Design an enrichment and scoring workflow for a hiring signal — GTM Engineer take-home

GTM Engineer45 minutes

Context

Northbeam Ledger is an invented accounting-automation product for mid-sized property management companies. The team believes the strongest buying signal is a company posting for a "Staff Accountant" or "AP Specialist" while running 500 or more units. Every Monday, a list of about 300 raw job-posting rows lands in a table. The team wants each row turned into a scored, routable lead with no manual lookup. You may build this in Clay, n8n, a spreadsheet with scripts, or write it out as a design if you have no such tool. Nothing is sent to anyone.

Materials

This week's input table is attached as postings.csv: a 30-row sample of the roughly 300 rows that arrive each Monday. Columns: posting_id, job_title, company_name_raw, location, posted_date (the run date is 2026-10-05), description_snippet (shortened here). Check the rows for these known quirks: some company names are staffing agencies posting on behalf of a client; some postings are reposts with a new posting_id; company_name_raw has spelling variants of one company. Unit counts, when present, appear only inside description_snippet. Available enrichment (assume each works but may return blank): company domain lookup, unit count and employee count from a company-data provider, and a contact finder for finance titles.

Data

Download materials.csv (30 rows, 6 columns)

Columns: posting_id, job_title, company_name_raw, location, posted_date, description_snippet.

The table scrolls sideways, so every column is there.

posting_id	job_title	company_name_raw	location	posted_date	description_snippet
P-1021	Staff Accountant	Harbor Point Mgmt	Tampa FL	2026-10-01	Manage AP for 1800 units across 14 communities
P-1022	AP Specialist	Talentbridge Staffing	Remote	2026-10-04	Our client a leading property manager seeks an AP specialist
P-1023	Staff Accountant	Harbor Point Management LLC	Tampa FL	2026-09-26	Manage AP for 1800 units across 14 communities
P-1024	Staff Accountant	Cedar Row Residential	Austin TX	2026-10-02	Own month-end for a portfolio of 640 apartment units
P-1025	AP Specialist	Lakeshore Property Group	Chicago IL	2026-09-30	High-volume invoice processing for 2200 doors
P-1026	Senior Accountant	Brightwater Communities	Denver CO	2026-10-03	Lead GL close for 5 HOA-managed neighborhoods
P-1027	Staff Accountant	Apex Search Partners	Dallas TX	2026-10-01	Our client is hiring for a staff accountant role
P-1028	AP Specialist	Maple & Vine Mgmt Co.	Raleigh NC	2026-10-04	Process vendor invoices for 950 units and 3 commercial sites
P-1029	Property Accountant	Stonebridge Realty Mgmt	Phoenix AZ	2026-09-22	Property-level reporting for 310 units
P-1030	AP Specialist	Lakeshore Property Grp	Chicago IL	2026-10-03	High-volume invoice processing for 2200 doors
P-1031	Staff Accountant	Talentbridge Staffing	Remote	2026-10-02	Immediate need for a staff accountant at a property firm
P-1032	Accounts Payable Clerk	Pinehurst Apartments	Charlotte NC	2026-10-04	Enter and code invoices for 120 units
P-1033	Staff Accountant	Redhawk Property Services	Nashville TN	2026-10-01	Support AP and AR across 1500 units in 3 states
P-1034	AP Specialist	Summit Ridge Living	Salt Lake City UT	2026-09-28	Manage AP workflow; portfolio of 780 units
P-1035	Controller	Harbor Point Mgmt	Tampa FL	2026-10-02	Lead finance team for 1800 units
P-1036	Staff Accountant	Corvus Staffing Group	Atlanta GA	2026-10-03	Seeking accountants for multiple property clients
P-1037	AP Specialist	Willowbend Residential LLC	Columbus OH	2026-10-04	Handle 1100 units worth of vendor payments
P-1038	Staff Accountant	Cedar Row Residential	Austin TX	2026-09-18	Own month-end for a portfolio of 640 apartment units
P-1039	Staff Accountant	Eastgate Property Mgmt	Boston MA	2026-10-02	Manage books for 520 units and a small retail center
P-1040	AP Specialist	Northfield Management Inc	Minneapolis MN	2026-10-01	AP for 2600 units; 40 vendors per day
P-1041	Staff Accountant	Parkview Mgmt	Kansas City MO	2026-10-04	Property accounting; units not listed
P-1042	Billing Analyst	Lakeshore Property Group	Chicago IL	2026-10-03	Tenant billing support for 2200 doors
P-1043	AP Specialist	Bluehaven Estates	San Diego CA	2026-10-02	Process invoices for 430 units
P-1044	Staff Accountant	Kestrel Talent Solutions	Remote	2026-10-04	Our client a real estate operator needs a staff accountant
P-1045	Staff Accountant	Brightwater Communities	Denver CO	2026-10-04	Lead GL close for 5 HOA-managed neighborhoods
P-1046	AP Specialist	Gatehouse Residential	Orlando FL	2026-09-29	Own AP for 1400 units across 9 properties
P-1047	Staff Accountant	Maple and Vine Management	Raleigh NC	2026-09-27	Process vendor invoices for 950 units and 3 commercial sites
P-1048	Accountant	Sunbelt Rentals Mgmt	Houston TX	2026-10-03	Part of a 4-person finance team; 3000 units
P-1049	AP Specialist	Oakline Housing Partners	Portland OR	2026-10-01	Affordable housing AP; 700 units with grant reporting
P-1050	Staff Accountant	Harborview Apartments	Seattle WA	2026-10-04	Single-property accounting; 90 units

Showing the first 25 of 30 rows. Download the CSV for all rows.

Task

Design the workflow end to end. Build it if you have the tool, otherwise sketch it as a table and a diagram. Explain your decisions as you go.

Order of steps: say why that order protects your enrichment budget (what you filter before you enrich).

Columns: a column-by-column table of every column the workflow adds, with name, source, type, example value, and the decision it feeds.

Scoring: a scoring rule with weights, using the signal (title match, unit count, posting freshness, repost status) and any enrichment output.

Failure handling: what happens on a blank enrichment, a duplicate, a staffing-agency row, a rate limit, and a wrong company match.

Output: what a sales rep sees for a top-scored row.

Test: show how the design handles rows P-1021, P-1022, and P-1023 and at least two other rows of your choice.

Stay inside 45 minutes.

Deliverables

A step-order diagram or list with the reason each filter sits where it does
A column-by-column table: name, source, type, example value, decision it feeds
A scoring rule with weights and a worked score for P-1021, P-1022, P-1023, and two other rows
A failure-handling table covering blank, duplicate, staffing-agency, rate-limit, and wrong-match cases
A mock of the final rep-facing output for one top-scored row
A short note on what you would monitor in week two

Rubric

Dimension	What good looks like	Evidence to look for	Levels, weak to excellent
Workflow design	The flow is repeatable and restartable over the whole weekly batch, with a stable key and cheap steps before expensive ones.	Ordered steps with a stated reason; posting_id or domain used as a key; the run can resume after a failure.	Weak Steps listed in arbitrary order; no stable key; a failure means rerunning everything. Adequate Ordered steps with no reason; key is the row number or missing; cannot resume after a failure. Good Sensible order and a key such as posting_id; restart behavior mentioned but not specified. Very good Ordered steps with reasons, a stable key, and a defined resume point after a partial failure. Excellent Idempotent weekly run with keys, checkpoints, and a cost estimate per run; shows how week two differs from week one.
Data hygiene	Dedupes reposts and normalizes company names before enriching; excludes staffing agencies; checks that matches are right.	Dedupe and name-normalization step placed before enrichment; a rule for agency rows; a spot-check of matched domains.	Weak Enriches every row as given; agencies and reposts pass through. Adequate Removes obvious duplicates only; name variants and agencies still reach enrichment. Good Dedupes and filters agencies before enrichment; name variants only partly normalized. Very good Normalizes names, dedupes reposts, excludes agencies, and spot-checks matched domains. Excellent All of the above, plus a rule for wrong company matches and a review sample that would catch drift.
Tool fluency	Uses the tool that fits and gets a real result out of it, or sketches it cleanly when no tool is available.	A working table or flow, or a precise sketch; no needless extra tools.	Weak No tool or sketch; hand-waves how steps run. Adequate Opens a tool but gets little working; or a vague sketch. Good A partial working flow or a precise sketch covering most steps. Very good A working flow or a precise sketch that another builder could follow. Excellent Uses the tool fluently, tests on real rows, and picks simpler tools where they fit.
Signal design	Treats the hiring posting as a signal with freshness, strength, and noise, and weights it accordingly.	Freshness decay or window; separate treatment of a stale or reposted posting; an argued weight on unit count.	Weak Scores on title alone or does not treat the posting as a signal. Adequate Counts postings but ignores freshness, reposts, and strength differences. Good Weights title, units, and freshness; stale postings and reposts treated the same. Very good Freshness window, repost handling, and an argued weight on unit count. Excellent Signal tiers with decay, a repost rule, and a stated way to learn from outcomes after launch.
Systems thinking	Spends credits only where a field changes a decision, and puts a human check at the points where the data is least reliable.	Enrichment gated on an earlier filter; a named review point; a stated budget per run.	Weak Enrichment spent on every row; no human checks. Adequate Some filtering before enrichment but no budget or review point. Good Enrichment gated on an earlier filter; one review point named. Very good Credits spent only where a field changes a decision; a stated budget per run and a review point. Excellent Budget tied to expected value per lead, fallbacks when credits run out, and cost per qualified lead monitored.
Clear write-up	The column table and failure rules are specific enough that another builder could implement them.	Precise column names and examples; failure cases with explicit outcomes.	Weak Column table missing or vague; a builder could not implement it. Adequate Columns named with no sources, types, or examples. Good Columns with source and example; failure cases listed but outcomes vague. Very good Precise columns with examples and explicit outcomes for each failure case. Excellent Table and failure rules exact enough to implement without a question; includes a rep-facing example.

Why unaided AI alone does not pass this

The 30 rows are in the data, so the work is decisions about those rows: which are agencies, which are reposts, which name variants are one company, and where a blank unit count changes the score. A generic "enrich then score" outline does not make those calls. The Harbor Point pair, the Lakeshore and Maple and Vine spelling variants, and the staffing agencies each need an explicit decision. A reviewer can verify the design by checking those rows against the planted problems and by asking the candidate to walk through two of their test rows and why the workflow treats them that way.

Notes for the reviewer

Planted problems: four staffing agencies (Talentbridge Staffing, Apex Search Partners, Corvus Staffing Group, Kestrel Talent Solutions), repeat postings or spelling variants of one company (Harbor Point, Lakeshore, Maple and Vine, Cedar Row, Brightwater), one row with no unit count (P-1041), and several small single-property postings. Candidates without Clay access are not disadvantaged: score design quality, not the tool. Strong submissions place dedupe and agency filtering before paid enrichment. To verify, ask the candidate to walk through two of their test rows. Not observed is informative, not punitive.

Adapt this in ChatGPT