Introducing the Expert: Lucas Perez, Senior Data Scientist at LedgerFlow

Lucas Perez works at LedgerFlow, a mid-market accounting SaaS platform with 38,000 monthly active users. He’s led multiple product analytics initiatives, including two activation overhaul projects and a systematic feature-adoption study. We asked Lucas to break down what connected product strategies really mean for mid-level data-science teams, especially for those seeking to drive data-informed decisions in accounting SaaS.

Why do “connected product strategies” matter for accounting SaaS? Where does data science fit?

Connected product strategies usually mean making sure every piece of your platform—from onboarding to reporting—feeds into a single understanding of the customer journey. For accounting SaaS, the challenge is that users have wildly different needs. Your data science team sits in the middle: you’re the ones segmenting users, measuring what leads to activation (not just acquisition), and eventually, tying feature usage to retention.

Without a strong data strategy, product decisions are often made on gut feeling. In accounting software, that’s risky. One poor onboarding tweak can cause a spike in churn or support tickets. Data science teams are the last line of defense, translating noisy clickstream data or NPS feedback into evidence for what’s actually working.

Where do you start? What data should teams focus on first?

Skip “vanity metrics” like raw signups. Start by defining activation metrics. For us, that’s “completed first invoice” or “reconciled first bank transaction.” These are the actions that actually predict long-term engagement. Instrument those events, then work backwards to see what steps block or accelerate them.

We mine product usage logs for sequences—what do users who activate in under 24 hours do differently? For example, we saw users who skipped the onboarding walkthrough took 2.8x longer to send their first invoice. That informed a change to make walkthroughs opt-out, not opt-in.

Qualitative data is essential. We use Zigpoll and Survicate to run micro-surveys right after onboarding and after a month of use. This helps explain the “why” behind the numbers, especially when feature adoption drops after a release.

How does experimentation play out in product-led growth for accounting SaaS?

A/B testing is table stakes now. But many teams make the mistake of only testing UI changes. We test pricing prompts, default settings (e.g., enable bank sync by default), and even the wording of in-app nudges. Accounting users are risk-averse, so incremental changes work better than big-bang redesigns.

One experiment: We swapped the order of features in the onboarding checklist to put “add bank account” before “invite accountant.” That single change improved Day 7 retention by 6.9%. It also drove a 29% increase in bank syncs completed within the first week. The lesson: data guides what to test, but context—knowing accounting users’ trust issues—guides what’s worth testing.

What data models or analyses do you rely on most for feature adoption?

Funnel analysis is basic but ignored. We create detailed funnels for every core feature—who starts, who completes, and where they drop off. Then we use time-to-event analyses (e.g., survival analysis) to understand not just if users adopt a feature but when.

Cohort analysis is also vital. For example, users onboarded after a UI change in March 2024 showed a 3.2% higher activation rate versus the previous quarter (internal analysis). But after segmenting by company size, only businesses with more than four employees benefited—the rest actually activated less.

We also use clustering (k-means or hierarchical, depending on the feature) to identify usage patterns that correlate with upgrades or churn.

What are some common traps or mistakes you see data-science teams make here?

The biggest one: reporting metrics with no action path. If you’re only tracking NPS or weekly active users, you’re not informing product changes. Metrics have to be tied to interventions. Otherwise, you’re just a reporting function.

Another: overfitting to power users. For accounting SaaS, the “enthusiast” segment can skew data. We saw one cohort (high-frequency invoicers) driving up average engagement stats, but 44% of our churn came from “silent” users who never made it past onboarding. If you only analyze the engaged, you miss the leaky bucket.

Lastly, underestimating qualitative input. Sometimes, the numbers don’t explain a drop-off. That’s where tools like Zigpoll and Hotjar heatmaps help; you see users rage-clicking the “Import CSV” button or leaving confused feedback.

How do you collect and act on user feedback for onboarding and new features?

We fire a two-question survey (Zigpoll or Typeform) at key moments: after onboarding ends, after first feature use, and after support interactions. One quantitative question (“How easy was X?”), one open-ended (“What’s missing?”). We bucket responses by user type and activation status.

For feature launches, we also use in-app feedback widgets (like Sprig) to collect context-specific feedback. For example, after rolling out a new payroll export, we tagged 140 pieces of feedback in the first week. The top complaint—confusing default date ranges—led us to patch the UI in days, not weeks.

But feedback has to close the loop. We aggregate results weekly in a dashboard owned by product and data science. Each month, we identify one “quick win” and one “structural change” to address.

Can you share a real example where data-driven strategy changed product direction?

Last year, we saw that only 11% of trial users connected their bank accounts. Churn post-trial was 24% higher for those who didn’t. Initial hypotheses blamed the UI, but survey data (via Zigpoll) pointed to trust: users worried about security.

Instead of a UI revamp, we ran an experiment swapping the bank connection step with educational content and a security FAQ. After the change, bank connections during trials jumped from 11% to 23% in two months. Trial-to-paid conversion increased by 8.1%. This wasn’t just a design fix; it was a data-driven response to an emotional barrier.

What are the best tools or frameworks you’ve used for experimentation and analytics?

For event tracking, we use Amplitude to build funnels and user journeys. For experimentation, Split.io and Optimizely are reliable; Split is better for backend or feature-flag experiments. For feedback, as mentioned, Zigpoll and Survicate are both flexible. Survicate works well for embedded surveys in product flows, while Zigpoll is used for email follow-ups because of its higher completion rates.

For deeper cohort and churn modeling, we rely on internal Python tools—lifelines for survival analysis, scikit-learn for basic clustering. The combination of product analytics platforms and custom scripts covers almost everything mid-level teams need.

Comparison Table: Onboarding & Feedback Tools

What’s one commonly overlooked metric or approach?

Tracking “time to first value” instead of just “time to first action.” For instance, users may log in and poke around, but until they’ve sent an invoice or reconciled a bank feed, they haven’t gotten value. Measuring the lag between signup and true product value gives a clearer picture of onboarding effectiveness.

Also, “feature decay” is crucial. We track how often users revisit advanced features after first use. A quick spike followed by long-term drop-off signals poor retention or bad feature fit.

Are there limitations to this approach?

Absolutely. Quantitative data can only tell you what’s happening, not why. There’s also a lag—by the time you see churn increasing, it’s too late for that cohort. And, high experiment velocity can burn users out or create inconsistent experiences.

Furthermore, most SaaS approaches assume self-service users. For companies with required sales or onboarding calls, the activation journey gets muddier. The frameworks here don’t directly apply—manual interventions skew the signals.

What advice would you give to mid-level data scientists pushing connected strategies?

Own the problem, not just the dashboard. Position your analyses in ways that ask: “What should we do next?” not just “What happened?”

Push for instrumentation that covers the full journey—trial, onboarding, first value, upgrade, and (crucially) re-engagement. Build relationships with support and product marketing; feedback from those teams fills gaps in your data.

Don’t be afraid to kill “pet” features. If your funnel or cohort analysis shows a feature is used by less than 2% of users, question if it needs to exist. Every extra button drags onboarding and increases confusion.

Finally, treat connected product strategies as iterative. The best teams see metrics rise and fall, but always know what’s causing the movement. That’s the difference between reporting and decision-making.

Final Thoughts: The Future of Connected Product in Accounting SaaS

A 2024 Forrester report found that SaaS companies tying onboarding metrics directly to activation saw 23% stronger retention at six months. Data-driven connected product strategy isn’t about dashboards; it’s about wiring product, support, and feedback into a single loop. For mid-level data scientists, the opportunity is to move from supporting role to strategic voice—provided you’re willing to connect the dots, not just count them.