COMPARE/Category comparison
Recommendation engine vs. next-best action
+17–35% ARPU against holdoutObserved range across Markin deployments, measured on treated cohorts.
A recommendation engine predicts which item or content a user will engage with and ranks by relevance. Next-best action selects the commercial treatment, an offer, a save, a channel, a hold, that maximises expected incremental revenue for a sized opportunity. A rec engine answers 'what should we show'; an NBA answers 'what is worth doing, and what is it worth'. The two are complementary, not substitutes.
What is at stake
A decision layer is not a tool line item. It moves ARPU on the whole base, every month.
Installed base
5.0M
customers at $14 ARPU / month
Addressable revenue
$462.0M
per year, reachable base
Verified ARPU uplift
+17% to +35% ARPU
on treated cohorts, against holdout
What that is worth
$78.5M – $161.7M
incremental revenue per year
Measured on treated cohorts against a randomised holdout, read over a full measurement window rather than the first weeks. Anonymised range across Markin deployments in large B2C bases; your own holdout is the number that decides. The figures above apply that range to the reachable share of the base on this page's assumptions; they are arithmetic, not a forecast for your business.
Run it on your own numbersWhat each one actually does.
01
Recommendation engine
A system that predicts and ranks the items, content or products a user is most likely to engage with, using collaborative filtering, content models or embeddings.
02
Next-best action
A layer that sizes the commercial opportunity for a customer and selects the treatment, message, offer, channel, timing or hold, that maximises expected incremental revenue.
Side by side
The differences that change outcomes.
| Dimension | Recommendation engine | Next-best action |
|---|---|---|
| Question it answers | What should we surface to this user next? | Which action is worth taking for this customer, and what is it worth? |
| Primary input | Item features, interaction history, user embeddings, catalogue metadata. | Opportunities, uplift models, margins, costs, contact history, constraints. |
| Primary output | A ranked list of items by predicted engagement or relevance. | One chosen action per customer, including a deliberate hold. |
| Usual owner | Product, data science. | Growth, data science, revenue leadership. |
| How it's measured | Click-through rate, take rate, watch time, relevance. | Incremental revenue and ARPU against a holdout. |
The unsolved part
Why a great recommendation can leave revenue on the table
Recommendation engines optimise engagement within a fixed catalogue. Relevance is not value: the most relevant next item is rarely the highest-value commercial action, and a rec engine has no concept of margin, cost or when not to act.
- A rec engine ranks items by predicted engagement; it attaches no revenue value, margin or cost to the action behind the recommendation.
- It cannot surface a commercial opportunity nobody modelled, an upgrade, a save, a reactivation, because its output space is the catalogue.
- It cannot decide to hold: there is always a next-best item, even when no item has positive expected value.
- Its impact is measured on take rate and engagement, not on incremental revenue against a control group.
The actual difference
Markin is not another decisioning engine.
Markin is not a decisioning engine. A decisioning engine ranks actions a human already defined. Markin works like a data science and growth team: it forms its own hypotheses about why ARPU is stuck, marketing, product, pricing or technical, sizes them, executes them inside the systems you already run, and reads each one against a holdout.
| A decisioning engine | Markin | |
|---|---|---|
| Where the hypothesis comes from | A human authors it. The engine chooses between options someone already approved. | Markin authors it. It reads the base, finds where revenue is leaking or unclaimed, and writes the hypothesis itself. |
| What it is allowed to question | Message, offer, channel, timing, inside the campaign surface it was given. | Anything that moves ARPU: onboarding friction, pricing and packaging, a feature nobody adopts, a payment failure spike, a broken deeplink. |
| Who does the analysis | Your analysts, before and after. The engine optimises; it does not investigate. | Markin does the analysis. Sizing, segment definition, experiment design and readout are automated end to end. |
| Where it stops | At the recommendation. Someone still has to build and launch it. | It launches. Markin executes inside your existing platforms and product surfaces, then closes the loop on the result. |
| Throughput | As many hypotheses as your roadmap has room for, typically a handful per quarter. | Hundreds in parallel, every one carrying a control group. |
| What happens when it is wrong | The programme keeps running until someone reviews it. | It is retired automatically. Failing to beat control is a normal, cheap outcome. |
A decisioning engine picks the best action from a list you wrote. Markin writes the list, and runs it in your stack.
Hypothesis space
Everything a human growth scientist would look at.
Most growth problems are not message problems. Markin is not restricted to the campaign surface: if something is holding ARPU back, it is in scope, and it gets tested the same way.
Marketing
The classic surface, but chosen per customer rather than per segment, and always against a holdout.
- Which offer this specific customer is worth making
- Channel and timing chosen per person, not per campaign
- Contact pressure and fatigue arbitrated across every programme
- Win-back economics: who is worth a discount and who is not
Product
Where the customer actually experiences the value, and where most silent revenue loss happens.
- Onboarding steps that lose customers before first value
- A feature with high retention correlation that half the base never discovers
- Paywall and upgrade prompt placement
- In-product surfaces used as a treatment arm, not just email and push
Commercial
Pricing, packaging and the shape of the offer itself, tested rather than argued about.
- Plan and bundle structure by cohort
- Discount depth against margin, not against conversion alone
- Annual versus monthly framing per customer
- Dunning and involuntary churn recovery sequences
Technical health
Anomalies nobody asked it to look for. This is the category no decisioning engine covers.
- A checkout error rate that rose on one device and one region
- Payment failures concentrated in a single issuer or method
- A broken deeplink quietly killing a high-value journey
- Latency or delivery degradation eating conversion before any message does
Think of Markin as a data science and growth team that never sleeps: it investigates, forms hypotheses, ships them into your own stack and proves each one against a control group, at a volume no human team can reach.
The loop
Execution is a step in the loop, not a hand-off.
- 01
Observe
Markin reads the behavioural, transactional and product signal you already collect, continuously.
- 02
Hypothesise
It writes the hypothesis itself, marketing, product, commercial or technical, and states the expected direction.
- 03
Size
Each opportunity is ranked by expected value, so the queue is ordered by money rather than by opinion.
- 04
Design
Segment, treatment, guardrails and a randomised holdout are set before anything ships.
- 05
Execute
It launches inside the systems you already run, your engagement platform, your product surfaces, your APIs. Nothing waits on a build queue.
- 06
Read
Results are measured against the holdout over a full window, so novelty is not mistaken for effect.
- 07
Scale or retire
What beats control is scaled across the base. What does not is switched off automatically.
Where Markin fits
Not a replacement. A growth-science team on top.
Markin treats the recommendation as one candidate action among many. The intelligence layer sizes the revenue opportunity behind it, the action layer chooses whether to recommend, to offer, to save or to hold, and the whole loop is measured against control.
Recommendations become a candidate, not the answer
A rec engine's output is one input to the decision; the action layer compares it against a save, an upgrade or a hold and picks the highest expected incremental value.
Ranked by revenue, not relevance
Actions are selected on expected incremental revenue net of margin and cost, so the highest-value move wins the customer's attention even when it is not the most relevant item.
Hold is a first-class output
When no action has positive expected value, the decision is to do nothing, something a rec engine, which always returns a next item, cannot express.
Operating model
The constraint is not ideas. It is how many you can test.
| Today, with recommendation engine | With Markin on top | |
|---|---|---|
| Revenue hypotheses tested per quarter | 4 to 8, whatever the roadmap had room for | Hundreds, generated and run in parallel |
| What can be hypothesised about | Messages, offers and audiences, the campaign surface | Marketing, product, pricing and technical health alike |
| From decision to live in the channel | A ticket, a build queue, a release window | Markin launches it in your existing platforms itself |
| Time from idea to a result you trust | 6 to 10 weeks of analysis, build and readout | Days, because sizing and design are automated |
| Share of decisions with a control group | The flagship programmes, when there is time | Every decision, by default |
| Coverage of the base | Top segments and the customers a rule caught | One decision per customer, across the whole base |
| Cost of testing the 500th hypothesis | Another analyst, another quarter | Effectively zero |
| What the team spends its time on | Pulling data, building lists, reconciling reports | Judgement: constraints, economics, what to scale |
Markin does not replace your data science team. It removes the ceiling on how much of the base that team can act on, and how fast it finds out whether it worked.
Evidence standard
Most of this category reports its own lift.
None of the major engagement, CDP or personalisation vendors publishes an independently verified uplift figure for its decisioning product. Where numbers exist, they come from vendor-commissioned studies or single-customer case studies with no disclosed holdout methodology. The most rigorous public research in the category is not flattering to anyone, including us, which is exactly why we build against it.
BCG reports that when organisations adopt rigorous incrementality testing, they typically find 20% to 40% of their active next-best-action programmes deliver marginal to negative lift.
Independent researchBCG, How Measurement Is Evolving in Next-Best Action (2026)The same research flags novelty effects, new programmes show inflated early results, and recommends 8 to 12 weeks before drawing conclusions.
Independent researchBCG, How Measurement Is Evolving in Next-Best Action (2026)Global-holdout, programme-level ROI measurements often overstate impact through halo effects, pull-forward effects and experiment contamination.
Independent researchBCG, How Measurement Is Evolving in Next-Best Action (2026)
How Markin holds itself to it
- Every decision Markin makes carries a control group. Uplift is reported against that holdout, not against the customers who did not qualify.
- Results are read over a full measurement window rather than in the first weeks, so novelty is not mistaken for effect.
- Programmes that fail to beat control are retired automatically. Killing decisions that do not pay is part of the loop, not an annual review.
- The one figure we quote about ourselves is a range, not an average: +17% to +35% ARPU on treated cohorts against a randomised holdout, across Markin deployments in large B2C bases. We publish no industry benchmark, because we could not source one we would be willing to defend. Your holdout is the number that matters.
Size it yourself
What a recommendation engine is worth once it has to earn revenue
Preloaded for a content or streaming business running recommendations at scale: a rec engine drives engagement and watch time. The figure below is the incremental margin available from deciding the commercial action behind the recommendation, a save, an upgrade, a hold, rather than the next item to surface, measured against a holdout rather than a take rate.
Your base
Accounts that generated revenue in the last 30 days. Not registered users.
Recurring plus non-recurring revenue divided by active customers.
Margin on the next unit sold, not blended company margin.
Your programme today
Consented, non-fatigued, reachable on at least one channel.
Revenue lost to cancellations each month, as a share of the base.
The bet
Licences, data, incentives and the people running it.
Before any incrementality haircut. 2–4% is a defensible planning assumption.
Verified annual impact
$4.2M
Net incremental gross margin in the central case, after the programme cost and after the share of decisioning programmes that independent research finds deliver no real lift.
Reported uplift
$13.9M
What a before/after dashboard would claim, with no control group.
Verified uplift
$9.7M
What survives a holdout in the central case.
Return on programme cost
4.9×
Payback
3 mo
If 20–40% of it does nothing
What it takes to prove it
To detect a 3% lift on revenue per customer you need roughly 40K customers in the control arm, about 1.5% of your addressable base, read over at least 8 weeks, so novelty is not mistaken for effect.
Addressable base
2.8M
Revenue at risk from churn
$325.3M
Annualised, at the current monthly rate.
Time to value
90 days to a number that survived a holdout.
No replatform, no data migration, no rebuild of the channels you already run. If the first cohorts do not beat control, nothing scales and you have lost a quarter, not a roadmap.
Weeks 0–2
Read the context you already have
Markin connects to the data and the channels you run today, recommendation engine included. No migration, no replatform, no new source of truth.
Weeks 3–6
First sized opportunities in test
Opportunities are ranked by expected value, treatments are chosen per customer, and the first cohorts go live with a randomised holdout attached.
Weeks 7–12
First verified incremental revenue
Results are read over a full measurement window. What beats control scales; what does not is retired. Nothing scales on a number that has not survived a holdout.
When you don’t need Markin.
- Your only commercial lever is surfacing content, and engagement is the business model: there is no treatment selection problem to solve.
- Your catalogue has no commercial action behind it, so there is nothing to size in revenue terms.
- You cannot connect a recommendation to an outcome, so the loop cannot be validated against control.
Questions buyers ask.
Is a next-best action system just a recommendation engine?
No. A recommendation engine predicts the next item a user will engage with. A next-best action system sizes the commercial opportunity for a customer and chooses the treatment, offer, save, channel, hold, that maximises expected incremental revenue. The rec engine answers 'what to show'; the NBA answers 'what to do, and what is it worth'.
Can a recommendation engine do retention?
It can recommend content that keeps a user engaged, which indirectly helps retention. It cannot decide who is worth saving, with which treatment, at what cost, or who to leave alone, those are commercial decisions that need uplift, margin and cost, not relevance.
Do the two compete?
No. They are complementary. A recommendation engine is a strong candidate-action source; the next-best action layer decides whether to recommend, to offer something commercial, or to hold, and proves the choice against a holdout.
How is Markin different from the decisioning or AI already inside recommendation engine?
A decisioning engine ranks actions a human already defined, inside the campaign surface it was given. Markin forms the hypotheses itself, marketing, product, pricing or a technical anomaly holding growth back, sizes them, executes them inside recommendation engine and your product surfaces, and reads each one against a randomised holdout. It behaves like a data science and growth team, not like an optimiser.
Does Markin only test messages and offers?
No. Anything a human growth scientist would investigate is in scope: onboarding friction, feature adoption, pricing and packaging, dunning, and technical health issues such as a checkout error rate or a broken deeplink quietly killing conversion. Marketing is one of four hypothesis domains, not the boundary.
What is the business case for adding Markin on top of recommendation engine?
On the assumptions preloaded above, 5.0M customers at 14 a month, a small move in ARPU is a large number in absolute terms, because it applies to the whole installed base every month rather than to a campaign. Across Markin deployments the verified range on treated cohorts is +17% to +35% ARPU against a randomised holdout. The point is not more messages: it is finding the highest-value action per customer, launching it, and proving it against control before it scales.
How long before it pays for itself?
First sized opportunities are in test within six weeks and the first holdout-verified result lands inside 90 days. Payback depends on your base, margin and programme cost, the calculator on this page computes it from your own numbers, after applying the 20% to 40% haircut BCG finds when next-best-action programmes are incrementality-tested.