Next Best Action instead of mass campaigns: what changes

The cost of mass campaigning

“Mass SMS campaigns cost nothing” is a phrase taken as fact. They cost more than they look.

The direct cost is the SMS itself, the operational load and the loss of customers who unsubscribe. These numbers can be calculated. They are small per campaign, but they accumulate.

The indirect cost is more significant. Every irrelevant communication lowers open rate and conversion of future messages. If a customer received five irrelevant campaigns, they stop reading SMS from the operator at all. When the operator finally sends an important message — about a SIM swap, about a critical operation — it can be ignored. This is a real cost that does not show up as a separate line in P&L but affects the effectiveness of all future communications.

Another indirect cost is brand perception. An operator that sends irrelevant offer SMS is perceived as “selling at me” rather than “helping me”. This affects NPS, retention and indirectly the customer’s willingness to buy new products.

NBA — Next Best Action — is the alternative. Instead of a mass campaign, the system selects for a specific customer, in a specific moment, a specific action. And that action does not have to be an offer — it can be a proactive notification, help, information.

What goes into NBA

NBA is not an ML model. NBA is an operational loop made of six components.

Catalogue of actions. A list of concrete actions the system can offer. Each action has a target audience, channel, content, business goal, success metric. Not “general marketing campaigns”, but concrete actions described.

Eligibility rules. For each action — who can receive it. Not “everyone”, but specifically. These rules can be simple (if ARPU above X and traffic pattern Y) or complex (multi-criteria scoring).

Suppression rules. Who must NOT receive the action. A customer who recently complained. A customer in a blackout window after a previous action. A customer with an opt-out. A customer flagged for fraud risk.

Ranking. When several actions are applicable, how to choose one. Ranking can be simple (priority order) or complex (a model predicting conversion for each).

Frequency caps. Not more than X actions per day, Y per week, Z per month. These limits prevent spam.

Audit and learning. What happened after each action. Conversion. Reactions. Used to improve the model and the rules.

Without any one of these six, NBA does not work. Most NBA projects focus on the model (ranking) and forget the other five. This explains why 70% of NBA projects stay in pilot.

What separates a working NBA from a model

A working NBA has several characteristics that pilots usually lack.

The action library is finite. Not “the model will decide what to do”, but “the model picks from 30 defined actions”. The library needs maintenance — adding new actions, retiring underperformers, adjusting content. Discipline that is often missing.

Eligibility is not just a data filter. It includes business judgement. “This customer technically qualifies for the offer, but they are in a billing dispute right now — do not send”. This requires human input into the rules.

Suppression is a separate contract with the customer experience team. The customer should not receive conflicting communications (a retention call from the contact centre simultaneously with a marketing push). This requires cross-functional coordination.

Frequency caps are not just “no more than X per day”. The cap must be segmented. A premium customer can take more frequent actions; a low-value customer once every two weeks. Without segmentation the cap is either too tight (opportunities are lost) or too loose (fatigue grows).

What most often breaks NBA

Marketing dominating customer care. When NBA is owned by marketing, the action library tilts toward sales and promo. Customer care actions (proactive notifications, help, retention contact) are under-represented. The customer feels NBA is spam in different packaging.

No action ownership. An action is added to the library, but nobody is responsible for its performance. Six months later there are 50 actions in the library, 20 of them underperforming, nobody cleans them up. The library turns into junk.

Modelling without operational integration. The ML team built a ranking model, but integration with the decision engine was not done, or events do not arrive in real time. The model predicts into the air.

Push in one channel. NBA sends everything through one channel (usually push or SMS). It does not account for different segments having different preferences. A premium customer may expect a call or an email, not a push.

No experimentation culture. NBA is effective when continuously tested — A/B tests on content, timing, ranking. If every change needs committee approval, iteration slows to the point where NBA is no better than a rule-based campaign.

A realistic first year for NBA

NBA cannot be built in a quarter. A realistic plan is one year.

First 90 days. Foundation. A catalogue of 10-15 base actions with full descriptions. Eligibility rules per action. Suppression rules — global and per-action. Frequency caps by segment. Audit log infrastructure.

Next 90 days. Pilot. NBA active on one segment (for example, premium postpaid). Daily ranking. 5-7 channels orchestrated. Daily review across marketing, customer care, finance. A measurement framework.

Next 90 days. Iteration. Underperforming actions removed. 5-10 new actions added based on identified gaps. Rule refinement. Model upgrade if actually needed. Expansion to a second segment.

Last 90 days. Scaling. Coverage expanded to 50%+ of the active base. Multi-channel orchestration. Continuous experimentation. Linked with the network experience and retention processes.

After a year a properly built NBA delivers a 30-100% conversion lift over baseline mass campaigns. This is not a single project — it is a rebuild of commercial work.

When NBA is premature

If the action library has fewer than 8-10 thoughtful options, building NBA does not make sense. First gather the library through a rule-based system, then add the model.

If master data does not reconcile between billing, CRM and the app, eligibility will be inaccurate, NBA will misfire, the team’s trust in the system will collapse.

If the contact centre and retention act independently of marketing and there is no shared dashboard, NBA produces conflicting communications with the same customer.

If consent management has no granularity (one general opt-in or nothing), customer preferences cannot be respected correctly.

If a regular operating routine does not form (no weekly review across marketing, sales, customer care), NBA quickly degrades — actions without owners, frequency caps without enforcement, suppression rules forgotten.

Discussion points for the committee

How many campaigns has the operator run last quarter and what was the economics? If the answer is hard to give — there is no baseline, NBA is too early.

Which 5 actions work best and which 5 worst? If the team does not know, NBA will not be an improvement, it will be a process rebuild.

What is the current NBA stack or orchestration system? If nothing — start small. If there is a legacy campaign system — refactoring.

Who will own NBA as a product? Not “marketing” — a specific person with authority to change the library.

How SamaraliSoft can help

Next Best Action / Offer Engine Blueprint — design of the action catalogue for a specific operator, design of eligibility, suppression and frequency caps based on the real customer base, choice of NBA architecture (build vs buy vs hybrid), pilot on one segment with a measurement framework, and a 12-month scaling plan.

Related reading

• /en/insights/telecom-subscriber-intelligence-operating-model/ — operating model around data
• /en/use-cases/telecom-churn-war-room-mnp/ — retention in the MNP era
• /en/insights/telecom-ai-native/ — where AI pays back
• /en/architecture/telecom-around-core-architecture/ — the growth layer

Sources

• Digital Government Portal — digital infrastructure UZ
• UzTelecom — example user interface
• Central Bank of Uzbekistan — biometric verification for digital banking
• UzDaily — telecom market