Predictive LTV
Predictive LTV modelling for B2B
Predictive LTV modelling scores prospect accounts on likely lifetime value using firmographic, usage and deal shape signals from your own CRM, then steers acquisition budget toward the segments the model says are worth the most, not the segments that convert most easily.
- Inputs
- CRM, billing and product usage data
- Output
- Account and segment level LTV score
- Cadence
- Model refresh monthly, review quarterly
What this covers
Most acquisition budget is allocated by cost per lead or cost per meeting. Neither number says anything about what the account is worth once it closes. Predictive LTV modelling exists to fix that gap: it scores prospects and segments on expected lifetime value using patterns from closed business, then feeds that score back into where spend actually goes.
This is a scoring and budget allocation exercise, not a forecasting exercise. It does not predict revenue for the board. It ranks segments so spend stops chasing the cheapest lead and starts chasing the account shape that expands, renews and does not churn in year one.
- Budget is allocated to whichever channel produces the most leads, regardless of what those leads become
- Sales spends equal time on a $5,000 logo and a $150,000 logo because nothing upstream told them the difference
- Expansion and churn data sits in a different system to acquisition data and the two never talk
- Segment definitions are built on job title and company size, not on what actually correlates with retained value
This is for you if
- You have at least 12 to 18 months of closed-won and churn data in your CRM or billing system
- You sell into more than one segment and suspect they are not equally valuable
- Marketing and sales already agree on what a qualified opportunity looks like
This is not for you if
- You have fewer than 50 to 100 closed deals to train against
- Your product has one flat price and no meaningful expansion or churn variance
- You need a lead scoring model for MQL routing rather than a value model
Verdict: If the data does not exist yet, the right first step is fixing CRM hygiene, not buying a model.
How the engagement runs
Model build flow
Data audit
CRM, billing, usage exports checked for completeness
Segment definition
Cohorts built on the fields that actually predict value
Score build
Model scores historical and open accounts
Budget mapping
Score tiers mapped to channel and spend decisions
Review cycle
Model refreshed as new closed data lands
- 1
Days 1 to 30
Pull closed-won, closed-lost and churn records for the last 12 to 24 months. Audit for missing fields, duplicate accounts and inconsistent stage definitions. Agree the value definition with finance: is it first year contract value, three year LTV, or gross margin adjusted LTV. Build the first version of the segment model against this history.
- 2
Days 31 to 60
Score the current open pipeline and the active prospect list against the model. Sit down with sales and marketing leadership to review which segments scored highest and whether that matches gut instinct. Where it does not match, dig into why before trusting the model. Set the first reallocation: which channels or lists get more budget, which get less.
- 3
Days 61 to 90
Run the reallocated budget for a full cycle. Track whether the accounts entering pipeline from high-score segments actually close at the rate and value the model predicted. Refresh the model with the new closed data. Document the scoring logic so it survives a change of tooling or a change of team.
What you get
Artefacts and rhythm
- ✓A written value definition agreed with finance before any scoring starts
- ✓The segment model itself, documented in plain language, not a black box
- ✓A scored list of current pipeline and target accounts, refreshed monthly
- ✓A budget reallocation recommendation tied to the score tiers
- ✓A monthly review meeting to check model output against actual closes
How it is measured
Segment value score
Score = (Predicted LTV x Win probability) / Cost to acquire
- Predicted LTV comes from the model, built on historical contract value, expansion and churn by segment
- Win probability is the segment's historical opportunity to closed-won rate
- Cost to acquire is fully loaded cost per opportunity for that segment's primary channel
Example: A segment with $60,000 predicted LTV, 22% win probability and $2,500 cost to acquire scores higher than a segment with $30,000 LTV, 35% win probability and $1,800 cost to acquire once the LTV gap is large enough.
| Segment | Predicted LTV | Win rate | Cost to acquire | Score |
|---|---|---|---|---|
| Enterprise | $180,000 | 18% | $6,000 | 5.4 |
| Mid-market | $54,000 | 24% | $2,200 | 5.9 |
| SMB | $14,000 | 31% | $650 | 6.7 |
What we will not do
- Build a model on fewer closed deals than the statistics can support and present it as reliable
- Promise a specific lift in pipeline value before the model has run a full cycle against real outcomes
- Replace your CRM or billing system to make the data cleaner; we work with what exists
- Build a model and hand it over without the plain language documentation of how it scores
Objections, answered
The strongest case against this
Our sales team already knows which accounts are worth more. Why do we need a model for that.
Sales intuition is usually right about which named accounts matter. It is much weaker at the segment level, where the pattern is buried across hundreds of deals nobody reviews in aggregate. The model does not replace the intuition, it tests it against the full data set and finds where the two disagree.
The strongest case against this
We do not have clean enough data for this.
Most companies do not, at first. The data audit in the first 30 days will tell you exactly what is missing and how much it matters. Sometimes the answer is to fix CRM hygiene for two quarters before a model is worth building. That is a legitimate outcome of the audit, not a failure of it.
The strongest case against this
Win rates keep falling. Does that make the model useless.
It makes the model more necessary, not less. Falling win rates mean every opportunity worked has to be worth more to hit the same revenue target, which is exactly the variable this model is built to sort on.
Common questions
- How much historical data do we need before this is worth doing?
- At least 50 to 100 closed-won deals with consistent stage and value fields, covering 12 to 18 months. Below that, the model has too little signal to separate real patterns from noise, and the honest recommendation is to fix data hygiene first.
- Does this replace lead scoring?
- No. Lead scoring predicts whether a prospect will convert to a qualified opportunity. This predicts what the account is worth once it closes. The two can run together: lead score decides who gets worked first, LTV score decides how much budget the segment they came from deserves.
- What counts as lifetime value in the model?
- Whatever finance agrees it should be before the model is built: first year contract value, three year value, or gross margin adjusted value. This gets written down in the first 30 days so the model and the finance team are scoring the same thing.
- Can this work if we only have one product tier?
- It works less well. If price is flat and there is no meaningful variance in expansion or churn across segments, there is little for the model to sort on. This service fits companies where segment value genuinely diverges.
- How often does the model get refreshed?
- Monthly as a default, so new closed and churned deals keep updating the score. The underlying segment definitions get reviewed quarterly, since those should move slower than the score itself.
- Who owns the model once it is built?
- You do. The scoring logic is documented in plain language, not left inside a proprietary tool, so it survives a change of CRM, a change of agency or a change of internal team.
- Does this change how sales is compensated?
- That is a decision for you, not something this engagement sets. Some teams do adjust quota credit once segment value becomes visible, but that is a compensation conversation the model informs, it does not run.
- What happens if the model disagrees with what sales believes?
- That disagreement gets investigated before any budget moves, not overridden automatically in either direction. Sometimes the model is missing a variable sales knows about. Sometimes sales is anchored on a handful of loud accounts. Either way it gets checked, not assumed.
- How does this connect to the wider budget conversation?
- The score feeds directly into the pipeline model and the channel mix decisions in demand generation. It is one input into where budget goes, alongside CAC payback and coverage targets, not a standalone number sitting outside the plan.
Read the thinking behind it
Guides
Want this run for you?
Tell us what you are spending and where the pipeline stalls. If this is not the right first step for you, we will say so.