Adapt this in ChatGPT

Build an outbound motion from a raw account list — GTM Engineer take-home

GTM Engineer45 minutes

Context

Harrowgate Pay is an invented payroll and tip-compliance product for multi-location restaurant groups in the US. It automates tip pooling, overtime rules by state, and payroll exports to the major accounting tools. The company has two AEs and two SDRs, no outbound motion yet, and 14 paying customers, all of them restaurant groups with roughly 5 to 40 locations. A marketing intern exported 50 accounts from a webinar, a trade show, a purchased list, and the web form into one sheet. Your job is to turn it into an outbound motion the team can run on Monday. Nothing is sent to anyone.

Materials

The attached table (accounts.csv) has 50 rows and these columns: company, domain, industry, headcount, country, last_funding, source, notes. It was exported straight from four sources and has not been cleaned. Treat every value as the only information you have about the account; do not look anything up. The only company facts you can rely on beyond the table are in the context above.

Data

Download materials.csv (50 rows, 8 columns)

Columns: company, domain, industry, headcount, country, last_funding, source, notes.

The table scrolls sideways, so every column is there.

Showing the first 25 of 50 rows. Download the CSV for all rows.

Task

Work in the order that makes sense to you and explain your decisions as you go.

1. Define the ICP from the context in a few sentences: who is in, who is out, and the one or two attributes that matter most.

2. Clean and score the 50 rows against your ICP. Show the rule and weights, flag the rows you removed or merged and why, and give the ranked list with a score for every kept account.

3. Propose routing for three outcomes: sales-ready, nurture, and disqualify. For each, name the owner (AE, SDR, marketing, nobody), the response SLA, and the reason. Say what you do with accounts whose data is too thin to decide.

4. Draft three first-touch messages for your top accounts, each in the channel you recommend and under 90 words, using only facts in the table or the context.

5. Write a short explanation of your approach: the calls you made, the ones you are unsure about, and what you would do with more time. Include one piece someone could open and inspect, such as a spreadsheet with your formula, a script, or a workflow.

Stay inside 45 minutes.

Deliverables

• A short ICP definition: who is in, who is out, and the attributes that matter most
• The cleaned list: rows removed or merged, with the reason for each
• Scoring rule with weights, plus the ranked list with a score on every kept account
• Routing for sales-ready, nurture, and disqualify, each with an owner, an SLA, and a reason
• Three first-touch drafts for top accounts, each under 90 words, with the channel chosen
• An inspectable artifact (spreadsheet, script, or workflow) that produced the ranking
• A one-page explanation of approach, open doubts, and next steps

Rubric

Dimension	What good looks like	Evidence to look for	Levels, weak to excellent
Data hygiene	Cleans before scoring: finds duplicates and junk rows, decides how to treat blanks, and records what was changed.	Merged or removed rows listed with reasons; a stated rule for blank headcount or funding; the ranking built on the cleaned set.	Weak Scores the raw list as given; duplicates and junk rows stay in the ranking. Adequate Spots some obvious junk but misses near-duplicates; blanks silently treated as zero. Good Removes junk, merges most duplicates, and states what was changed; blanks handled inconsistently. Very good Finds all duplicates and junk, handles blanks with a stated rule, and keeps an audit list of changes. Excellent Finds everything, treats each blank by its consequence (score, flag, or hold), and makes the cleaning repeatable.
ICP judgment	Derives the ICP from the context, picks the attributes that predict fit, and scores accordingly with argued weights.	ICP tied to the customer profile in the context; headcount and industry used with reasons; near-fit accounts handled deliberately.	Weak No ICP stated; score is arbitrary or based on headcount alone. Adequate Generic ICP (restaurants, mid-size) that ignores the customer profile; weights not explained. Good ICP matches the context; weights stated; near-fit accounts such as hotels and caterers lumped in or out without argument. Very good ICP specific and defended; near-fit accounts placed deliberately; non-US and oversize accounts excluded for a reason. Excellent ICP with tiers, an argued call on each ambiguous group, and a note on what data would change the ranking.
Routing decisions	Three outcomes with owners and SLAs that fit a four-person team, plus a path for thin-data accounts.	Distinct owner and SLA per outcome with reasons; a hold or enrich path for blanks; capacity of two SDRs and two AEs respected.	Weak Everything goes to one queue; no owners or SLAs. Adequate Outcomes named but owners or SLAs missing, or identical for all three. Good Owner and SLA per outcome; reasons thin; no plan for accounts with missing data. Very good Distinct owners and SLAs with reasons; a path for thin-data accounts; sized to a two-AE, two-SDR team. Excellent Adds absence and overload handling, a recycle rule out of nurture, and a way to check the routing after a week.
First-touch writing	Each message opens on one true detail from the table or context, makes one ask, stays under 90 words, and invents nothing.	Message cites a real field (locations, new CFO, tipped staff); one clear ask; no invented claims about the prospect.	Weak Generic template with a name swapped in; invented claims about the prospect. Adequate Personalized with a surface detail; feature list; more than one ask or over the word limit. Good Built on one real fact per message and under the limit; clear ask; the three messages read alike. Very good Each message uses a different relevant fact and role; honest about what is known; channel choice justified. Excellent Messages differ by account situation, claim only what the table proves, and say why this account and not another.
Clear write-up	A teammate can pick up the ranking, routing, and messages and act without asking questions.	Summary first; consistent terms; open doubts named; the artifact is labeled and opens.	Weak Disorganized; cannot tell what was decided. Adequate Decisions present but scattered; terms change between sections. Good Readable explanation with decisions and reasons; doubts mostly implicit. Very good Clear structure, named doubts, and a teammate could run it Monday. Excellent Readable in two minutes, with trade-offs and next steps explicit and an artifact that explains itself.
Working with AI	Uses an assistant to speed up cleaning, scoring, and drafting, then checks its output against the rows.	Spot-checks shown in the work; corrections to an assistant mistake; prompts that include the actual rows.	Weak Pastes the assistant output unedited; wrong counts or invented rows go unnoticed. Adequate Uses an assistant but accepts results without checking against the data. Good Uses an assistant for drafts and scoring and spot-checks a few rows. Very good Directs the assistant with the rows and rules, verifies counts, and fixes its mistakes. Excellent Uses the assistant inside a repeatable process, with checks that would catch drift on the next list.

Why unaided AI alone does not pass this

Reading the file is easy with any tool, so the test is the judgment calls on this data. Three accounts appear twice under different spellings and domains, two rows are test entries, and five rows have blank headcount or funding, and each of those needs a stated decision. Scoring also requires choices about this ICP (hotels, a caterer, a staffing agency, a Canadian and a UK group, and oversize chains all sit near the edge). A reviewer can verify the work by checking the rows against the planted problems and by asking the candidate to walk through two of their decisions and the reasons behind them.

Notes for the reviewer

Planted problems: three duplicate pairs (Copperleaf and Copper Leaf Hospitality, Marlow & Finch Bistros and Marlow+Finch Bistros Inc, Northgate Taproom Group and NorthGate Taproom Group LLC), two test rows (Test Co, asdf), and five rows with blank headcount or funding. Fit questions (hotels, caterer, food distributor, non-US) have no single right answer; score the argument. To verify, ask the candidate to walk through two decisions, for example one merged pair and one edge-of-ICP account. Not observed is informative, not punitive.

Adapt this in ChatGPT