02 · Portfolio & Multi-Product Strategy¶
With one product you prioritise. With three you allocate — and the two are different disciplines. Prioritisation asks which item is most valuable. Allocation asks where the marginal engineer produces the most, which is a question about diminishing returns, not about which idea is best.
This is the arithmetic that decides whether a multi-product company compounds or fragments. Get it wrong in the obvious direction and you starve a young, fast-growing line to protect a mature one that is already throwing off cash and cannot absorb the investment. Get it wrong in the fashionable direction and you underfund the core that pays for everything, and discover eighteen months later that your $27M line stopped growing while you were busy.
The portfolio at ListUp¶
| Line | ARR | YoY growth | Engineers | R&D cost | ARR per engineer | R&D as % of line |
|---|---|---|---|---|---|---|
| Publish | $27,200,000 | 19% | 14 | $2,730,000 | $1,942,857 | 10.0% |
| Decide | $14,700,000 | 71% | 15 | $2,925,000 | $980,000 | 19.9% |
| Console | $10,500,000 | 62% | 13 | $2,535,000 | $807,692 | 24.1% |
| Platform | — | — | 10 | $1,950,000 | — | 3.7% of total ARR |
| Total | $52,400,000 | 37.9% | 52 | $10,140,000 |
The immediate temptation on seeing this table is to move engineers from
Publish (cheap to run, low growth) to Decide and Console (expensive,
fast-growing). Sometimes right. But ARR per engineer is a historical
ratio, not a marginal one, and it says nothing about what the next engineer
would produce. Publish's 19% growth on $27.2M contributes $5,168,000 of
new ARR next year — more than Console's 62% on a smaller base contributes at
$6,510,000, and not far off it. Percentage growth flatters small lines.
Three lenses, none sufficient alone¶
| Framework | What it asks | Use it for | Where it misleads |
|---|---|---|---|
| Three horizons (H1 core / H2 emerging / H3 bets) | How mature is each line? | Communicating a balanced portfolio | Encourages treating H3 as a permanent hobby with no bar |
| 70/20/10 allocation | What proportion goes to core, adjacent, new? | A starting split when you have no data | It is a heuristic, not a finding — do not defend it as analysis |
| Growth/share matrix | Where do we lead, in what markets? | Deciding where to compete | Market share is unknowable in small or undefined categories |
| Marginal ROI | What does the next engineer produce, per line? | The actual allocation decision | Requires evidence you have to go and collect |
The first three are communication tools. The fourth is the decision tool, and it is the one nobody does because it requires estimating something uncomfortable.
Marginal ROI: the number that should drive allocation¶
For each line, estimate what one additional engineer, sustained for a year, would produce in incremental ARR. The estimate must come from evidence, not enthusiasm — the best source is what happened the last time you added or removed people from that line.
| Line | Evidence used | Marginal ARR per engineer | Confidence |
|---|---|---|---|
| Publish | Last 4 engineers added went to reliability and channel coverage; two new channels added $1.4M combined over 18 months | $180,000 | High — well-measured |
| Decide | Attach rose 15.5% → 34% over 2 years as the squad grew from 6 to 15; each acceptance-rate point is worth ~$310k in attach | $620,000 | Medium — confounded with the agency channel |
| Console | Every major agency workflow shipped has moved agency NRR; the last two added $2.1M with 4 engineers over 12 months | $540,000 | Medium |
| Platform | Cuts cycle time across 42 line engineers; a 2.5% throughput gain ≈ 1 engineer-equivalent per 40 | Indirect | Low — must be argued on flow, not ARR |
Three disciplines make this credible rather than a wish list:
- Show the evidence column. A marginal ROI with no basis is a preference wearing a number.
- Show the confidence column. Decide's $620,000 is confounded with the agency channel's growth; some of that ARR would have arrived anyway.
- Assume diminishing returns. These are estimates for the next few engineers, not for twenty. Re-estimate at every planning cycle, and expect the number to fall as a line matures.
The multi-product effect on retention¶
The strongest argument for a portfolio is usually not revenue per line. It is what happens to retention when a customer uses more than one thing.
ListUp's 6,150 direct accounts ($18.2M of the $52.4M; the remaining $34.2M comes from 214 agencies at an average of $159,873 each):
| Product combination | Accounts | Share | ARPA (annual) | ARR | Net revenue retention |
|---|---|---|---|---|---|
| Publish only | 3,210 | 52.2% | $1,560 | $5,007,600 | 97% |
| Publish + Decide | 1,640 | 26.7% | $3,480 | $5,707,200 | 116% |
| Publish + Console | 520 | 8.5% | $4,020 | $2,090,400 | 121% |
| All three | 780 | 12.7% | $6,900 | $5,382,000 | 134% |
| Total / weighted | 6,150 | $2,957 | $18,187,200 | 116.7% |
Two things follow immediately:
- 52.2% of accounts are single-product and retain below 100%. They are losing revenue every year. That population is the single largest identified opportunity in the company.
- Moving 10% of Publish-only accounts (615) to Publish + Decide is worth 615 × ($3,480 − $1,560) = $1,180,800, or 2.3% of total ARR — and it moves those accounts from 97% to 116% NRR, which compounds every year afterwards.
The mandatory caveat, which separates a portfolio strategy from a correlation deck: accounts that buy a second product may simply be better accounts. The honest test is a matched comparison — take accounts of similar size, tenure and channel count, and compare NRR before and after attach. Until that exists, present the number as the size of the prize if the relationship is causal, and say so out loud.
Allocation decision — ListUp's next year¶
The question: 52 engineers today, 8 more approved for next year. Where do the 60 go?
Option A — proportional to growth. Fund fast-growing lines most: Publish 12, Decide 19, Console 18, Platform 11.
Option B — proportional to marginal ROI. Decide and Console take everything: Publish 12, Decide 21, Console 17, Platform 10.
Option C — the chosen allocation:
| Line | Now | Next year | Δ | Rationale |
|---|---|---|---|---|
| Publish | 14 | 13 | −1 | Reliability is met; channel coverage is competitive. Below 13, incident load rises — a floor, not an optimum |
| Decide | 15 | 19 | +4 | Highest marginal ROI, and it is the reason the other two lines retain |
| Console | 13 | 16 | +3 | Highest NRR combination; agency revenue is 65% of the company |
| Platform | 10 | 12 | +2 | Cross-cutting; funded on flow evidence, not ARR |
| Total | 52 | 60 | +8 |
Why not pure marginal ROI (Option B):
| Reason | Detail |
|---|---|
| Floors exist | Publish below 13 engineers pushes incident load onto the group that carries 52% of revenue |
| Marginal estimates decay | Decide's $620k assumes the next 4 engineers behave like the last 4. Going to +6 stretches an already-confounded estimate |
| Platform has no ARR line | Funding strictly by attributable ARR guarantees platform is never funded, then everything slows |
| Portfolios need optionality | Zero allocation to new bets means no H3 pipeline, and the current lines will mature |
The modelled result, at the marginal ROI estimates and each line's existing base:
| Line | Base ARR | Base growth | Δ engineers | Marginal contribution | Modelled next-year ARR |
|---|---|---|---|---|---|
| Publish | $27,200,000 | $5,168,000 | −1 | −$180,000 | $32,188,000 |
| Decide | $14,700,000 | $10,437,000 | +4 | +$2,480,000 | $27,617,000 |
| Console | $10,500,000 | $6,510,000 | +3 | +$1,620,000 | $18,630,000 |
| Total | $52,400,000 | $22,115,000 | +8 | +$3,920,000 | $78,435,000 |
That models 49.7% growth, well above the 38% plan. Two corrections before anyone takes it to a board: growth rates decelerate as bases grow (Decide at 71% on $14.7M will not hold 71% on $25M), and marginal ROI estimates are optimistic by construction. The planning number is 38%, with this model shown as the upside case and the deceleration assumption stated explicitly. A leader who presents the unadjusted model once loses the room permanently.
When to launch a second product¶
| Test | Pass condition | ListUp when Decide launched |
|---|---|---|
| Core is healthy | Core growth durable, NRR above 100%, no existential competitive threat | Publish at 41% growth, NRR 108% |
| Same customer | Sold to the same buyer through the same channel | Yes — identical buyer |
| Real demand evidence | Customers already paying someone else for it | 31% paying $89/mo elsewhere (Level 3, Module 1) |
| Shared advantage | The new product is better because of the first | 4.1M labelled price changes |
| Funded properly | A dedicated team, not a side project | 6 engineers, one PM, ring-fenced |
| Kill criterion | A number and a date agreed at launch | Under 12% attach at 18 months → stop |
Failing any single one is usually fatal. Failing "funded properly" is the most common: second products staffed from spare capacity almost never reach escape velocity, because they are the first thing cut whenever the core has a bad quarter.
When to kill a line¶
| Signal | Threshold | Action |
|---|---|---|
| Growth below company average for 4 quarters | And no strategic reason | Cap investment; harvest |
| Attach stalled below the kill criterion | Missed the pre-agreed date | Sunset or spin down |
| Negative contribution after fully loaded costs | 2 consecutive years | Kill |
| It exists to serve one customer | Any | Convert to services, or exit |
| Blocks the core roadmap | Persistently | Kill even if profitable |
Killing well is a skill. Announce internally before externally, give customers 12 months and a migration path, redeploy the team visibly and generously, and write down what the bet taught you. Teams watch how you retire a failed product far more closely than how you launch a successful one, and it sets the price of taking a risk next time.
Portfolio governance¶
| Cadence | Review | Decision it can force |
|---|---|---|
| Quarterly | Line-level metrics: ARR, growth, NRR, attach, marginal ROI re-estimate | Shift up to 10% of capacity |
| Half-yearly | Portfolio review with the exec team | Shift more than 10%; open or close a line |
| Annually | Allocation for the year, with the bridge (Level 3, Module 10) | Headcount plan |
| Continuous | Kill criteria on new bets | Automatic stop when a threshold is breached |
The most valuable rule is the last: kill criteria fire automatically. Something that must be argued for at the moment of failure will not be argued for, because by then everyone in the room is invested.
How It Actually Works: diminishing returns and the NRR compounding engine¶
Why marginal ROI, not average ROI, is the correct allocation variable.
Every line's output as a function of headcount roughly follows a
concave production function — each additional engineer adds less than the
last, because the highest-value work gets done first. If f(n) is a line's
ARR contribution with n engineers, ARR per engineer = f(n)/n is the
average product, while the allocation decision needs f(n+1) - f(n),
the marginal product. These diverge sharply once f is concave:
Publish's average is $1.94M/engineer (mature, most of f already
captured), but its marginal is $180k — the 15th engineer is nowhere near as
valuable as the 1st. Decide's average is lower ($980k) but its marginal is
higher ($620k), because it is earlier on its curve, with more high-value
work still undone. Allocating by the average metric routes investment
away from the steepest part of every line's curve; allocating by the
marginal metric routes it toward the steepest part. This is the same
logic as marginal utility in economics: rational allocation equalizes
marginal returns across uses, not average returns.
Why the confidence column matters mechanically, not just rhetorically. Decide's $620k marginal estimate was fit on a period where headcount rose 6→15 and the agency channel grew simultaneously — two moving variables explaining one outcome (attach going 15.5%→34%). Without holding the channel effect constant, the regression of "engineers added" against "ARR gained" has an omitted-variable bias: some of the coefficient on headcount is actually capturing the channel's own growth. Reported confidence (High/Medium/Low) is a proxy for how many other explanatory variables moved during the measurement window — the more that moved together, the wider the true error bars around the point estimate, even though the arithmetic ($620k) looks precise to the dollar.
The NRR blended-rate mechanism. Net revenue retention compounds
multiplicatively, not additively: an account at 116% NRR is worth
1.16^t times its starting ARR after t years with zero new sales,
against 0.97^t for a 97%-retaining single-product account. Over 5 years
that is a 2.10x multiplier on the 116% cohort versus 0.86x on the
97% cohort — a 2.4x gap in terminal value from the retention rate alone,
before counting a single new sale. This is exactly why "move 10% of
single-product accounts to two products" is analyzed as an NRR-shift
problem rather than a one-time upsell: the $1,180,800 immediate ARR gain is
the smaller of the two effects. The larger, uncounted effect is that those
615 accounts now compound at a blended rate close to 116% instead of 97%
for as long as they stay customers — the real prize is the change in the
exponent, not the change in year-one revenue.
The kill-criterion mechanism is about removing a sunk-cost decision
node, not about being strict. A threshold set before launch is
evaluated by a rule (attach < 12% at month 18 → stop); a threshold
argued at the moment of failure is evaluated by the people who chose the
bet, staffed the team, and told the board it would work — the same
cognitive-dissonance and sunk-cost mechanisms covered in Level 2's
prioritization bias material apply here at the portfolio level. Automating
the trigger converts an emotionally loaded judgment call into a
pre-committed fact-check, which is the only way kill criteria actually
fire on schedule instead of being renegotiated indefinitely.
Exercise¶
- Build the portfolio table for your company: ARR, growth, headcount, cost, ARR per head and R&D as a percentage of each line. If a line has no ARR, state what it is funded on instead.
- Compute dollar growth, not percentage growth, per line. Note which line looks best on each measure, and which one you would have starved by reading only percentages.
- Estimate marginal ROI per line with an evidence column and a confidence column. Where the evidence is weak, say what you would measure in the next two quarters to strengthen it.
- Build the attach table: accounts by product combination, ARPA and NRR for each. Compute the value of moving 10% of your single-product accounts to two products.
- State the confound in that number, and design the matched comparison that would settle it.
- Produce three allocation options — proportional to growth, to marginal ROI, and your recommendation — and write the specific reasons the pure-ROI option is wrong, including any floors.
- Model the outcome of your recommendation, then apply a deceleration assumption and state the planning number separately from the upside case.
- Score your newest bet against the six launch tests, and write its kill criterion as a number and a date if it doesn't have one.
- Write the sunset plan for your weakest line as if the decision were already made: internal announcement, customer notice period, migration path, and where the team goes.