tuning Hot/Warm/Cold thresholds
What goes wrong: Default 75+ = Hot. Your data shows actual conversion happens at 85+. SDRs chase 'Hot' contacts at 75-84 that rarely convert. SDR pipeline drops by 20% because they are working warmer-than-actual leads. Cost: $50-100K in lost pipeline per quarter.
How to avoid: After 30 days, pull the conversion rate of contacts at each Freddy score range. Set thresholds where the actual conversion-rate inflection happens.
Using Freddy AI Writer drafts as-is without rep review
What goes wrong: Reps send Freddy-drafted emails verbatim. The drafts have subtle hallucinations: wrong product features, made-up customer references, generic phrasing. Reply rate drops because emails sound AI-generated. Recipients call it out. Cost: brand damage + 10-20% reply rate decline.
How to avoid: Treat Freddy drafts as a starting point. Reps must edit at least 30% of the draft before send. For high-stakes emails (proposals, exec outreach), human review is non-negotiable.
Not recalibrating after a major motion change
What goes wrong: You launch a new product line or pivot ICP. Freddy is still trained on the old motion's data. Scores predict the old customer profile, not the new one. SDRs and AEs work the wrong leads for 3-6 months before anyone notices. Cost: 6 months of misallocated outreach.
How to avoid: After any major motion change (new product, new ICP, new pricing), re-train Freddy with the last 90 days of data only. Re-audit signal weights and thresholds.
Recap
Done — what's next
How to set up Freshsales deal pipelines that produce accurate forecasts
Read the next tutorial
Hand it off
Freddy AI is the most-bought, least-deployed Freshsales feature. Teams pay the Pro/Enterprise premium and use 20% of what they bought. A specialist who has tuned Freddy for 30+ portals knows which signals to weight, which thresholds to tighten, and how to drive rep adoption. EverestX Freshsales specialists run Freddy optimization at $14-16/hr — typically $400-800 for a from-scratch configuration.
See specialist rates
Minimum: 6 months of historical Contact data, 200+ Customer-stage records, Lifecycle Stage values + stamped timestamps, Lead Source completeness >70%. Below this, Freddy will produce scores but they will be near-random. Wait until you have the data — early activation creates rep distrust that takes 6-12 months to recover.
24-48 hours for initial model training. Then 30 days of monitoring to calibrate thresholds. After 90 days you should see Score Accuracy >65%. Freddy retrains automatically every 30 days on new data, but you can force a retrain via Admin Settings → Freddy AI → Predictive Scoring → Retrain.
Yes — Pro+ Freddy AI Predictive Scoring uses both default and custom fields as signals. In Admin Settings → Freddy AI → Predictive Scoring → Signals, you can add custom fields (Dropdown, Boolean, Number) as scoring inputs. Custom Modules (Enterprise only) get a separate Freddy scoring config.
Freddy is more focused on predictive analytics (lead scoring, deal health, forecasting). Breeze (HubSpot 2024+) leans heavier on generative AI (content writing, AI agents, chatbots). For pure sales-motion ML, Freddy is more mature; for marketing + sales content generation, Breeze is stronger. See our [HubSpot tutorials](/tutorials/crm-sales/hubspot-crm) for parallel guidance.
Yes — each Freddy feature (Copilot, Predictive Scoring, Deal Insights, Writer, Forecast Insights) has its own toggle in Admin Settings → Freddy AI. Disable the ones you do not have data quality for; keep the ones that work. Most teams enable Copilot + Deal Insights immediately and wait 60-90 days before enabling Predictive Scoring.
Freshsales
A Freshsales pipeline is the single most-referenced screen in the CRM. Set up wrong, the forecast number is fiction and reps stop trusting it. Set up right, your VP Sales walks into Monday with a real number. Here is the discipline that holds up.
Freshsales
Freshsales ships with 40+ default reports. They cover 30% of what your team actually needs. The Custom Report Builder is where the real leverage lives — and where most owners build dashboards nobody opens. Here is the design discipline that makes them load-bearing.
Freshsales
Freshsales data quality problems are quiet killers. They do not break the CRM — they just make every report wrong, every Freddy score garbage, and every campaign target the wrong people. Here is how to diagnose what is broken and fix it without making it worse.
Freshsales
DIY Freshsales is a great idea — until it isn't. This is the honest framework for when the cost of self-managing exceeds the cost of hiring help, and how to tell which side you are on.