Developing a Data-Driven Product Strategy

Use growth economics to choose customers, channels and product investments. Define the numbers, segment them by buying motion and distinguish attributed demand from demand the activity actually caused.

Why this matters

Moving upmarket, adding a free tier or investing in partners each assumes a route to profitable growth. Test the acquisition cost, payback, retention and conversion behind that assumption.

Blended averages hide different businesses. Attribution misses untracked buying activity and cannot establish causation alone. Find the binding funnel constraint before adding demand. It may be onboarding, security review or another product problem.

Growth economics: the numbers strategy runs on

CAC, LTV and payback

Skok’s “SaaS Metrics 2.0” (2011) frames recurring-revenue economics around customer value and time to recover acquisition cost. Use three measures:

• Customer acquisition cost (CAC): total sales and marketing cost for a period divided by the new customers won in it. Say whether it’s blended (all channels) or paid only, and whether it’s fully loaded (salaries, tools, events, partner commissions).
• Lifetime value (LTV): average revenue per account, multiplied by gross margin, divided by the churn rate. Customer lifetime is one divided by churn.
• CAC payback: CAC divided by the monthly gross margin a new customer brings in (new monthly recurring revenue per customer, multiplied by gross margin). The answer is in months.

Skok suggests LTV above three times CAC and payback under twelve months, allowing around twenty months for enterprise land-and-expand. Use these as discussion references rather than universal thresholds.

Three adjustments matter for B2B and platform products:

• Expansion breaks the simple LTV formula. With net revenue retention above 100%, LTV computed from churn alone runs toward infinity. Cap the lifetime (five years is a common choice) and say so, or model expansion with the retention methods in Chapter 09.
• Gross margin is a product variable. AI features carry inference cost. If gross margin falls, payback lengthens and LTV shrinks even when price and CAC don’t move, so shipping an agentic feature is also a growth decision.
• Partner economics belong in CAC. Margin shared with a reseller or managed service provider is a cost of acquiring that customer; leaving it out flatters the partner motion.

B2B pipeline metrics

SiriusDecisions popularized MQL (marketing-qualified lead), SAL (sales-accepted lead), SQL (sales-qualified lead), opportunity and close. Track conversion, days per stage, win rate and pipeline coverage: pipeline divided by quota. Pipeline velocity combines opportunity count, win rate and deal size divided by cycle length.

Count accounts and buying groups carefully. Several contacts downloading a white paper can inflate lead-based demand. Agree stage definitions and systems with sales and marketing operations. Product usually owns trial, activation and product-qualified signals.

Product-led growth and the PQL

OpenView Partners defines product-led growth (PLG) as a go-to-market strategy in which the product itself is the primary driver of acquisition, conversion and expansion. By OpenView’s own account, its partner Blake Bartlett coined the term in 2016.

Wes Bush defines a product-qualified lead (PQL) through experienced product value, often in a trial or free plan. Marketing engagement such as opening an email supports an MQL; completing a valuable task supports a PQL.

Bush’s method for defining a PQL has three steps:

1. Find the behaviors that correlate with upgrading, usually tied to the “aha” moment when users first get value.
2. Review the candidate signals: product interest, number of users, features used, usage and spend patterns, and adoption velocity.
3. Start with a minimum viable PQL, a simple threshold such as a number of logins, and refine it toward the behaviors that best predict conversion.

Track signups, activation, time to value, trial conversion, PQL conversion, expansion and net revenue retention. A product-qualified account (PQA) combines value signals from several users within an enterprise account.

Enterprise PLG often supports product-led sales: product creates demand and sales closes it. Forrester’s 2026 buying report says over 60% of buyers use trials, rising to 78% for purchases above $10 million. Treat trial activation as a product funnel measure.

Designing journeys with data

The funnel by journey stage

Add measures to Chapter 01’s buying and usage journeys. McClure’s AARRR provides lifecycle stages. Gartner’s six buying jobs capture the non-linear work of enterprise buying groups.

A journey funnel model records, for each stage:

• the buyer’s question at that stage (“Is this problem worth solving?” “Will security approve it?”);
• the persona who owns that question;
• the touchpoints: owned (website, docs), paid, earned (press, reviews, community), partner, product and AI answer engines;
• the entry definition: what has to be true for an account to count as in this stage;
• a leading metric and the conversion rate to the next stage;
• median time in stage;
• whether the stage is observable (yes, partly or no);
• the owner and the data source, with its blind spots.

The entry definition does most of the work. “Trial started” means nothing until you decide whether it’s a sign-up, a connected data source or a first real outcome. Write it so two analysts would count the same accounts.

The stages you can’t see

G2 surveyed over 1,000 software buyers in August 2025 and reported half started in an AI chatbot, its leading shortlist source. Treat this as vendor research. Forrester’s 2026 report similarly describes AI answers followed by validation with trusted people.

Add unobserved AI conversations, peer referrals and community activity to the funnel. Estimate them through self-reported attribution, win/loss interviews and assistant audits. The first tracked page view may occur well after discovery.

Measuring channel effectiveness

What ad metrics measure

Ad platforms measure delivery with impressions and CPM, response with click-through and cost per click, and attributed outcomes with cost per lead, opportunity and ROAS. B2B decisions usually benefit from opportunity-level measures. These remain credited outcomes. Incremental cost and ROAS require causal estimates.

Incrementality is the question

Incrementality asks what would not have happened without the activity. It requires a comparable counterfactual. Gordon, Zettelmeyer, Bhargava and Chapsky compared observational methods with fifteen Facebook experiments in Marketing Science (2019). Observational estimates often missed experimental lift, in either direction. Targeting already-likely buyers creates selection bias.

Lewis and Rao studied 25 retailer and brokerage experiments in QJE (2015). Even millions of users produced imprecise advertising-return estimates because purchase variation exceeded ad effects. Expect wide intervals in small B2B samples. Plan how inconclusive evidence will affect the decision.

Lift tests at concept level

Three designs cover most cases:

• User-level randomized lift tests (holdouts). A random treatment group sees the ads and a control group doesn’t. Garrett Johnson, Randall Lewis and Elmar Nubbemeyer’s “ghost ads” method (2017) logs the ads the control group would have seen instead of paying to show them placebo ads, which cuts cost and sharpens the comparison.
• Geo experiments (matched markets). Turn spend on or off in comparable regions and compare outcomes, with no user tracking. In B2B this works only where a region holds enough target accounts to show a difference.
• Account-level holdouts. The B2B adaptation: randomly withhold a share of target accounts from a campaign and compare opportunity creation. It’s a practice rather than a published framework, and it suits account-based marketing because the target list already exists.

Whatever the design, write the hypothesis, unit of randomization, primary metric, guardrails, minimum detectable effect, duration and decision rule before the test starts. Chapter 05 covers experiment readouts in depth.

Marketing mix modeling and multi-touch attribution

Marketing mix modeling (MMM) relates aggregate channel spend to outcomes over time, controlling for trend and seasonality. It needs history and struggles at low volumes. Google’s Meridian, generally available in January 2025, uses Bayesian geo models. Meta’s Robyn uses ridge regression. Calibrate both with experiments and retain statistical review.

Multi-touch attribution (MTA) works bottom-up, assigning credit across the touchpoints it can observe for each person or account before a conversion. It’s fast and granular, which suits tuning campaigns in flight. It’s also correlational, sees only tracked touches and keeps losing ground to tracking restrictions and consent rules.

Use triangulation: MMM for allocation, experiments for causal estimates and calibration, attribution for campaign optimization. When a well-powered experiment conflicts with the other tools, investigate their bias.

Attribution models and their biases

The models

Attribution rules divide conversion credit among observed touches. Their biases affect budget, especially across long enterprise buying journeys.

The tooling has already moved. In 2023 Google retired first click, linear, time decay and position-based models from Google Ads and Google Analytics 4, leaving data-driven and last-click attribution; it said fewer than 3% of Google Ads web conversions used the rules-based models. Those models now live mainly in the CRM and marketing automation tools where B2B teams build pipeline attribution.

Biases that apply to every model

Four biases cut across all of them:

• Only observed touches count. Peer referrals, community threads, podcasts and AI assistant answers are “dark” touches, and their credit goes to whatever the model can see.
• Selection bias. Targeting puts ads in front of people already likely to buy, so exposure and conversion are correlated before the ad does anything. This is the bias Gordon and colleagues measured.
• Double counting. Every ad platform attributes conversions to itself, so the platforms’ reports often add up to more conversions than actually happened.
• The wrong unit. Person-level models spread credit across individuals when a group made the decision.

Google’s data-driven model compares observed exposures with similar holdback users for each advertiser and conversion. Use it for within-platform optimization. Use experiments to assess strategic allocation to the platform.

Measurement after the privacy shifts

For several years, product and marketing teams planned for the end of third-party cookies in Chrome. It never came. The position as of this edition, based on Google’s most recent statement (October 2025):

• Chrome kept third-party cookies. Google dropped its plan to deprecate them in July 2024. In April 2025 it said it would not roll out a standalone prompt for third-party cookies either, leaving users to choose in Chrome’s privacy settings; Incognito mode blocks them by default. In October 2025 Google retired most of the Privacy Sandbox technologies, including the Attribution Reporting API, citing ecosystem feedback and low adoption, while committing to keep working on an interoperable attribution standard through the W3C. The only restriction actually rolled out was a test on 1% of Chrome users in 2024.
• iOS requires permission to track. Since iOS 14.5, enforced from April 2021, apps must use Apple’s AppTrackingTransparency framework to ask permission before tracking users across other companies’ apps and sites or reading the device’s advertising identifier. Without consent, the identifier returns zeros.

Consent, platform policy, blockers and browser settings leave journeys incomplete. Plan around partial observation using:

• First-party data: CRM records, product telemetry and billing data, which describe the account that buys.
• Aggregate models: MMM needs no user-level data.
• Experiments: account-level and geo holdouts need no cross-site tracking.
• Self-reported attribution: “How did you first hear about us?” at sign-up or the first sales call captures touches no tracking will.

How strategy uses these numbers

Use growth evidence for where-to-play, how-to-win and investment decisions.

Where to play. LTV, CAC and payback by segment and motion show which customers and routes to market are worth growing. A motion that pays back in 15 months and expands at 130% deserves a different investment from one that pays back in 30 months and churns.

How to win. Conversion and time-in-stage data from the journey funnel model show the binding constraint: the stage where the most value is lost. Very often it’s a product problem. Trials that never activate point at onboarding and time to value; deals that sit in security review point at missing controls and documentation. Fixing the constraint usually beats buying more demand that will pile up at the same stage.

Investment allocation. MMM and experiments inform budget shifts. The strategy memo states the expected incremental return as a range, and the test result that would change the decision.

Protect long-term demand. Dawes’s 95:5 rule is a heuristic, not a measured constant: most B2B buyers are out of market. Attribution centered on demand capture can underfund the activities that make future buyers remember you.

The guardrail. Goodhart’s law warns that a measure turned into a target stops being a good measure. Set an MQL target and you’ll get MQLs. Pair every growth target with a quality measure from the next stage (MQL-to-opportunity rate, PQL-to-paid conversion, first-year retention) so that gaming one shows up in the other.

How to apply it

This method builds the growth evidence behind a product strategy over roughly one quarter. It assumes an AI harness connected to the CRM, product analytics and billing, and marketing and sales operations willing to work with you on definitions.

Weeks 1 and 2: Agree the definitions.

1. List every growth metric your strategy depends on (CAC, payback, LTV, NRR, stage conversions, activation, PQL or PQA) with its definition, owner, source system and time window. Where teams define a term differently, record both and agree one.
2. Decide the cuts: at minimum segment and motion (direct, partner, self-serve, cross-sell), plus region if economics differ by region.

Weeks 2 to 4: Build the scorecard and the journey funnel.

3. Ask the harness for CAC, payback and LTV by motion and segment, each figure traced to its source and the gross margin it assumes. Fill the growth metrics scorecard.
4. Draw the journey funnel model with sales, marketing and customer success. Write each stage’s entry definition, pull conversion and median time in stage, and mark how observable each stage is.
5. Name the binding constraint, the stage with the largest loss or longest wait, and ask what product change would move it.

Weeks 4 to 6: Audit the attribution.

6. Fill an attribution assessment for the report that budget decisions actually use.
7. Add a self-reported attribution question at sign-up or first meeting if none exists, and compare its answers with the tracked view.

Weeks 6 to 12: Test what matters.

8. Pick the one or two bets where the money is large and the evidence weakest, and write a channel experiment plan for each. A B2B test may need two quarters to read.
9. If the company runs an MMM, ask whether it has been calibrated with experiments, and offer yours as calibration points.

Week 12: Decide and write it down.

10. Write the data-driven strategy memo, with impact as a range and the result that would change your mind. Chapter 34 covers memo writing; Chapter 29 covers the review where it gets decided.
11. Put the scorecard and funnel on a monthly review cadence. Refresh the numbers each month, and change a definition only when the review agrees to, with the old and new versions recorded.

The AI lens

Building AI-powered products

AI products change the growth math in ways that classic SaaS metrics hide.

• Inference cost moves payback. Gross margin sits in the denominator of payback, so when an AI feature lowers margin every customer takes longer to pay back. Put gross margin by product line on the scorecard and recompute payback when it moves. Chapter 15 covers pricing that keeps margin aligned with usage.
• Activation means trust. For an AI product, a login is a weak activation signal. Define the PQL around accepted outputs (an accepted diagnosis, a merged suggestion, an approved action).
• Usage pricing makes LTV a range. When revenue depends on consumption or outcomes, average revenue per account varies more from month to month. Report LTV and payback as ranges with the assumption behind each end.
• AI products meet longer validation. Gartner reported in May 2026 that 69% of B2B buyers turn to sales reps to validate AI-generated insights. Buyers apply the same skepticism to an AI product’s own claims, so expect security, legal and AI governance reviews to add a stage to the journey, and measure it.
• AI assistants describe your product to buyers. If an assistant gets your product wrong, buyers carry the error into the evaluation, which makes accurate public documentation a product concern.

Answer engines and acquisition in 2026

Google reported 1.5 billion monthly AI Overview users in 200 countries and territories in May 2025. Pew’s March 2025 study of 900 US adults found summaries on 18% of Google searches. Traditional-result clicks were 8% with summaries versus 15% without; summary links received clicks on about 1% of visits. Gartner’s predicted 25% search decline remains an unconfirmed forecast. January 2026 press reports described OpenAI plans to test US Free and Go ads. The product implications:

• Fewer, later, less attributable clicks. Buyers arrive with a shortlist already formed in an assistant, and your attribution credits whatever they touched next.
• An invisible stage. The AI conversation belongs in the journey funnel model, marked not observable and estimated from self-reported attribution and audits.
• Share of answer. A practitioner metric is emerging: how often, and how accurately, AI assistants mention and recommend your product for the questions your target buyers ask. Run a fixed set of buyer questions across the major assistants on a schedule and score presence, position and accuracy.
• Evidence feeds answers. Assistants draw on public documentation, review sites, community discussion and third-party material, so clear, current documentation is now an acquisition asset.

Pranjal Aggarwal and colleagues named this generative engine optimization (GEO) in a KDD 2024 paper. In their benchmark, some content changes improved visibility in generative engine responses by up to 40%, varying by domain. That shows visibility can be influenced; vendor GEO playbooks remain untested tactics.

AI in attribution and growth analytics

The measurement platforms now use machine learning throughout. Beyond Google’s data-driven attribution and open-source Bayesian MMM, Meta has added an incremental attribution setting that uses holdouts and machine learning to estimate which conversions an ad caused, for bidding and reporting. Expect reported conversions to fall when you turn it on, because it counts fewer of them. Each of these tools is trained on what its owner can observe, and no platform is a neutral judge of its own effectiveness.

Doing the work with AI

An approved harness can draft payback and conversion analyses from CRM, telemetry, billing and ad systems. Define metrics and evidence requirements before requesting the numbers:

• the denominator and time window of every rate, and the segment cuts behind any average;
• the source system and filter for every figure that will reach a decision memo;
• a range wherever a number comes from a model or a small sample;
• what the data can’t see: dark touches, unobservable stages and accounts missing from the join.

Check for wrong account joins, renewals counted as new customers, conflicting metric definitions and conclusions stronger than noisy data supports. Spot-check joins against sources. Chapter 19 covers verification.

When the analysis involves customer records, pipeline data or contract values, use only AI tools your company has approved for that data.

The Beacon case: finding where Beacon’s growth money should go

The ARR bridge. You start by splitting last year’s growth. Prior-year ARR was about $34.8M ($48M divided by 1.38). With NRR at 118%, existing customers now contribute about $41.0M, so expansion net of churn added about $6.3M and new logos about $7.0M. Nearly half of Beacon’s net new ARR came from the installed base, yet the proposal puts all the incremental money into net-new demand.

The scorecard. The harness pulls a year of closed-won deals and sales and marketing cost by motion. Beacon won about 90 new logos at an average first-year contract value of about $78K, on about $12M of new-logo sales and marketing spend: a blended CAC of about $133K. At the old 81% gross margin, payback would be about 25 months; at today’s 72% it’s about 28. Both exceed the roughly 20 months Skok allows for enterprise land-and-expand businesses, and the agentic margin drop alone added about three months without anyone in sales or marketing doing anything differently.

The blended number hides three different businesses:

Cross-sell pays back fastest. The MSP motion pays back slowly on the first contract but expands fastest, because each partner adds tenants. Direct enterprise is the largest and slowest. The cross-sell result also sharpens a standing tension: Beacon’s cheapest growth comes from Northlight Observe customers, which pulls toward building on Observe data and away from staying tool-agnostic.

The journey funnel. You map the direct and cross-sell journey with sales and customer success. Priya asks an AI assistant how to cut alert noise, reads Beacon’s docs and a community thread, and starts a proof of value on one environment. Activation is the first correlated incident; an account becomes a PQA when three or more responders accept a Beacon diagnosis. Then Sam runs a security review, Marcus builds the ROI case for his CFO and procurement closes.

Over the past year, about two thirds of 210 proofs of value reached a first correlated incident within two weeks, and accounts that became PQAs converted at more than twice the rate of those that didn’t. The median deal then spent 74 days in Sam’s security review and 19 in Marcus’s ROI case. Deals rarely died in the trial; they waited in security review, where Sam’s team asked the same questions about least privilege, audit logs, data residency and AI risk documentation every time. That’s the binding constraint, and more paid demand would only lengthen the queue.

The attribution assessment. Last-touch reporting credits event registrations and branded search. First-touch reporting credits documentation, assistant referrals and Observe prompts. Self-reported attribution adds colleagues and communities. Each view shows part of the journey and rewards different teams. None alone establishes which activity caused conversion.

The channel experiment. Most of the proposed increase would fund a paid social and webinar program aimed at about 600 target accounts. You propose an account-level holdout: randomly withhold 20% of them (about 120) for two quarters and compare opportunity creation rates, with the metric and decision rule registered in advance. With so few accounts only a large effect will be detectable, so you plan to read the result alongside self-reported attribution.

The answer engine audit. The harness runs a fixed set of buyer questions (“best tool to find the root cause of production incidents,” “AIOps for managed service providers”) across the major AI assistants each month. One assistant describes Beacon as autonomous by default. That’s wrong (actions run behind a human approval gate, and only low-risk actions a customer allowlists run on their own), and it’s exactly the claim that alarms reviewers like Sam. The fix is clearer public documentation of approval gates, audit logs and allowlists.

The memo. Your data-driven strategy memo recommends holding paid demand flat until the holdout reads out, and moving the proposed increase toward three things:

• Northlight Observe cross-sell and MSP expansion, where payback or expansion is strongest.
• A self-serve trust center (security documentation, audit log samples, data residency details and AI risk documentation) to shorten Sam’s review, with median days in security review as the success metric.
• Continuing the move to per-resolved-incident pricing, which ties price to the value delivered and to the inference cost that drives margin (Chapter 15).

It names the evidence that would reverse the call: a clear lift in opportunity creation from the holdout, or a security-review median that doesn’t fall after the trust center launches.

Product 360 artifacts

# Growth metrics scorecard · [Product] · [Period] · v[n]

**Owner:** [PM]   **Definitions agreed with:** [marketing ops, sales ops, finance]   **Last refreshed:** [Date]

## Definitions (one line each, with source system)
- CAC: [blended or paid only; fully loaded or not; includes partner margin?] · Source: [system]
- Gross margin used: [%, by product line] · Source: [finance]
- LTV: [formula; lifetime cap in years; expansion treatment] · Source:
- Payback: CAC / (new monthly revenue per customer x gross margin) · Source:
- Activation: [event that counts] · PQL / PQA: [threshold] · Source:

## Scorecard by motion and segment
| Motion / segment | New logos | First-year ACV | CAC (fully loaded) | Gross margin | Payback (months) | LTV:CAC | NRR | Activation rate | PQL or PQA rate | Win rate | Median cycle (days) |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Direct enterprise | | | | | | | | | | | |
| Partner / MSP | | | | | | | | | | | |
| Self-serve | | | | | | | | | | | |
| Cross-sell | | | | | | | | | | | |
| **Blended** | | | | | | | | | | | |

## Reference points
LTV:CAC above 3 and payback under 12 months (Skok); up to about 20 months for enterprise land-and-expand. Note any motion far outside these and why.

## What changed since last period
| Metric | Motion | Change | Cause (evidence) | Action |
|---|---|---|---|---|
| | | | | |

## Known gaps
[Missing data, definition disputes, motions too small to read]

# Journey funnel model · [Product] · [Segment] · [Date]

**Owner:** [PM]   **Built with:** [marketing, sales, customer success]   **Unit:** account / user

| Stage | Buyer question | Persona | Touchpoints (owned, paid, earned, partner, product, AI answer) | Entry definition | Leading metric | Conversion to next | Median days in stage | Observable? (yes / partly / no) | Owner | Data source and blind spots |
|---|---|---|---|---|---|---|---|---|---|---|
| Problem framing | | | | | | | | | | |
| Exploration | | | | | | | | | | |
| Trial / proof of value | | | | | | | | | | |
| Activation | | | | | | | | | | |
| Validation (security, ROI) | | | | | | | | | | |
| Purchase | | | | | | | | | | |
| Expansion and advocacy | | | | | | | | | | |

## How we estimate the unobservable stages
- Self-reported attribution question: [wording, where asked]
- Win-loss interviews: [cadence]
- AI answer audit: [cadence, question set]

## Binding constraint
Stage: [ ]   Evidence: [conversion, wait time]   Product hypothesis to fix it: [ ]   Metric to watch: [ ]

# Attribution assessment · [Report or dashboard] · [Date]

**Owner:** [PM]   **Report owner:** [name, team]   **Model in use:** first touch / last touch / linear / time decay / position-based / data-driven / other

## The decision this report supports
[Budget allocation, campaign tuning, channel reviews, comp plans]

## Bias review
| Question | Answer |
|---|---|
| What does this model over-credit? | |
| What does it under-credit? | |
| Which touches can't it see? (peers, community, AI answers, events, partner) | |
| Is the unit right? (person versus account) | |
| Does any platform self-attribute here? Double counting risk? | |
| Is targeting creating selection bias? | |

## Cross-checks
| Method | Available? | What it says | Agrees with the report? |
|---|---|---|---|
| Experiment or holdout | | | |
| MMM (calibrated?) | | | |
| Self-reported attribution | | | |
| Win-loss interviews | | | |

## Verdict
Fit for: [ ]   Not fit for: [ ]   Change needed: [ ]   Review date: [ ]

# Channel experiment plan · [Channel or program] · [Date]

**Owner:** [PM or marketing lead]   **Analyst:** [name]   **Decision maker:** [name]

## Hypothesis
We believe [program] causes [outcome] in [segment], and we'll know we're right if [metric] is at least [x] higher in the treatment group than in the holdout.

## Design
- Unit of randomization: user / account / geo
- Population: [list, size]
- Treatment: [what they get]   Holdout: [share, what they don't get]
- Randomization method and who runs it:
- Duration: [weeks or quarters, with the reason]

## Metrics
- Primary (pre-registered): [ ]
- Guardrails: [ ]
- Minimum detectable effect at this sample size: [ ]

## Decision rule (written before launch)
- If lift ≥ [x]: [scale / keep]
- If no detectable lift: [cut / redesign / combine with other evidence]
- If inconclusive: [what we'll do, and what extra evidence we'll use]

## Threats and mitigations
[Contamination between groups, small sample, seasonality, sales outreach to holdout accounts]

## Readout
Date: [ ]   Result with interval: [ ]   Decision: [ ]   Logged in: [decision log]

# AI answer engine audit · [Product] · [Month]

**Owner:** [PM or product marketing]   **Assistants checked:** [list]   **Run by:** [harness, date]

## Fixed buyer question set
| # | Buyer question (as a buyer would ask it) | Persona | Journey stage |
|---|---|---|---|
| 1 | | | |

## Results
| # | Assistant | Product mentioned? | Position in answer | Accurate? | Errors or risky claims | Sources the answer drew on |
|---|---|---|---|---|---|---|
| | | | | | | |

## Share of answer
[Mentions / questions asked, by assistant, against last month]

## Corrections
| Error | Likely source | Fix (docs, reviews, public content) | Owner | Recheck date |
|---|---|---|---|---|
| | | | | |

# [Decision] · Data-driven strategy memo · [Date]

**Author:** [PM]   **Decision maker:** [name]   **Decision needed by:** [Date]

## The decision we're asking for
[One or two sentences.]

## Context and numbers
[The three to five figures that matter, each with definition and source. Link to the scorecard and funnel model.]

## Options
| Option | Expected incremental impact (range) | Cost | Payback effect | Main risk |
|---|---|---|---|---|
| A | | | | |
| B | | | | |
| Do nothing | | | | |

## Recommendation
[Option, and why the evidence supports it.]

## What the data can't see
[Unobservable stages, attribution limits, small samples, open definitions.]

## What would change our mind
[The experiment result or metric movement that reverses this call, and by when we'll know.]

## Review date
[Date, forum]

Prompts

Goal: Get CAC, payback and LTV by motion and segment, with every number traced to a source.
Paste in (or connect the harness to): closed-won deals for the period, sales and marketing cost by motion, gross margin by product line, and your written metric definitions. Use only tools approved for CRM and financial data.

Using the data provided for [product] over [period], build a growth metrics scorecard with one row per motion ([direct, partner, self-serve, cross-sell]) and segment ([list]).

For each row compute: new logos, first-year ACV, fully loaded CAC (include partner margin or commission in partner CAC), gross margin, CAC payback in months, LTV with a [5]-year lifetime cap, LTV:CAC and NRR.

Rules:
- State the definition, denominator, time window and source system for every metric.
- Show the calculation for payback and LTV for one row so I can check it.
- If expansion makes the churn-based LTV formula meaningless, say so and use the cap.
- Flag any motion with fewer than [20] deals as too small to read reliably.
- Mark missing inputs as "no data" and list the source needed.

Check the output: recompute payback for one motion by hand from the figures shown. Confirm the new-logo count matches the CRM report finance uses.

Goal: Put numbers on each stage of the buying journey and find where value is lost.
Paste in: stage definitions, opportunity and trial data with stage timestamps, product activation events, and any self-reported attribution answers. Use an approved tool for customer data.

Build a journey funnel model for [product] in [segment], with the account as the unit. Use these stages: [list].

For each stage give: the entry definition I provided, the number of accounts that entered, conversion to the next stage, median and 80th-percentile days in stage, and whether the stage is observable in this data (yes, partly or no).

Then:
1. Name the binding constraint: the stage with the largest loss of accounts or the longest wait, and show the evidence.
2. Compare accounts that reached [PQL or PQA threshold] with those that didn't on later-stage conversion.
3. List which stages this data can't see and how I could estimate them.
4. Suggest two product hypotheses that might relieve the constraint, labeled as hypotheses.

Check the output: confirm that the entry definitions match what you supplied and haven't been reinterpreted. Spot-check five accounts' stage dates against the CRM.

Goal: Know what an attribution report can and can't support before money moves on it.
Paste in: the attribution report or dashboard export, the name of the model it uses, and the decision it's being used for. Pipeline data is confidential, so use an approved tool.

Act as a skeptical measurement analyst. For this attribution report on [product]:

1. Name the attribution model and explain in two sentences how it assigns credit.
2. List what it over-credits and what it under-credits for a B2B buying cycle of [length] with a buying group of [size].
3. List the touches it can't see (peer referrals, community, AI assistants, events, partners) and estimate how material each might be, labeled as a judgment.
4. Check for double counting across platforms and for person-level credit applied to account-level purchases.
5. Say which decisions the report is fit for and which it isn't, and what cross-check (experiment, calibrated MMM, self-reported attribution) would settle the largest open question.

Describe attribution as observed touches. State which causal claims require experimental evidence.

Check the output: if the model claims a channel "drove" pipeline, ask it to restate the claim as "touched." If the restated claim no longer supports the decision, you've found the problem.

Goal: Plan a channel experiment that can give a causal answer at B2B volumes.
Paste in: the program description, the target account list size, historical opportunity creation rates for similar accounts and the budget at stake. Use an approved tool for account and pipeline data.

Design an account-level holdout experiment for [program] targeting [n] accounts in [segment].

Include: the hypothesis; randomization method (stratified by [size, region, existing customer]); holdout share and the reason; primary metric and guardrails; duration; the minimum detectable effect at this sample size, with the arithmetic shown; threats such as sales reps contacting holdout accounts; and a decision rule for a positive, null and inconclusive result.

Be honest about power. If the sample can't detect a plausible effect, say so and propose how to combine the test with other evidence (self-reported attribution, longer duration, a geo design).

Check the output: confirm the minimum detectable effect is computed from the baseline rate you supplied. Make sure the decision rule was written for the null case, not only the win case.

Goal: Measure share of answer and catch inaccurate AI descriptions before buyers repeat them.
Paste in: a fixed list of 15 to 30 buyer questions by persona and stage, your product's accurate positioning and key facts, and the list of competitors.

For each question in the list, ask [assistants available to you] as a buyer would and record: whether [product] is mentioned, its position in the answer, the competitors mentioned, whether the description of [product] is accurate against the facts provided, any risky or false claims, and the sources the answer cites.

Then summarize share of answer by assistant and by persona, list every inaccuracy with its likely source, and propose a documentation or content fix for each.

Run the questions exactly as written, with no product name added unless it's in the question.

Check the output: rerun three questions yourself and compare. Answers vary between runs, so treat a single month as a sample and watch the trend.

Goal: Turn the scorecard, funnel and experiment evidence into a decision a leader can make.
Paste in: the growth metrics scorecard, the journey funnel model, the attribution assessment, any experiment readouts and the decision being asked for. These hold confidential financials, so use an approved tool.

Draft a two-page strategy memo for [decision maker] on [decision].

Structure: the decision asked for; the three to five numbers that matter, each with its source; options including "do nothing," each with expected incremental impact as a range and its effect on payback; the recommendation; what the data can't see; the result that would change the recommendation; and the review date.

Rules:
- Use only numbers in the material provided. Cite the artifact each comes from.
- Separate measured effects (experiments) from correlations (attribution) in the wording.
- Preserve the ranges and their assumptions.

Check the output: trace every number back to the source artifact. Delete any sentence that claims a channel or feature "drove" an outcome without experimental evidence.

Anti-patterns

Checklist

• Every growth metric in the strategy has one written definition, an owner, a source system and a time window.
• CAC, payback, LTV and NRR are reported by motion and segment, with partner margin in partner CAC and payback recomputed when gross margin moves.
• The PQL or PQA is defined by experienced value, and its link to paid conversion has been checked.
• The journey funnel model has entry definitions, conversion, time in stage and observability for every stage, including AI assistant and peer stages.
• The binding constraint is named, with evidence, and has a product hypothesis attached.
• Every attribution report used for budget has an attribution assessment listing its biases and blind spots.
• A self-reported attribution question is in place at sign-up or first meeting.
• Large budget or product bets have a channel experiment plan with a pre-registered metric, minimum detectable effect and decision rule, and any MMM in use is calibrated against experiments.
• A monthly AI answer engine audit tracks share of answer and accuracy.
• The strategy memo states expected impact as a range, what the data can’t see and what would reverse the call.
• Every number in the memo traces to a reproducible source, and AI-generated figures have been spot-checked.