Buy vs. build

Should you build your own customer engagement stack? Here's what it really takes across agentic lifecycle messaging and in-app personalization.

Lifecycle agent

Where building in-house makes the least sense

Wiring an LLM to your email provider takes an afternoon. Deciding what to send every user, and proving it works, is an ML, data and safety problem.

Next-best-action decisions

Every user's agent decides whether to send, when, on which channel and what to say, working toward the objectives and weighted outcomes you set.
Build

Models to predict send propensity, timing, channel preference and likely response for every user, plus the feature pipelines, training jobs and serving infrastructure behind them.

Typically needs ML and data engineers you would otherwise hire for your core product.

Why does it matter?
Getting a working prototype is easy. Getting predictions good enough to beat a simple journey takes far more data, iteration and specialist skill than most teams expect, and a bad model doesn't just underperform, it sends the wrong message to real users.

Training data

Agents start from patterns learned across consumer apps, then learn from how each of your users responds to every message.
Build
Your models only learn from your own history, which is usually a handful of journeys, sent to broad segments, with outcomes nobody recorded at the user level.
Why does it matter?
Good next-best-action models need lots of examples of different messages sent to different users at different moments, and what happened next. Most companies have never generated that data, so an in-house model starts cold and stays there for months.

Data quality and access

Agents read the events and attributes you send to Trophy, plus everything Trophy's in-app experiences already know about each user: streaks, progress, rankings and more.

Build
Product data lives in your warehouse, your app database and your analytics tool, each with different identifiers, latency and gaps. Someone has to clean, join and stream it to wherever decisions get made.
Why does it matter?
An agent is only as good as what it can see. Most teams don't have clean, real-time, per-user product data in one place, and the marketing team that owns messaging often can't access it at all.

Writing on brand, at scale

Agents write every message inside your brand voice, guidelines and guardrails. Add a rule once and every agent follows it from then on.
Build

Prompting, retrieval of user context, output validation, guardrail checks, retries and human review for every message an LLM writes.

An evaluation suite to catch off-brand copy, false claims and hallucinated offers before users see them.

Why does it matter?
One off-brand or inaccurate message to a single user can become a screenshot. Keeping AI-written copy on brand needs layered checks around every generation, not a better prompt, and those checks need maintaining every time the underlying models change.

Model costs

Usage-based pricing on monthly active users. Model costs are included, and agents only spend effort on users and moments worth reaching.
Build
Token costs grow with every user, every decision and every draft, multiplied by the extra calls you need for planning, checking and rewriting copy within guardrails.
Why does it matter?
Running capable models for every user, every day gets expensive fast, and costs are hard to forecast. Trophy spreads that work across many customers and routes each step to the cheapest model that can do it well.

Delivery and channels

Email from your own domain and push through your own FCM, APNs or Expo setup, with per-user sending limits and preferences built in.
Build
Integrations with an email provider and push services, domain authentication, bounce and unsubscribe handling, frequency caps and per-user preferences.
Why does it matter?
A great message is worthless if it lands in spam or comes one too many times in a week. Deliverability and fatigue control are their own discipline, and an agent that ignores them burns your list.

Proving it works

A holdout group that never sees agent messaging, with lift reported on every objective over time and by channel.
Build
Randomized holdouts, incrementality analysis and reporting built and maintained by your data team, separate from opens and clicks.
Why does it matter?
Without a clean holdout you can't tell whether messaging drove a result or just took credit for it. Lift, not engagement, is what justifies the investment, and it's the first thing a CFO will ask for.

Keeping up

Trophy's agents improve as models, techniques and best practice move on, without any work from your team.
Build
New models, new provider APIs and new failure modes arrive every few months. Each one means re-testing prompts, evals and guardrails.
Why does it matter?
AI moves faster than any internal roadmap. Building in-house means committing a team to keep pace indefinitely, for something that is not your core product.

In-app experiences

Simple to sketch, hard to run

Streaks, points and leaderboards look like a few database tables. In production they're high-frequency systems full of edge cases.

Integration time

1-3 days
Build
2-3 months for a first version
Why does it matter?
Every experience has its own logic, edge cases and UI states. With Trophy you send events and focus on the user experience instead of the plumbing.

Edge cases

Time zones, streak freezes and restores, leaderboard resets, ties, level-ups, retroactive changes and multi-tenancy, handled and tested.

Build
Every edge case is a bug report waiting to happen. A user who loses a 300-day streak because of a time zone change won't wait for next sprint.
Why does it matter?
These features are built on users' sense of fairness and progress. Getting them wrong does more damage to trust than not having them at all.

Reliability

99.9% uptime SLA. Powering experiences for 1M+ users.

Build
Processing every user action in real time raises problems with atomicity, idempotency, retries and race conditions.
Why does it matter?
Double-counted points or a missed streak day is visible to the user immediately. Reliable event processing is hard to get right, and harder to retrofit.

Scalability

Built to handle hundreds of millions of events a day and scale with you.
Build
Each step change in usage means infrastructure reviews, optimizations and more engineering time. Real-time leaderboards are a common first bottleneck.
Why does it matter?
With Trophy we handle optimization and scaling so your team can stay focused on the user experience.

Connection to messaging

Every in-app experience feeds the lifecycle agent automatically, so a near-miss on a streak or a new rank becomes a reason to reach out.
Build
Separate systems for in-app experiences and messaging, joined together with webhooks, syncs and journey logic someone has to keep in step.
Why does it matter?
The best messages reference what a user just did or is about to lose. When in-app experiences and messaging live in one platform, that context is there by default.

The whole stack

The total cost of ownserhip

Building one piece is a project. Owning the whole stack is a permanent commitment.

People required

Initial integration: 1 developer

Ongoing: 0 dedicated engineers

Build

Initial build: 3-6 engineers across product, data and ML

Ongoing: 2-3 to maintain, tune and extend

Why does it matter?
Customer engagement isn't the core of your business. Every engineer working on it is one not working on the product your users came for.

Tools to buy anyway

Lifecycle agent, in-app experiences, analytics, experimentation and holdouts in one platform.
Build
An LLM provider, an email provider, push services, a feature store, an experimentation tool and an analytics stack, each with its own contract and integration.
Why does it matter?
Building in-house rarely means building everything. It usually means stitching together half a dozen vendors and owning the glue between them.

Cost

Usage-based pricing on monthly active users, with zero upfront investment. Trophy only charges for active users, so you never pay for churn.

Build
High upfront investment in salaries and infrastructure, then ongoing costs for model usage, tooling, monitoring and support.
Why does it matter?
With Trophy you only pay for what you use, and benefit from our economies of scale on infrastructure and model costs.

Support and maintenance

24/7 support with dedicated Slack channels for customers who need it, focused on best practice, not fixing bugs.
Build
On-call for messaging and event pipelines, a backlog of user-reported issues, and ongoing patching and upgrades.
Why does it matter?
Building in-house adds a permanent maintenance cost that competes with your core roadmap, forever.

New capabilities

New features ship to Trophy regularly, shaped by our vision and what dozens of consumer apps need.

Build
Every new channel, experience or objective is another project to scope, staff and maintain.
Why does it matter?
We're building Trophy to be the customer engagement platform for consumer web and mobile apps. Whatever your users or your team ask for next, it's likely we've already thought about it.