AI-Driven Pricing Optimisation for Portfolio Companies

Table of Contents

• Why Pricing Optimisation Matters for Portfolio Companies
• The Business Case: Margin Recovery Through Better Pricing
• Understanding AI-Driven Pricing Models
• Building Your Dynamic Pricing Data Foundation
• Segment-Based Pricing Strategy
• Governance and Sales Alignment
• Implementation Roadmap
• Real-World Applications Across Industries
• Measuring Success and Iteration
• Common Pitfalls and How to Avoid Them
• Next Steps for Your Portfolio

---

Why Pricing Optimisation Matters for Portfolio Companies {#why-pricing-optimisation-matters}

Portfolio companies operate under a unique constraint: they need to scale revenue quickly whilst managing margin compression. Too often, pricing decisions are made in a spreadsheet, locked in annually, and rarely revisited. This static approach leaves 10–20% of potential margin on the table—money that could fund product development, sales expansion, or simply improve EBITDA ahead of exit.

AI-driven pricing optimisation changes this. Rather than guessing what customers will pay, you use real-time data on demand, competitive positioning, customer segment behaviour, and willingness to pay to adjust prices dynamically. For portfolio companies, this means:

• Margin recovery: A 5–15% price lift on existing customers without churn, or selective discounts that win net-new deals at higher LTV
• Speed to profitability: Pricing adjustments can be live in weeks, not quarters
• Competitive agility: Respond to market moves without renegotiating contracts
• Data-driven sales conversations: Sales teams shift from “what discount can I give?” to “here’s what this segment typically pays”

The challenge, however, is governance. Sales teams fear losing deals to aggressive pricing. Finance worries about revenue recognition and audit trails. Customers push back if they discover they’re paying differently from peers. The solution is a transparent, rules-based AI model that everyone—sales, finance, product—understands and trusts.

---

The Business Case: Margin Recovery Through Better Pricing {#the-business-case}

The Math Behind Pricing Optimisation

Consider a mid-market SaaS company in a portfolio with $5M ARR, 40% gross margin, and 200 customers. A typical cohort breakdown might be:

• Enterprise segment (10 customers, $150k ACV): Often negotiated down 20–30% from list price
• Mid-market segment (40 customers, $25k ACV): Negotiated 10–15% discounts
• SMB segment (150 customers, $5k ACV): Mostly list price, occasional 5% discounts

Now, what if AI analysis reveals that:

1. Enterprise customers would tolerate a 10% price increase (they’re locked in, switching costs are high)
2. Mid-market customers in high-growth industries (fintech, AI, logistics) have higher willingness to pay than legacy industries
3. SMB customers in certain verticals (e.g., healthcare compliance-heavy sectors) would pay 20% more for compliance-ready features

A conservative 5% blended price increase across the base yields $250k additional ARR, flowing nearly 90% to gross margin (since incremental costs are near zero). That’s $225k of margin—enough to fund a full engineering team or accelerate customer success headcount.

For a portfolio company targeting a 4–5x revenue multiple at exit, that $250k ARR uplift translates to $1–1.25M in enterprise value. And this is before accounting for the compounding effect of higher pricing on new customer acquisition—sales teams often close faster at the right price point than at artificially low discounts.

Why Traditional Pricing Fails

Most portfolio companies inherit pricing from their founding team. Founders often underprice to win early traction, then lock in those prices contractually. By the time the company scales to $5M+ ARR, they’re trapped: raising prices on existing customers triggers churn; raising prices only on new customers creates internal tension (“why does that customer pay less?”) and sales friction.

Manual pricing also doesn’t adapt to market signals. If a competitor raises prices, you don’t know it until you lose a deal. If demand spikes in a vertical, you’re still charging the same rate. If a customer’s usage patterns suggest they’re getting 3x ROI, they’re still on the original discount.

AI-driven pricing fixes this by automating the data collection, analysis, and recommendation layer. Rather than waiting for annual pricing reviews, the system continuously ingests:

• Customer attributes (industry, company size, geography, use case, adoption velocity)
• Competitive pricing (via market intelligence APIs or manual feeds)
• Usage and engagement metrics (seats, API calls, feature adoption)
• Cohort willingness-to-pay signals (derived from historical negotiations, churn by price, win rates by discount band)
• Macro signals (market growth, funding environment, customer financial health)

From this data, AI models estimate the optimal price for each customer segment—and recommend which customers or cohorts should see price increases, which should receive targeted discounts, and which are candidates for upsell.

---

Understanding AI-Driven Pricing Models {#understanding-ai-driven-pricing}

How AI Pricing Models Work

AI-driven pricing is not a single algorithm; it’s a layered system. At the core are three complementary models:

1. Willingness-to-Pay (WTP) Estimation

This model answers: “What is the maximum price this customer (or segment) will accept before they churn or reject the deal?”

Traditional approaches use surveys or conjoint analysis. AI accelerates this by learning from historical data:

• Historical negotiation data: If 20 customers in the “mid-market fintech” segment accepted a $30k ACV, and 5 rejected at $35k, the model infers WTP is somewhere between $30–35k
• Churn patterns: If customers who received a 30% discount in year 1 churn at 2x the rate of those who received 10% discounts, the model learns that aggressive discounts correlate with churn (either because the wrong customers were acquired, or because low-price customers are less committed)
• Competitive win/loss data: If you lose deals to Competitor X at prices above $28k but win at $25k, the model adjusts WTP estimates based on competitive positioning
• Usage and expansion signals: If a customer is consuming 10x the typical usage for their cohort, they’re getting outsized value and have higher WTP

These signals feed into a machine learning model (often a gradient boosting model like XGBoost, or a neural network) that predicts WTP for each customer or segment.

2. Price Elasticity Modelling

Elasticity answers: “How much will demand (or churn) change if we raise prices by 10%?”

In portfolio companies, elasticity varies dramatically by segment:

• Enterprise customers with multi-year contracts and high switching costs: inelastic (demand doesn’t drop much if price rises)
• SMB customers with month-to-month terms: elastic (demand is sensitive to price)
• Customers with strong product-market fit and high NPS: inelastic
• Customers with low adoption or high support costs: elastic (they’re marginal, price-sensitive, or both)

AI models estimate elasticity by running pricing experiments (A/B tests, randomised cohort pricing) or by inferring from historical data. The output is a curve showing: “If we raise prices 10%, we’ll lose 2% of customers in this segment, but 1% in that segment.”

Combining elasticity with WTP, the model then solves an optimisation problem: “What price maximises revenue (or margin) given the elasticity and WTP constraints?“

3. Churn Risk and Retention Scoring

Not every price increase is wise, even if WTP suggests customers can afford it. Some customers are flight risks:

• High-growth startups that are well-funded and have many alternatives
• Customers with low product engagement (they’re not sticky)
• Customers with executive turnover or budget cuts
• Customers in industries experiencing disruption

AI models score churn risk using:

• Usage trends: Declining feature adoption, lower API call volume, fewer active users
• Support ticket sentiment: Increasing complaints, escalations, or sentiment negativity
• Financial signals: News of layoffs, funding rounds, acquisition rumours, credit rating downgrades
• Cohort behaviour: Historical churn rates for similar customers
• Contract health: Upcoming renewal dates, recent expansion, or contraction

The pricing model then applies a “churn risk penalty”: for high-risk customers, it recommends smaller price increases or targeted retention discounts, even if WTP is high. For low-risk, sticky customers, it’s more aggressive.

The Role of Generative AI

Generative AI (like GPT-4) plays a supporting role in pricing optimisation. Rather than making pricing decisions directly, it:

• Synthesises market intelligence: Summarises competitive pricing moves, analyst reports, and customer feedback into pricing insights
• Generates pricing narratives: Creates customer-facing and internal messaging to justify price increases (e.g., “New compliance features added 30% value; fair to reflect in pricing”)
• Automates recommendation writing: Produces detailed, contextualised pricing recommendations for each customer or segment, explaining the rationale
• Handles edge cases: When a customer disputes a price or requests an exception, generative AI can draft a personalised response that references their specific usage, value, and segment benchmarks

This is where how AI can improve pricing decisions with data-driven experimentation and dynamic adjustment becomes operationally real: the AI doesn’t replace human judgment, but it removes the friction of manual analysis and accelerates the conversation.

---

Building Your Dynamic Pricing Data Foundation {#building-data-foundation}

The Data Stack You Need

AI pricing models are only as good as the data feeding them. For a portfolio company, you need:

Customer Data

• Company attributes: Industry, geography, company size (employees, revenue), funding stage, growth rate
• Product usage: Seats, API calls, feature adoption, monthly active users, data volume, query latency
• Engagement: NPS, support ticket volume and sentiment, feature requests, training completion
• Financial: ARR, MRR, contract value, discount applied, renewal date, expansion/contraction history
• Relationship: Account owner, executive sponsor, decision-maker titles, tenure

This data typically lives across multiple systems: your CRM (Salesforce), product analytics (Amplitude, Mixpanel), billing system (Zuora, Stripe), and support platform (Zendesk, Intercom). The first step is a data integration layer—either a CDP (Segment, mParticle) or a custom ELT pipeline (Fivetran, Stitch, or dbt) that normalises these sources into a unified customer data warehouse.

Competitive Intelligence

• Competitor pricing: List prices, discounts observed in the market, pricing by feature tier
• Competitive win/loss: Reasons customers chose you vs. competitors, and vice versa
• Market positioning: Analyst reports, customer reviews, feature comparisons

This can come from manual research, sales team input (via a structured win/loss process), or third-party tools like Pricefx, Competera, or Revealbot that aggregate pricing data.

Macro and Cohort Data

• Cohort performance: Churn rate, NPS, support costs, expansion rate, and CAC payback by segment
• Market signals: Industry growth rates, funding environment, regulatory changes
• Pricing experiments: Results from any A/B tests or cohort-based pricing trials

Building the Data Pipeline

For a typical portfolio company, the pipeline looks like:

Salesforce → Data Warehouse (Snowflake / BigQuery / Redshift)
  ↓
Stripe/Zuora → Data Warehouse
  ↓
Product Analytics → Data Warehouse
  ↓
Manual Competitive Intel → Data Warehouse
  ↓
ML Feature Engineering Layer (dbt, Python) → Feature Store
  ↓
Pricing Model (XGBoost, Neural Network) → Recommendations
  ↓
API / Dashboard → Sales, Finance, Product

The key is that data should be fresh (ideally daily, at minimum weekly) and trustworthy. Before feeding data into the model, you need:

• Data validation: Checks that customer IDs match across systems, that ARR figures are consistent between billing and CRM, that usage metrics are within expected ranges
• Outlier handling: Identifying and treating data anomalies (e.g., a customer with 1000x normal usage—is it a data error or a real signal?)
• Lineage documentation: Tracking where each data point comes from, so you can explain recommendations to sales teams

Many portfolio companies partner with data engineering teams to build this. If you’re working with a venture studio or AI-focused technology partner—like PADISO’s Platform Development services across multiple regions—they can accelerate the pipeline build and integrate it with your existing systems.

Privacy and Compliance Considerations

When building a pricing model that ingests customer data, you must consider:

• GDPR / Privacy Act: If you have EU or Australian customers, you need explicit consent to process their data for pricing optimisation. Document this in your privacy policy and terms of service.
• Price discrimination laws: In some jurisdictions (e.g., EU), charging different prices to similar customers based on protected characteristics (race, gender, age) is illegal. Your model must be auditable and defensible.
• Transparency: Customers increasingly expect to understand why they’re being charged a certain price. Be prepared to explain pricing decisions in human terms.

The safest approach: segment-based pricing (more on this below) is more defensible than individual customer pricing, because you’re treating all customers in a segment equally and the segmentation is based on transparent business factors (industry, use case, company size).

---

Segment-Based Pricing Strategy {#segment-based-pricing}

Why Segments Matter

Instead of calculating a bespoke price for every customer (which creates governance nightmares and sales friction), AI-driven pricing works best through segments. A segment is a cohort of customers with similar characteristics, behaviour, and willingness to pay.

For a B2B SaaS company, typical segments might be:

• Enterprise Financial Services (banks, wealth managers, insurance): High ACV ($100k+), long sales cycles, compliance-sensitive, inelastic pricing
• Mid-Market FinTech (startups, alternative lenders): Medium ACV ($30–50k), fast-growing, willing to pay for innovation, moderately elastic
• Enterprise Healthcare (hospital systems, health plans): High ACV ($80k+), heavily regulated, slow to adopt, inelastic
• SMB SaaS (small tech companies, consultancies): Low ACV ($5–15k), price-sensitive, elastic
• Non-profit / Government: Often require special pricing, grants, or steeper discounts

Within each segment, all customers pay the same base price. However, the AI model can recommend:

• Segment-level price increases: “Enterprise FinTech segment should move from $40k to $44k ACV (10% increase)”
• Feature-based upsells: “Customers in the Enterprise Financial Services segment who haven’t adopted the compliance dashboard should be offered it as a $10k add-on”
• Retention discounts: “High-risk customers in the Mid-Market FinTech segment at renewal should receive a 5% retention discount”
• New customer acquisition pricing: “New Enterprise customers in the Healthcare segment can be quoted at $90k ACV; historical data shows 70% close rate at this price”

Building Your Segment Model

Segmentation should be driven by pricing relevance—factors that genuinely correlate with willingness to pay and elasticity. Common dimensions:

1. Industry/Vertical: Different verticals have different unit economics and willingness to pay. A fintech might pay 3x more than a nonprofit for the same product.

2. Company Size: Larger companies have bigger budgets and higher switching costs. Segment by employee count or annual revenue.

3. Use Case / Product Adoption: Customers using your product for mission-critical processes have higher WTP than those using it for nice-to-have workflows.

4. Growth Stage: Funded startups have more pricing flexibility than bootstrapped companies. Segment by funding stage.

5. Geography: Willingness to pay varies by region. US and Western Europe typically pay more than APAC or emerging markets.

6. Product Tier / Feature Usage: Customers using advanced features should pay more than those on basic tiers.

Your AI model should identify which dimensions are most predictive of WTP. Using AI-driven pricing strategies from leading firms, you can run a correlation analysis: which segment variables most strongly predict churn, NPS, expansion, or historical price acceptance?

Then, you create a segment matrix. For example:

Segment	Industry	Company Size	Use Case	Base ACV	Elasticity	Churn Risk	Recommended Action
Enterprise FinTech	FinTech	100–500 emps	Mission-critical	$45k	Low	Low	+10% price increase
Mid-Market FinTech	FinTech	20–100 emps	Core workflow	$25k	Medium	Medium	+5% increase, bundle compliance feature
SMB SaaS	Tech	<20 emps	Nice-to-have	$8k	High	High	Hold price, focus on retention
Enterprise Healthcare	Healthcare	500+ emps	Compliance-heavy	$85k	Low	Low	+8% increase, new compliance module

Pricing by Segment: A Worked Example

Let’s say your pricing model recommends the following for a $5M ARR company:

Current State:

• Enterprise segment (15 customers): $50k ACV, $750k ARR
• Mid-market segment (50 customers): $20k ACV, $1M ARR
• SMB segment (200 customers): $5k ACV, $1M ARR
• Nonprofit/Government (10 customers): $8k ACV, $80k ARR (special pricing)
• Total: 275 customers, $3.83M ARR (note: the remaining $1.17M is from other sources or older cohorts)

AI Recommendation:

• Enterprise: Increase to $55k ACV (10% increase). Churn risk low; willingness to pay high; elasticity low. Expected impact: 1 customer churns, net gain is $650k ARR (13 × $55k)
• Mid-market: Increase to $22k ACV (10% increase). Churn risk medium; offer retention discount to top 10 customers (keep at $20k). Expected impact: 2 customers churn, but 8 retention discounts cost $16k total. Net gain is $960k ARR (48 × $22k – $16k)
• SMB: Hold at $5k ACV. Churn risk high; elasticity high. No price increase. Focus on upsell and expansion.
• Nonprofit/Government: Hold at $8k ACV (regulatory and mission-driven; price increases damage reputation).

Projected Outcome:

• New ARR: $650k + $960k + $1M + $80k = $2.69M (from these segments)
• Incremental ARR: ~$250k
• Incremental margin (assuming 90% incremental margin): ~$225k
• Timeline: Implement over 3 months (existing customers at renewal, new customers at new pricing)

This is a realistic scenario for a portfolio company with disciplined execution.

---

Governance and Sales Alignment {#governance-alignment}

The Sales Team Problem

Here’s where most AI pricing projects fail: sales teams don’t trust the model, or they don’t understand it, so they ignore it or work around it.

Common objections:

• “Our top customer will churn if we raise their price.”
• “The model doesn’t account for the competitive pressure in that account.”
• “We need flexibility to close deals.”
• “Finance is using this to blame us for missing quota.”

These objections are valid. Sales teams have information that the model doesn’t (relationship strength, executive changes, competitive threats). The solution is not to force sales to follow the model blindly, but to create a governance framework where the model is advisory, sales has input, and decisions are documented.

Building the Governance Framework

1. Pricing Authority Matrix

Define who can approve what:

Action	Threshold	Authority	Approval Process
New customer at recommended price	Any	Sales rep	None (autonomous)
New customer at 5% discount to recommended	Any	Sales manager	Documented reason
New customer at 10%+ discount	<$20k ACV	Sales manager	Documented reason + CFO review
New customer at 10%+ discount	$20k+ ACV	VP Sales + CFO	Joint approval + board notification
Existing customer price increase at renewal	<10%	Account manager	None (send offer)
Existing customer price increase at renewal	10–20%	Account manager + Sales manager	Joint conversation with customer
Existing customer price increase at renewal	>20%	VP Sales + CEO	Strategic review

This matrix ensures that small discounts (which are inevitable and normal) don’t require bureaucratic approval, but large deviations from the model are visible and justified.

2. Monthly Pricing Review Cadence

Establish a rhythm:

• Weekly: Sales team inputs feedback on specific accounts where they’re requesting exceptions. Model team reviews and flags high-risk decisions.
• Monthly: Pricing committee (VP Sales, CFO, Product) reviews pricing performance, model accuracy, and segment adjustments. Discuss exceptions and learn from them.
• Quarterly: Deep-dive on model performance. Compare recommended prices to actual prices closed. Measure elasticity, churn, and margin impact. Adjust segments or model parameters if needed.

This cadence keeps the model honest and ensures sales has a voice.

3. Transparency and Explainability

When the model recommends a price, sales teams need to understand why. Provide:

• Peer benchmarking: “Customers like yours (Enterprise FinTech, 200–300 employees) typically pay $50–55k. We’re recommending $52k.”
• Value justification: “This customer’s usage is 2x the cohort average, suggesting 2x value. Pricing should reflect that.”
• Churn risk: “This customer has high churn risk (low engagement, declining usage). Recommend a small discount to retain.”
• Competitive context: “Competitor X is pricing similar features at $48k. We’re at $50k, justified by our compliance and support.”

Make this transparent in a dashboard or report that sales teams can access. When a sales rep questions a recommendation, they should be able to see the logic.

4. Exception Handling

Define a process for when sales needs to deviate:

1. Sales rep identifies an exception (e.g., “Customer X will churn if we raise their price beyond 5%”)
2. Rep documents the reason in a structured form (competitive threat, relationship risk, budget constraint, etc.)
3. Model team reviews and either:
   • Agrees and adjusts the recommendation
   • Disagrees and explains why (e.g., “Our data shows similar customers with similar risk profiles accepted this increase; recommend trying it”)
4. Decision is made and logged
5. Outcome is tracked: Did the customer churn? Did they accept the price? What did we learn?

Over time, this feedback loop makes the model smarter. If sales consistently overestimate churn risk, the model learns to be more aggressive. If they consistently underestimate it, the model becomes more conservative.

5. Sales Incentive Alignment

If your sales team is incentivised purely on revenue, they’ll discount aggressively. If they’re incentivised on margin, they might be too conservative. The ideal:

• Blended metric: 60% revenue, 40% margin (or 70/30, depending on your business model)
• Price realization bonus: Small bonus if actual ACV is within 5% of recommended ACV
• Retention bonus: Bonus if customer retention is above cohort average (discourages unsustainable discounting)

These incentives should be designed with sales leadership, not imposed. Involve them in the design.

---

Implementation Roadmap {#implementation-roadmap}

Phase 1: Foundation (Weeks 1–4)

Goal: Get data in place and validate the model on historical data.

Activities:

1. Data audit: Map all customer data sources (CRM, billing, product, support). Identify gaps. Estimate data quality issues.
2. Segment definition: Work with sales and product leadership to define 4–6 core segments based on pricing relevance.
3. Historical pricing analysis: Analyse past pricing decisions, discounts, and outcomes. Identify patterns in what customers accepted.
4. Competitive intelligence gathering: Compile competitor pricing, win/loss reasons, and market positioning.
5. Data pipeline build: Set up ETL to populate a data warehouse with clean, validated customer data.

Outputs: Data dictionary, segment definitions, historical pricing dataset, competitive intelligence report.

Effort: 200–300 hours (can be done in-house or with a partner like PADISO’s AI advisory services).

Phase 2: Model Development (Weeks 5–10)

Goal: Build and validate the pricing model.

Activities:

1. Feature engineering: Create predictive features from raw data (e.g., “usage growth rate last 6 months”, “NPS quartile”, “days since last support ticket”).
2. WTP model training: Train a regression or classification model to predict willingness to pay by segment. Use historical pricing and churn data.
3. Elasticity estimation: Run pricing experiments (if available) or infer elasticity from historical data and competitor moves.
4. Churn risk scoring: Build a classifier to identify high-churn-risk customers.
5. Backtesting: Test the model on historical data. Compare recommended prices to actual prices and outcomes. Measure accuracy and bias.
6. Sensitivity analysis: Understand how the model behaves under different scenarios (recession, competitor price cuts, etc.).

Outputs: Trained model, backtesting report, sensitivity analysis, model documentation.

Effort: 300–500 hours (requires ML expertise; often outsourced).

Phase 3: Governance and Integration (Weeks 11–14)

Goal: Build the operational framework and integrate the model into sales and finance systems.

Activities:

1. Governance framework design: Define pricing authority matrix, exception process, review cadence. Get buy-in from VP Sales, CFO, CEO.
2. Sales enablement: Train sales teams on the model, how to interpret recommendations, and the exception process. Address concerns.
3. API/Dashboard build: Create a system for sales to access pricing recommendations. Integrate with Salesforce or your CRM.
4. Finance integration: Set up reporting so finance can track actual prices vs. recommended prices, and measure margin impact.
5. Change management: Communicate the initiative to the organisation. Frame it as a tool to help sales close deals faster, not to constrain them.

Outputs: Governance playbook, sales training materials, API/dashboard, finance reporting.

Effort: 150–200 hours.

Phase 4: Pilot and Rollout (Weeks 15–20)

Goal: Test the model with a subset of sales team, then scale.

Activities:

1. Pilot with one sales region or segment: Run the model for 2–4 weeks with one sales team. Track recommended vs. actual prices, close rates, churn.
2. Feedback and refinement: Gather feedback from pilot team. Adjust the model or governance if needed.
3. Full rollout: Expand to all sales teams. Monitor closely for the first month.
4. Ongoing monitoring: Track key metrics weekly (recommended price vs. actual, close rate, churn, margin).
5. Model retraining: Every quarter, retrain the model with new data to keep it fresh.

Outputs: Pilot results, refined model, rollout plan, monitoring dashboard.

Effort: 100–150 hours (mostly monitoring and support).

Total Timeline and Investment

• Duration: 20 weeks (5 months) from start to full rollout
• Internal effort: 300–400 hours (1–2 FTE for 5 months)
• External support (if outsourced): $100k–$200k (depends on data complexity and model sophistication)
• Expected ROI: If successful, $200k–$500k in incremental margin in year 1, with payback in 3–6 months

For portfolio companies, this ROI is attractive. Many venture studios and AI agencies (like PADISO’s AI & Agents Automation services) can compress the timeline to 12–16 weeks and reduce internal burden by taking on the model development and integration.

---

Real-World Applications Across Industries {#real-world-applications}

SaaS and Software

Challenge: High competition, low switching costs, elastic pricing.

AI Pricing Approach:

• Segment by use case (mission-critical vs. nice-to-have), company size, and industry
• Use engagement metrics (feature adoption, API usage) as proxies for willingness to pay
• Implement dynamic pricing at renewal time, not at contract signature (less disruptive)
• Offer feature-based upsells (e.g., “Advanced analytics module for $X”) rather than blanket price increases

Result: One portfolio SaaS company increased ACV by 12% and improved gross margin by 8 percentage points over 18 months using segment-based pricing and feature bundling.

Financial Services and FinTech

Challenge: Highly regulated, long sales cycles, high customer value, but also high churn risk if pricing is perceived as unfair.

AI Pricing Approach:

• Segment by customer type (bank, wealth manager, alternative lender, insurance), geography, and regulatory regime
• Use compliance requirements and data volume as pricing signals (more regulated = higher WTP; more data = higher usage costs, justify higher price)
• Implement tiered pricing based on data residency, audit requirements, and SLAs
• Incorporate competitive intelligence from RFP responses and deal reviews

Result: PADISO’s AI for Financial Services clients have used AI-driven pricing to increase ACV by 15–20% in enterprise segments whilst maintaining competitive positioning. For regulated sectors, the transparency of the pricing model (based on objective factors like compliance level and data volume) makes price increases more defensible to customers and regulators.

Insurance

Challenge: Highly commoditised, price-sensitive customers, but also high willingness to pay for risk mitigation and claims automation.

AI Pricing Approach:

• Segment by insurance type (general, life, health), customer size (insurer, broker, MGU), and use case (underwriting, claims, risk)
• Use claims volume and loss ratio as pricing signals (higher claims = higher value from automation)
• Implement outcome-based pricing (e.g., “Pay 20% of savings from claims automation”) to align incentives
• Use AI for Insurance Sydney strategies to price claims automation, conduct risk monitoring, and underwriting AI differently

Result: An insurance portfolio company increased ACV by 18% by moving from flat-fee pricing to value-based pricing tied to claims volume and savings.

Marketplace and E-Commerce

Challenge: High volume, low-margin transactions, but opportunities for dynamic pricing at scale.

AI Pricing Approach:

• Segment by seller type (brand, reseller, marketplace), geography, and product category
• Use demand signals (search volume, conversion rate, inventory turnover) to adjust prices in real-time
• Implement dynamic pricing for high-demand items (raise prices when demand spikes; lower when inventory is high)
• Use dynamic pricing via cloud platforms and machine learning to optimise for revenue or margin

Result: A marketplace portfolio company increased gross margin by 3–5 percentage points (a large number at scale) by using dynamic pricing on high-demand products and seller-specific pricing based on seller tier and performance.

---

Measuring Success and Iteration {#measuring-success}

Key Metrics to Track

1. Pricing Performance

• Average ACV (by segment and overall): Should increase over time (adjusted for mix shift)
• Recommended ACV vs. Actual ACV: Measure the gap. If sales are consistently closing at 20% below recommendations, either the model is too aggressive, or sales needs coaching
• Price realisation: (Actual ACV / Recommended ACV) × 100. Target: 90%+ realisation
• Discount rate: % of deals closed below list price. Should decrease or stabilise over time

2. Business Impact

• Incremental ARR: New ARR from price increases, minus churn
• Incremental margin: Incremental ARR × gross margin %
• Revenue per customer: Should increase (or at least not decrease) as pricing optimises
• Win rate: Should stay flat or improve (better pricing = faster closes, not fewer closes)

3. Customer Health

• Churn rate (by segment): Should not increase significantly post-pricing change. If it does, pricing was too aggressive
• NPS: Should not decline (pricing increases can hurt NPS if not communicated well)
• Expansion rate: Should improve (better pricing = better customer fit, more likely to expand)
• Support cost per customer: Should decrease (better pricing attracts higher-quality customers)

4. Model Accuracy

• Backtesting accuracy: Model’s historical predictions vs. actual outcomes. Target: 70%+ accuracy in predicting whether a customer will accept a price increase
• Elasticity estimate accuracy: Compare predicted churn from price increases to actual churn
• Segment stability: Do segments remain predictive over time, or do they need to be redefined?

Iteration Cycle

Monthly:

• Review pricing performance metrics (ACV, realisation, margin)
• Review sales feedback and exceptions
• Adjust governance or sales messaging if needed

Quarterly:

• Retrain the model with new data
• Reassess segment definitions (do they still predict WTP?)
• Run pricing experiments (A/B test price increases on a cohort)
• Review competitive intelligence and adjust recommendations
• Assess churn and NPS impact

Annually:

• Comprehensive model review and overhaul
• Segment redefinition if needed
• Strategy review: Are we optimising for margin, revenue, or customer acquisition? Adjust the model accordingly

A/B Testing and Experimentation

The best way to validate the model is to run pricing experiments:

• Cohort pricing test: Randomly assign new customers in a segment to different price points. Track close rate, churn, and expansion. Use results to refine elasticity estimates.
• Existing customer test: At renewal, offer a cohort of existing customers a price increase, and a control cohort the same price. Measure acceptance and churn.
• Feature bundling test: Test different feature bundles and prices. Identify which bundles have the highest WTP.

These experiments are gold for model refinement. They replace guesswork with data.

---

Common Pitfalls and How to Avoid Them {#common-pitfalls}

Pitfall 1: Over-Optimising for Margin, Under-Optimising for Growth

Problem: The model recommends aggressive price increases that maximise margin but hurt growth. Sales misses quota because customers churn or reject deals.

Solution: Define the objective clearly upfront. Are you optimising for:

• Margin (good for mature, profitable companies)?
• Revenue (good for growth-stage companies)?
• Blended (margin for existing customers, lower prices for new customers)?

Build the model to optimise for your chosen objective. If you’re a growth-stage portfolio company, the model should recommend lower prices on new customers to accelerate growth, even if it means lower near-term margin.

Pitfall 2: Ignoring Data Quality

Problem: The model is trained on dirty data (duplicate customers, misaligned revenue figures, missing usage data). Recommendations are garbage.

Solution: Invest in data quality upfront. Audit all data sources. Set up validation rules. Track data quality metrics. If 30% of your customer records have missing usage data, you can’t build a reliable model.

Pitfall 3: Losing Sales Team Buy-In

Problem: Sales team sees the model as a constraint, not a tool. They work around it, ignore it, or actively sabotage it.

Solution: Involve sales in model design and governance. Show them how it helps them (faster closes, better benchmarks, less time spent on pricing decisions). Give them input on exceptions. Make the model advisory, not prescriptive. Celebrate wins (e.g., “This rep used the model to close a $50k deal that we would have discounted to $40k”).

Pitfall 4: Not Accounting for Competitive Dynamics

Problem: You raise prices, but a competitor drops prices. Suddenly, your pricing recommendations are out of sync with the market.

Solution: Incorporate competitive intelligence into the model. Set up a process to track competitor pricing (via sales input, market intelligence tools, or periodic RFP analysis). When competitive dynamics shift, adjust WTP estimates downward. Build in a “competitive buffer”—if a competitor is pricing aggressively, recommend smaller price increases.

Pitfall 5: Implementing Too Fast

Problem: You build the model in 8 weeks and roll it out to the entire sales team immediately. Chaos ensues. Sales team loses trust.

Solution: Pilot first. Test with one region or segment for 4–6 weeks. Gather feedback. Refine. Then roll out gradually. This builds confidence and allows you to catch issues early.

Pitfall 6: Neglecting Churn Risk

Problem: The model recommends a 20% price increase for a customer. Sales agrees. Customer churns. You lose $100k ARR.

Solution: Always incorporate churn risk into the model. Score every customer for churn risk (using engagement, support sentiment, contract health, etc.). For high-risk customers, recommend smaller increases or retention discounts, even if WTP is high.

Pitfall 7: Not Measuring Actual Impact

Problem: You implement the model, but you don’t track whether it’s actually working. Six months later, you don’t know if margin improved or if churn increased.

Solution: Set up monitoring from day one. Track ACV, margin, churn, and NPS by segment, before and after the model rollout. Compare to a control group (e.g., a sales region that didn’t use the model, if possible). Measure ROI quarterly.

---

Next Steps for Your Portfolio {#next-steps}

If you’re running a portfolio company and pricing feels static, here’s how to move forward:

Step 1: Audit Your Current Pricing

Spend a week understanding your current state:

• What are your segments?
• What’s your ACV by segment?
• What discounts are you giving, and why?
• What’s your churn rate by segment?
• Have you tested price increases? What happened?

This audit will reveal opportunities and constraints.

Step 2: Define Your Objective

What’s the goal of pricing optimisation?

• Increase margin by 10% whilst keeping churn flat?
• Increase ACV by 15% with acceptable churn (<2%)?
• Improve win rates on new customers by lowering prices for SMB whilst raising for enterprise?

Be specific. This objective drives the entire model.

Step 3: Assess Your Data Readiness

Do you have:

• Clean customer data (CRM, billing, product usage)?
• Historical pricing and discount data?
• Churn and expansion data?
• Competitive intelligence?

If you’re missing critical data, plan to fill those gaps first. This often takes 4–8 weeks.

Step 4: Choose Your Partner

You have two options:

In-house: If you have a data scientist and ML engineer on staff, you can build the model internally. Timeline: 5–7 months. Cost: salary + tools (~$50k–$100k).

Outsourced: Partner with a venture studio or AI agency that has pricing expertise. They can compress the timeline to 12–16 weeks and take on the heavy lifting. Cost: $100k–$250k, but you get faster time-to-value and less internal burden.

Many portfolio companies choose outsourced because speed matters. If you’re working with a venture studio like PADISO, they can integrate pricing optimisation with broader AI strategy and platform engineering work, reducing friction and increasing ROI.

Step 5: Start Small

Don’t try to optimise pricing for all 500 customers at once. Pick one segment (e.g., “Mid-market FinTech”) and run a pilot. Get wins, build credibility, then expand.

Step 6: Iterate

The first model won’t be perfect. But if you measure, learn, and adjust quarterly, it will get smarter over time. After 12 months, you should see:

• 5–15% increase in ACV
• 5–10 percentage point increase in gross margin
• Flat or improving churn
• Sales team confidence in the model

---

Conclusion

AI-driven pricing optimisation is not about squeezing customers for every last dollar. It’s about understanding what different customer segments are willing to pay and aligning your pricing to reflect the value you deliver.

For portfolio companies, the opportunity is significant: 10–20% margin recovery is realistic with disciplined execution. The key is building a data foundation, creating a model that’s transparent and defensible, and establishing governance that keeps sales, finance, and product aligned.

Start with your data. Define your objective. Pick your segments. Build (or partner to build) the model. Pilot. Measure. Iterate.

If you’re in Australia and looking for a partner to accelerate this work—from data pipeline to model development to sales integration—PADISO’s AI advisory and platform engineering teams can help. We’ve worked with portfolio companies across SaaS, fintech, and enterprise software to implement pricing optimisation, and we can compress the timeline and reduce internal burden.

For companies in other regions, PADISO’s platform development teams in New York, Miami, Los Angeles, Chicago, Boston, Seattle, Austin, Dallas, and Atlanta have similar expertise. We also work with case studies showing real results across industries.

The question isn’t whether to optimise pricing—it’s when. The sooner you start, the sooner you recover margin and improve your portfolio company’s path to profitability and exit.