time-to-close-prediction

成約までの期間予測

あなたは、会話パターンに基づいて商談の成約時期を予測する営業ボットの構築に精通しています。エンゲージメントのシグナルと会話の動きを分析し、商談の成約時期を予測するシステムを開発者が作成できるよう支援することが目標です。

予測が重要な理由

予測の課題

Traditional forecasting:
- Rep guesses: "Should close this month"
- Based on hope, not data
- Consistently wrong
- Pipeline becomes fiction

Result:
- Missed forecasts
- Bad resource planning
- Surprised leadership

予測

Data-driven prediction:
- Analyzes conversation patterns
- Compares to historical deals
- Updates in real-time
- Confidence intervals

Result:
- Accurate forecasts
- Better resource allocation
- Proactive deal management

予測シグナル

進捗速度のシグナル

How fast is the deal moving?

Accelerating indicators:
- Response times decreasing
- Meeting frequency increasing
- Stakeholders being added
- Questions becoming more specific
- Pricing discussion initiated

Decelerating indicators:
- Response times increasing
- Meetings being postponed
- Stakeholders going quiet
- Vague or generic responses
- "We'll revisit this later"

エンゲージメントのシグナル

How engaged is the prospect?

High engagement (faster close):
- Multiple stakeholders involved
- Detailed questions asked
- Proactive outreach from them
- Documentation requested
- Reference calls requested

Low engagement (slower close):
- Single contact only
- Surface-level questions
- Slow responses
- No proactive contact
- Vague interest

ステージ完了のシグナル

Where are they in the journey?

Discovery completed:
+Needs articulated
+Decision process shared
+Timeline discussed
→ Expect 60-90 days to close

Evaluation in progress:
+Demo completed
+Trial/POC active
+Technical questions addressed
→ Expect 30-60 days to close

Decision pending:
+Proposal reviewed
+Pricing negotiated
+Contract discussed
→ Expect 0-30 days to close

購買シグナルの密度

Count buying signals over time:

Week 1: 2 signals
Week 2: 4 signals
Week 3: 7 signals
Week 4: 10 signals

Increasing density = approaching decision

Signals to count:
- Pricing questions
- Timeline questions
- Implementation questions
- Contract questions
- "What's the next step?"

予測モデル

過去のパターンとの照合

def predict_from_history(deal):
    # Find similar historical deals
    similar_deals = find_similar_deals(
        industry=deal.industry,
        company_size=deal.company_size,
        deal_size=deal.estimated_value,
        product=deal.product_interest
    )

    # Get their actual close times
    close_times = [d.days_to_close for d in similar_deals]

    # Current deal's stage
    stage_days = get_average_remaining_days(
        stage=deal.current_stage,
        similar_deals=similar_deals
    )

    return {
        "estimated_days": stage_days,
        "confidence": calculate_confidence(similar_deals),
        "range_low": percentile(close_times, 25),
        "range_high": percentile(close_times, 75)
    }

会話パターンモデル

def predict_from_conversation(deal):
    conversations = get_conversations(deal)

    # Extract features
    features = {
        "avg_response_time": calc_avg_response_time(conversations),
        "response_time_trend": calc_response_trend(conversations),
        "message_count": len(conversations),
        "stakeholder_count": count_stakeholders(conversations),
        "buying_signal_count": count_buying_signals(conversations),
        "objection_count": count_objections(conversations),
        "question_specificity": assess_question_depth(conversations),
        "sentiment_trend": analyze_sentiment_trend(conversations)
    }

    # Predict using trained model
    prediction = close_time_model.predict(features)

    return {
        "estimated_days": prediction,
        "confidence": close_time_model.confidence(features),
        "key_factors": get_key_factors(features)
    }

複合予測

def predict_close_time(deal):
    # Historical comparison
    historical = predict_from_history(deal)

    # Conversation analysis
    conversation = predict_from_conversation(deal)

    # Stage-based baseline
    stage_baseline = get_stage_baseline(deal.current_stage)

    # Weighted combination
    weights = {"historical": 0.3, "conversation": 0.5, "stage": 0.2}

    predicted_days = (
        historical["estimated_days"] * weights["historical"] +
        conversation["estimated_days"] * weights["conversation"] +
        stage_baseline * weights["stage"]
    )

    # Confidence is minimum of components
    confidence = min(
        historical["confidence"],
        conversation["confidence"]
    )

    return {
        "predicted_close_date": today() + timedelta(days=predicted_days),
        "predicted_days": predicted_days,
        "confidence": confidence,
        "factors": {
            "historical_pattern": historical,
            "conversation_signals": conversation,
            "stage_baseline": stage_baseline
        }
    }

ステージ別の予測

初期ステージの予測

Discovery stage:
- High uncertainty
- Wide range
- Base on industry averages

Prediction:
"Based on similar companies, expect 60-120 days
to decision. Current engagement suggests
you're on the faster end."

Focus on:
- Qualifying properly
- Identifying decision process
- Understanding timeline drivers

中期ステージの予測

Evaluation stage:
- Moderate uncertainty
- Can compare to historical
- Engagement signals matter

Prediction:
"Deal is moving 20% faster than average.
If current pace continues, expect close
in 35-50 days."

Focus on:
- Addressing remaining objections
- Stakeholder alignment
- Competitive positioning

後期ステージの予測

Negotiation/Decision stage:
- Lower uncertainty
- Clear remaining steps
- Timing more predictable

Prediction:
"Pending contract review and legal approval.
Historical similar deals close within 2 weeks
of this stage. Target: January 25-February 5."

Focus on:
- Removing blockers
- Expediting approvals
- Closing logistics

動的な更新

リアルタイムの予測更新

Update prediction when:
- New conversation occurs
- Meeting completed
- Stage changes
- New stakeholder added
- Objection raised/addressed
- Buying signal detected

"After today's call where CFO asked about
implementation timeline, prediction moved
from 45 days to 30 days."

予測アラート

def check_prediction_alerts(deal):
    current = deal.current_prediction
    previous = deal.previous_prediction

    alerts = []

    # Significant acceleration
    if current.days < previous.days * 0.7:
        alerts.append({
            "type": "acceleration",
            "message": f"Deal moving faster—close date moved up {previous.days - current.days} days"
        })

    # Significant slowdown
    if current.days > previous.days * 1.3:
        alerts.append({
            "type": "slowdown",
            "message": f"Deal slowing—close date pushed out {current.days - previous.days} days"
        })

    # At risk of slipping
    if deal.target_close_date and current.predicted_close > deal.target_close_date:
        alerts.append({
            "type": "at_risk",
            "message": f"May miss target close date by {(current.predicted_close - deal.target_close_date).days} days"
        })

    return alerts

予測の活用

パイプライン予測

Aggregate predictions for pipeline:

Pipeline: $1,000,000
- Deal A: $200k, 90% in Jan (weighted: $180k)
- Deal B: $300k, 60% in Jan (weighted: $180k)
- Deal C: $500k, 30% in Jan (weighted: $150k)

January forecast: $510,000

Confidence range:
- Best case: $750,000
- Expected: $510,000
- Worst case: $300,000

リソース計画

Use predictions for:
- Implementation team scheduling
- Customer success allocation
- Rep capacity planning
- Inventory/fulfillment

"Predicted $2M closing in Q1.
Implementation needs 3 additional engineers."

商談コーチング

Alert managers to at-risk deals:

"Deal ABC predicted to close 2 weeks
later than forecast. Engagement dropped.
Suggest manager review."

"Deal XYZ accelerating faster than expected.
May need to pull in implementation team
earlier."

予測精度

精度の測定

Track:
- Predicted vs actual close date
- Prediction accuracy by stage
- Accuracy by deal type
- Accuracy improvement over time

Metrics:
- Mean Absolute Error (MAE)
- Prediction within 2 weeks: X%
- Prediction within 30 days: Y%

予測の改善

Feedback loop:

1. Make prediction
2. Deal closes (or doesn't)
3. Compare prediction to actual
4. Update model weights
5. Refine signal importance

Over time:
- More accurate predictions
- Better signal identification
- Improved confidence calibration

実装

予測データモデル

{
  "deal_id": "12345",
  "prediction": {
    "predicted_close_date": "2024-02-15",
    "predicted_days_remaining": 30,
    "confidence": 0.75,
    "range": {
      "optimistic": "2024-02-01",
      "pessimistic": "2024-03-01"
    },
    "factors": {
      "accelerating": ["buying_signals_increasing", "stakeholder_engaged"],
      "decelerating": ["response_time_slow"],
      "neutral": ["typical_stage_duration"]
    },
    "updated_at": "2024-01-15T10:00:00Z",
    "previous_prediction": "2024-02-20"
  },
  "history": [
    {
      "date": "2024-01-01",
      "predicted_close": "2024-03-01",
      "confidence": 0.5,
      "stage": "discovery"
    },
    {
      "date": "2024-01-08",
      "predicted_close": "2024-02-20",
      "confidence": 0.65,
      "stage": "evaluation"
    }
  ]
}

統合ポイント

Where to surface predictions:

CRM:
- Deal record
- Pipeline views
- Forecast reports

Bot conversations:
- Adjust urgency based on timeline
- Recommend next actions
- Alert on at-risk deals

Dashboards:
- Rep-level predictions
- Team-level forecast
- Accuracy metrics

予測に基づくメッセージング

商談の加速

"Based on our conversations, you're moving
faster than most companies your size.
If we aim for [predicted date], here's what
we'd need to accomplish in the next 2 weeks..."

停滞した商談

"We had good momentum, but I sense things
have slowed. Is there something holding you
back that I can address?"

リスクのある商談

"You mentioned wanting this done by [target].
To hit that, we'd need to [specific actions].
Is that timeline still realistic?"