Plaid Online Assessment: 90-Minute OA, Dedup & Categorization Tasks

Last updated: October 2026

Quick Answer

The Plaid online assessment is reported as a timed coding test of about 90 minutes with two practical problems on financial data, such as categorizing bank transactions and detecting duplicate transactions across sources that disagree on format. The algorithms are easy; the messy data is the test. Drill it in OphyAI Coding Interview on Premium, and rehearse explaining your edge cases in OphyAI Interview Practice.

Part	What candidates and guides report
Where it sits	After the recruiter call, before the live CoderPad screen
Length	About 90 minutes, taken on your own schedule (reported)
Problems	Two, themed on transactions
Typical tasks	Categorization, duplicate detection, recurring charges
Platform	Not named in any source we found; ask your recruiter
Scored on	Correct, clean code that survives dirty input

What Plaid Is Testing in This Round

Plaid normalizes data from thousands of banks that format it differently, so the assessment checks whether you write code that holds up when the input is wrong. Guides report that clean, well-structured code matters more than algorithmic cleverness.

Red flags that fail candidates:

• Solving the happy path and never handling null merchants, mixed date formats or negative amounts.
• Comparing money as floats instead of integer cents.
• Matching duplicates on exact equality when timestamps drift by seconds.
• One giant function with no helpers, so a second rule means a rewrite.

The Plaid interview guide covers the full loop and why this stage cuts most candidates.

What to Expect

Plaid has not published the assessment format, so treat everything here as candidate-reported and confirm with your recruiter.

• Format. A self-scheduled, timed coding session with two problems.
• Length. About 90 minutes in third-party guides.
• Data. Lists of transaction records: amount, date or timestamp, merchant string, sometimes a source or account ID.
• Language. Reports do not mention a required language; pick the one you parse data fastest in.
• Next step. Passing leads to a 45-90 minute live screen on CoderPad where requirements are added as you go.

Example Questions and How to Answer Them

These match reported themes, not a leaked list.

1. “Merge transactions from two aggregators and drop duplicates where merchant and amount match but timestamps differ by a few seconds.”

Approach: Normalize first: trim and lowercase merchants, convert amounts to integer cents, parse timestamps to UTC. Then group by (merchant, amount) and, within a group sorted by time, treat records inside a tolerance window as one. Say what you key on: a bank transaction ID when it exists, otherwise a composite of amount, date and normalized description. Ask whether two genuine identical purchases a few seconds apart are possible.

2. “Find recurring transactions.”

Approach: One anonymous LeetCode post describes a rule of at least 3 transactions with the same company, amount and spacing. Group by normalized merchant, sort by date, compute intervals, and accept a group when intervals and amounts stay within a tolerance. Keep the tolerance as a parameter, because the follow-up is usually “now allow amounts to vary by 5%“.

3. “Categorize these transactions using a set of rules.”

Approach: Build an ordered rule table (merchant keyword, amount sign, category) and apply the first match, with an explicit “uncategorized” fallback. Plaid’s product does this at scale with its personal finance categories, so mention that real systems version their taxonomy, as Plaid’s PFC migration guide shows.

4. “What happens when a field is missing?”

Approach: Decide per field: skip, default or flag. Write it down in a comment and test it. Reports say reviewers watch how your code reacts when a null merchant breaks the first version.

How to Prepare in 3 Days

Day 1: parsing drills. Write helpers to normalize money, dates and merchant strings. Test each against ugly samples.

Day 2: two timed problems. Do a dedup problem and a recurring-transactions problem in 90 minutes total, then add one new requirement to each without restarting. A timed run in OphyAI Coding Interview keeps you honest on the clock.

Day 3: explain it. Run a technical mock in OphyAI Interview Practice and talk through your edge cases, since the live screen that follows asks you to defend them. Start practicing →

Common Mistakes

Floats for money. Rounding errors turn equal amounts into non-matches.

Ignoring timezones. The same purchase can land on two dates.

No tests on bad rows. Hidden test cases reportedly target the mess.

Over-engineering. A clean, readable solution beats a clever one you cannot extend.

Related Guides

• Plaid interview guide
• Plaid system design interview
• Affirm HackerRank online assessment
• Fidelity HackerRank and Codility assessment

Frequently Asked Questions

What is on the Plaid online assessment?

Third-party guides and candidate reports describe two practical problems on transaction data, such as categorizing transactions, detecting duplicates across sources and finding recurring charges. The logic is simple but the input is deliberately messy. Plaid does not publish its question bank, so treat these as patterns rather than a list.

How long is the Plaid OA?

Third-party guides report about 90 minutes for two problems, taken on your own schedule. Plaid has not published the length or the scoring, and formats change by role and level. Check the invitation email and ask your recruiter before you start, so you can plan your time per problem.

Is the Plaid OA on HackerRank or CodeSignal?

None of the sources we checked name the platform for the online assessment; the one named tool is CoderPad for the later live screen. The invitation link will show the platform. Practice in a plain editor without autocomplete so either platform feels familiar on the day.

How hard is the Plaid online assessment?

Candidates describe the algorithms as easy to medium. The difficulty is handling inconsistent merchant strings, near-duplicate amounts and dates, missing fields and timezone drift. Candidates who write a fast happy-path solution reportedly score worse than slower candidates whose code survives every edge case.

Can I use AI tools during the Plaid assessment?

We found no published Plaid policy on candidate AI use, so read the assessment instructions and ask your recruiter. Our tracker of companies that allow AI in interviews lists what employers say publicly. You will also need to explain your code in the live screen that follows.

Sources and verification notes

Checked October 2026. The fetch tool could not reach the source sites from this environment, so they were read through search summaries. Plaid does not publish its assessment format; everything about the OA is candidate-reported or third-party.

• TechPrep, Plaid’s interview process 2026: reported 90-minute, two-problem OA. Third-party.
• techinterview.org, Plaid interview guide 2026: reported categorization and duplicate tasks, duplicate keys. Third-party.
• techinterview.org, Plaid software engineer interview: clean-code emphasis, CoderPad screen. Third-party.
• LeetCode discuss, Plaid phone interview: anonymous recurring-transactions report. Self-reported.
• Plaid docs, Transactions and PFC migration guide: recurring streams and versioned categories. First-party product docs.
• Plaid interview guide: the hub, with its own sources.