Lead scoring and next-best-offer
A sales floor can work a few thousand leads before the season starts. This model scores a simulated pipeline of 14,040 prospects twice — the probability they buy, and the probability a rep contact changes that — then assigns each one a product to sell. Ranking that pipeline by likelihood to buy and ranking it by incremental effect produce different call lists, and the two lists are worth different amounts.
Targeting strategy
Campaign
Rep outreach capacity
1,200 leads
How many leads inside sales can personally work before the season opens.
Cost per contact
$35
Loaded rep time plus any incentive attached to the call.
Pipeline
First purchase during the run · 4.1% baseline · +2.9 pts lift
Two or more buys last season · 3.4% baseline · +0.7 pts lift
12-game holders, plan-eligible · 40.0% baseline · +2.7 pts lift
Named contacts off manifests · 0.5% baseline · +2.4 pts lift
Asked for a callback · 14.3% baseline · +3.8 pts lift
Coefficients
Log-odds weights. The response column predicts buying without outreach; the uplift column is what a contact adds on top.
3 weights flip sign between the models, marked in red. Fans who read every email and already hold a plan are the most likely to buy and the least moved by a call. Group-night contacts are the reverse: they almost never buy on their own, and the phone is the only channel that reaches them.
Incremental plan revenue · by uplift
$579,454
above what the same leads would have produced untouched, from 1,200 calls costing $42,000
1,200
Leads contacted
99
Incremental conversions
5.0x
Margin return on $42,000
What the ranking is worth
Same pipeline, same 1,200 calls, same $35 per contact. Only the sort order changes.
By propensity
45 conversions · 3.5x margin return
$412,975
By uplift
99 conversions · 5.0x margin return
$579,454
By expected value
92 conversions · 5.5x margin return
$645,499
Uplift ranking beats propensity ranking by $166,479 at this capacity. Expected-value ranking adds a further $66,045 by sorting on uplift times plan value, at the cost of 8 conversions: the same calls go to fewer, larger accounts.
Baseline against uplift
Every 17th lead in the pipeline. Horizontal is the probability they buy with no contact; vertical is what the contact adds. Red points are the leads the current strategy calls.
The four groups in the pipeline
Cuts at a 25% baseline and 4 points of uplift. Value is what contacting the whole group would add — or, below zero, remove.
Persuadables
uplift ≥ 4 pts
3,437
avg baseline 9.4% · avg uplift +6.0 pts · $922,020
Contact is what moves them. Rep hours here create revenue that would not otherwise arrive.
1,200 in the current call list
Sure things
baseline ≥ 25%, uplift < 4 pts
247
avg baseline 40.4% · avg uplift +1.9 pts · $47,120
They convert through the portal on their own. A call bills time against revenue already coming.
0 in the current call list
Lost causes
baseline < 25%, uplift < 4 pts
8,851
avg baseline 3.4% · avg uplift +1.9 pts · $463,983
They move a little, but from a low base — the smallest expected gain per lead of the four groups.
0 in the current call list
Sleeping dogs
uplift < 0
1,505
avg baseline 14.1% · avg uplift -1.6 pts · -$121,333
Contact lowers their probability of buying. Working this group subtracts revenue.
0 in the current call list
Next-best-offer mix in the call list
Each lead is matched to the largest product its behavior supports — attendance, plan history, engagement, distance. Bars are incremental revenue by product across the leads being called.
Full season
178 leads · $3,530/seat · 41 games
$179,613
Mini plan
692 leads · $1,030/seat · 12 games
$155,471
Premium club
64 leads · $6,400/seat · 41 games, club level
$122,471
Half season
199 leads · $1,810/seat · 21 games
$101,211
Group plan
67 leads · $184/seat · 4 nights, per seat
$20,687
The call list — top ten, by uplift
| Lead | Source | Games | Baseline | Uplift | Offer | Plan value | Expected |
|---|---|---|---|---|---|---|---|
| L-23891 | Waitlist6 mi · 45% email | 14 | 17.4% | +25.3 | Half season4 seats | $7,240 | $1,833 |
| L-23734 | Waitlist4 mi · 52% email | 13 | 18.1% | +23.6 | Half season4 seats | $7,240 | $1,712 |
| L-23695 | Waitlist4 mi · 60% email | 15 | 21.9% | +22.2 | Full season2 seats | $7,060 | $1,567 |
| L-23934 | Waitlist11 mi · 47% email | 14 | 16.3% | +21.5 | Half season4 seats | $7,240 | $1,558 |
| L-22576 | Mini-plan4 mi · 38% email | 15 | 43.3% | +21.0 | Full season4 seats | $14,120 | $2,966 |
| L-22164 | Mini-plan3 mi · 36% email | 17 | 48.2% | +21.0 | Full season4 seats | $14,120 | $2,964 |
| L-22531 | Mini-plan4 mi · 35% email | 14 | 37.8% | +20.7 | Full season4 seats | $14,120 | $2,919 |
| L-22389 | Mini-plan13 mi · 31% email | 16 | 40.7% | +20.6 | Full season4 seats | $14,120 | $2,910 |
| L-23512 | Waitlist9 mi · 61% email | 14 | 20.1% | +20.6 | Half season2 seats | $3,620 | $745 |
| L-23573 | Waitlist8 mi · 58% email | 11 | 16.4% | +20.0 | Half season4 seats | $7,240 | $1,450 |
Expected is uplift times plan value: the revenue attributable to making the call, not the revenue the account would produce. A list ranked on the second number credits reps with accounts that were closing either way.
Reading the model
- 01
Ranking the same 1,200 calls by uplift instead of propensity is worth $166,479 of incremental plan revenue — 40% more for identical rep hours and identical cost.
- 02
A propensity list spends 57% of capacity outside the persuadable group: 247 sure things who convert through the portal anyway, 357 whose probability drops when contacted, and 76 who were never going to buy. That is $23,800 of the $42,000 program spent on outcomes the call does not change.
- 03
1,505 leads in the pipeline have negative uplift, averaging -1.6 points; working all of them would remove $121,333 of expected revenue. At $35 a contact and a 36% margin, a lead has to carry $97 of incremental revenue to pay for its own call — the 3,437 persuadables average $268, the 8,851 lost causes $52.
- 04
Full season carries the most incremental revenue in this call list — 178 of the 1,200 leads and 31% of the total. The offer attached to each lead comes from what that lead already did, not from the largest product they could theoretically afford.
Method
Each lead carries eight features: games attended last season, weeks since last purchase, a playoff-cohort flag, distance from the arena, group versus individual origin, email engagement rate, prior plan history, and party size. Two logistic models run over them. The response model gives P(convert) with no outreach. The treatment model adds a second set of weights to the same log-odds and gives P(convert) after a rep contact. Uplift is the difference, which is why it behaves differently from either score alone: attendance raises both, engagement and plan history raise the first and cut the second, and group origin cuts the first and raises the second. Games are capped at 20 in the response model and 14 in the uplift model, and party size at 6 seats, so one 40-person group manifest cannot dominate a score.
Next-best-offer is a commitment index — attendance, plus a bonus for holding a plan already, plus engagement, minus distance — mapped onto the product ladder and priced at the suite’s per-seat values, which put a two-seat full season at $7,060 — in line with the average account on the existing season-ticket book. Program return is computed on the contribution margin of incremental revenue (36%, the rate the renewal model uses) rather than the gross plan price, since crediting a $35 call with the full face value of a season ticket would flatter the program.
The caveat worth stating plainly is that uplift is never observed for an individual: no lead is both called and not called. Production requires a randomized holdout in every campaign, with the model fit as a two-model or transformed-outcome estimator on those experiments. Without a holdout, both models are fit on the same observational data and the gap between them is an assumption rather than a measurement.
The pipeline is simulated (seeded, stable across visits) and scaled to the audiences used elsewhere in this suite: 9,500 playoff-cohort new buyers, 3,100 mini-plan holders, and a 28,000-account single-game base. In production these leads would assemble from ticketing, CRM, and email platforms, with consent and contact-frequency rules applied before anything reaches a rep queue, and the quadrant cuts set by sales leadership rather than by the modeler.