AI TOOLKIT — PRIMARY ASSET
9 prompts that do the strategic hiring work before you ever brief an agency.
Most seed founders do one of two things when a role opens. They hire fast off a gut feel and pay for it in month four when the wrong person is still ramping. Or they freeze, because there’s no planning layer underneath the decision, and the next call they make is to a recruitment agency at 20%.
I ran the Frontiers EMEA commercial build this way. £1M in agency spend eliminated over 20 months. We never posted a single job board ad. The difference wasn’t sourcing harder. It was having a planning layer before sourcing ever started. These 9 tools are that planning layer, run as prompts instead of a £180-300K/year hire.
Each tool below produces a scored table or a clear verdict, not vague advice. Run them in order for a full hiring plan review, or run any single one standalone when you only need one decision.
What it does: Scores every role on your hiring plan and tells you which to fund, defer, or cut.
Why it works: At Frontiers EMEA, the commercial hiring plan had more roles on it than the runway justified. Scoring every role against the same four criteria, before sourcing started on any of them, is what let us commit fully to the ones that mattered instead of spreading effort thin across all of them.
Paste this prompt into Claude:
You are an AI Head of Talent triaging a hiring plan. My context: Company stage: [fill in] Runway: [fill in] Current hiring plan: [list every role, one per line] Comp bands: [fill in, or "not set"] Dealbreakers: [fill in] For each role on the plan, score 1-10 across these four criteria: 1. Mission criticality: does the business goal die without this hire? 2. Replaceability: 1 means must-hire, 10 means easy to skip or delay 3. Time-to-impact: how fast does this role start shipping value once hired? 4. Cost-fit: does the likely salary fit the runway I gave you? Output a table with columns: Role | Score (each of the 4) | Total /40 | Recommendation | Reasoning. Recommendation bands: - Under 24/40: KILL. Cut it from the plan entirely. - 24-32/40: DEFER 6 months. Revisit once runway or priorities shift. - 32+/40: FUND NOW. This is a real, funded role. Do not hedge. Give a clear recommendation for every role, even if the case is close.
How to read the output: Look at the total column first, not the individual scores. If two roles are close (within 2-3 points of each other), read the reasoning column before deciding, since the model will often flag which one is closer to the line. If everything comes back FUND NOW, your plan probably has too few roles listed, not too many good ones. Re-run with your full wish list, not just the roles you’ve already mentally committed to.
What it does: Checks what percentage of a role’s actual work Claude or automation can already do, before you commit to a full-time hire.
Why it works: At the Attorney General’s Office, we rebuilt the recruitment function from zero and cut time-to-hire by 57% and agency spend by 39%. Part of that came from not hiring for work that didn’t need a dedicated human yet. Capability moves fast. A role scoped a year ago as fully human might be 70% automatable today, and most hiring plans never get re-checked against that.
Paste this prompt into Claude:
You are auditing a proposed role to check how much of the actual work can be automated before I commit to hiring someone full-time. The role: [job title and a plain-English description of what this person would actually do day to day, not the job ad version] Break the role into its component tasks and estimate what percentage of the total workload each task represents. For each task, state whether Claude or a similar AI tool can already do it, partially do it with human review, or cannot do it at all. Then give me a single overall automation percentage for the role, and a verdict: - Above 70% automatable: skip the full-time hire. Deploy an AI workflow plus one human reviewer, possibly on a contract basis. - 30-70% automatable: this is fractional or part-time work, not a 40-hour role. - Under 30% automatable: this is genuinely human-judgment work. Hire full-time. Be specific about which tasks fall into which bucket. Don't just give me a number, show your working.
How to read the output: Pay attention to which specific tasks the model flags as automatable versus not. If the "cannot automate" tasks are the ones that actually matter for the role’s success, a low automation percentage is the right call even if the total task count looks automatable on paper. Re-run this every 6-12 months on roles you’ve already filled. Capability shifts, and a role you hired for at 20% automatable might be 60% automatable a year later.
What it does: Tests every "I need to hire someone" instinct against three cheaper alternatives before you commit to a full-time salary.
Why it works: Most "I need a hire" requests are actually project-shaped or tool-shaped work wearing a job description. Defaulting straight to full-time costs roughly three times what the work actually needs, and nobody checks the other three paths unless something forces the comparison.
Paste this prompt into Claude:
You are auditing a role on my hiring plan to check whether it should be a full-time hire, or something cheaper. The role: [job title and description of the work] My context: [company stage, runway, comp bands from earlier] Score this role against four paths: 1. HIRE FULL-TIME: the work is core, ongoing, and needs to be embedded in the team's identity long-term 2. CONTRACT (3-6 months): the work is project-shaped with a clear end date 3. FRACTIONAL: the work is strategic but doesn't need 40 hours a week 4. BUY A TOOL OR SAAS: an existing vendor product already solves this Score each path out of 10 for fit. If HIRE FULL-TIME doesn't win by at least 2 clear points over the next-best option, flag the role for re-evaluation and explain which of the other 3 paths is the stronger fit and why. Give me a final recommendation, not just the scores.
How to read the output: If full-time wins by a clear margin, that’s your confirmation to proceed as planned. If it doesn’t, don’t override the model’s recommendation just because "hiring someone" feels like the default. Run Tool 4 (Cost-Per-Hire Estimator) next on whichever path wins, so you know the real number before you commit either way.
What it does: Calculates the real, fully-loaded cost of a hire, not just the base salary line you’ve been budgeting against.
Why it works: At Nationwide Building Society, we made 52 specialist hires in 9 months across a £1B transformation programme, every one sourced directly. Running the true cost of each hire before committing budget is what kept that programme inside its numbers. Base salary is roughly half the real cost. Founders make headcount decisions on the wrong number constantly, because nobody adds up the rest of it until the invoices land.
Paste this prompt into Claude:
Calculate the true, fully-loaded first-year cost of this hire. Role and proposed base salary: [fill in] Country/region: [UK, EU, etc, for tax and benefits assumptions] Break the cost down across these lines: 1. Base salary 2. Benefits and payroll tax (use a reasonable UK/EU estimate if I haven't told you the exact figure, and state your assumption) 3. Equity or bonus, if relevant (ask me if you don't have this, otherwise assume none) 4. Sourcing cost: if using an agency, assume 20% of base. If sourcing in-house, estimate hours at a loaded hourly rate and ask me how many hours I expect to spend. 5. Tool stack (laptop, software seats, standard estimate is fine) 6. Ramp opportunity cost: assume 3 months of partial productivity unless I tell you otherwise 7. Replacement risk: ask me for an expected retention rate, or assume 75% if I don't have one, and calculate the risk-weighted cost of a potential re-hire Output a table with each cost line and a final TOTAL YEAR 1 figure, plus a TOTAL 2-YEAR figure. State every assumption you made clearly so I can correct it and re-run.
How to read the output: The number that matters is the gap between what you’d budgeted (usually just base salary) and the TOTAL YEAR 1 figure. If that gap is large, that’s the number to bring back into Tool 3’s build-vs-buy comparison, since a fractional or contract option often looks a lot more competitive once the full cost of full-time is visible.
What it does: Picks the actual sourcing channel mix for a specific role, instead of defaulting to "post it on LinkedIn and hope."
Why it works: Wide-net sourcing wastes budget on channels where your ideal candidate doesn’t live. Channel-led sourcing concentrates effort where they actually are, and that’s the difference between a role that fills in 30 days and one that’s still open at 90.
Paste this prompt into Claude:
Recommend a sourcing channel mix for this specific role. Role: [title, seniority level, function] What's worked before, if anything: [fill in, or say "nothing tried yet"] Budget available for sourcing: [fill in] Score the following channels 1-10 for this specific role across: talent density (does my ideal candidate actually live here), time-to-first-screen, cost per qualified candidate, likely conversion to hire, and effort cost in team hours: - LinkedIn outbound (direct sourcing) - LinkedIn job board post - Referral programme - Niche community or forum specific to this role's discipline (name the most relevant one for this role) - Direct outreach via email Recommend a mix, not a single channel: something like Primary 60% / Backup 30% / Test 10%. Name which channel fills each slot and why, and explicitly flag which channel to avoid for this specific role and why.
How to read the output: If the model recommends the job board as your primary channel for a senior or specialist role, push back and ask it to reconsider. Senior individual contributors rarely apply to job boards; they need to be found. Use this output to decide where you spend the first week of sourcing effort, not to build a channel strategy you never revisit.
What it does: Turns an approved role into a week-by-week sprint with sourcing days, interview windows, decision gates, and Friday checkpoints.
Why it works: "We need to hire ASAP" has no checkpoints and no decision gates built into it. At the Attorney General’s Office, replacing that vague urgency with a Friday-stamped calendar is what forced the accountability that got time-to-hire down 57%.
Paste this prompt into Claude:
Build a 30-day hiring sprint for this role, with a target close date and a specific Friday checkpoint each week. Role: [title] Target start date for sourcing: [fill in] Who's involved: [you, hiring manager, anyone else screening or interviewing] Structure the sprint as: Week 1: Setup. Lock the scorecard and comp band, activate sourcing across the channels I give you: [paste your channel mix from Tool 5, or say "not run yet"] Friday checkpoint: what number of candidates in pipeline counts as on-track by end of week 1? Week 2: Screens. Recruiter and hiring manager screens run. Friday checkpoint: how many confirmed finalists should be ready for panel by end of week 2? Week 3: Panels. Final round, debriefs, reference checks. Friday checkpoint: decision committed and offer drafted. Week 4: Close. Offer sent, negotiation, signature. Friday checkpoint: signed offer and onboarding kicked off. For each week, give me the specific number or milestone that tells me I'm on track versus falling behind, not just a vague description of the week's theme.
How to read the output: The Friday checkpoints are the entire point of this tool. If you hit week 2 and you’re below the checkpoint number, that’s your signal to rebalance sourcing immediately, not to wait and see. A sprint that slips in week 1 rarely recovers on its own by week 4.
What it does: Designs a role-specific interview panel, stages, and question set instead of generic behavioural questions that test nothing in particular.
Why it works: Generic behavioural questions test generic skills. A role-specific loop surfaces the signal that actually matters for the scorecard you’re hiring against, which is the difference between a panel that takes four hours and produces a confident decision, and one that takes four hours and produces a split vote.
Paste this prompt into Claude:
Design a full interview loop for this role. Role: [title] The 3-5 things this person absolutely must be able to do in the first 6 months: [fill in] Team size and who they'd report to: [fill in] Design a 4-stage loop: 1. Recruiter screen (15-20 min): basic fit and dealbreaker check 2. Hiring manager screen (30-45 min): tests the single most important scorecard category 3. Panel (3-4 interviews x 45 min): distributed coverage of every scorecard category with no overlap between panellists 4. Final/values round (30 min): sign-off and values alignment For each stage, tell me: who should run it, which specific scorecard categories it tests, 3-5 questions specific to this role (not generic behavioural questions), and what a strong versus weak answer looks like for each question.
How to read the output: Check that no two panel stages are testing the same scorecard category. If they are, ask the model to redistribute coverage. A panel where three interviewers all end up assessing "communication skills" and nobody tests the actual hard skill the role needs is the single most common failure mode this tool exists to catch.
What it does: Runs the finalist candidate through a structured risk check before the offer goes out, to catch the late-stage failures that turn into a counter-offer loss or a surprise quit at month four.
Why it works: The last-mile failures (counter-offer, late no-show, month-4 quit) are almost always preventable with five minutes of pre-offer signal reading. Most hiring managers skip this step entirely and send the offer cold.
Paste this prompt into Claude:
Run a pre-offer risk audit on this finalist before I send an offer. What I know about the candidate: [current company and tenure there, stated reason for leaving, what they said they want next, salary expectations, anything about their current process elsewhere that they mentioned, reference feedback if you have it] Check across five risk dimensions and flag anything concerning: 1. Resume integrity: does the timeline, seniority, and scope they described add up cleanly? 2. Tenure pattern: does their job history suggest they're near a natural "itch to move" point, or a cliff (like a vesting date) at their current company? 3. Motivation read: is their stated reason for leaving coherent with what they say they want next? 4. Reference signal: if you have reference notes, are there any soft warnings buried in otherwise positive feedback? 5. Offer-stage risk: how likely is a counter-offer from their current employer, and is comp aligned with what they said they need? Give me an overall risk score: GREEN, YELLOW, or RED. List the top 3 specific risks, and for each one, a specific mitigation action I can take before or during the offer call.
How to read the output: A YELLOW or RED score doesn’t mean don’t hire them, it means don’t send the offer blind. If counter-offer risk is flagged, that’s your cue to ask the direct question on the closing call rather than finding out after the offer’s already out. This tool is most valuable exactly when the panel is confident, since that’s when people skip due diligence.
What it does: Maps a new hire’s first month week by week, so they’re shipping something real before month two instead of still finding the right Slack channels.
Why it works: Generic onboarding is an HR checklist: equipment, accounts, e-learning modules. It doesn’t ramp anyone. A role-specific roadmap with weekly checkpoints is what turns a new hire into someone contributing on a visible timeline, not an open-ended one.
Paste this prompt into Claude:
Build a 30-day onboarding roadmap for this new hire. Role: [title] What they'll own in month 1 if things go well: [fill in] Any risk flags from the pre-offer audit worth building into week 1 (e.g. counter-offer risk, needs early wins): [paste from Tool 8, or say "none"] Structure as 4 weeks: Week 1: Learning and integration. What should they understand, who should they meet, what systems should be set up. Week 2: First wins. 2-3 small, visible things they should ship. Week 3: Owning something. Their first real project with clear success criteria. Week 4: Leading something. Running a meeting or owning a decision independently. For each week, give me: what should ship, and the specific question their manager should ask in the week's 1:1 (not a generic "how's it going"). Also give me an early-warning checklist: 4-5 specific signals that suggest this hire is at risk of not working out (e.g. missed 1:1s, no questions asked, avoiding a specific type of problem), so it can be caught in week 2 rather than month 4.
How to read the output: The early-warning checklist is the part most onboarding plans skip entirely. If any of those signals show up in week 1 or 2, that’s worth a direct conversation immediately, not a wait-and-see approach. Waiting until the 30-day mark to notice a problem means you’ve already lost most of the runway to fix it.
These 9 tools cover the strategic planning layer honestly and completely. What they don’t cover: the actual hours of sourcing conversations, the follow-up sequences that keep a pipeline warm, the judgment calls in a live negotiation when a counter-offer lands mid-call, and the pattern-matching that comes from having run this exact process across hundreds of hires.
If you run these tools and the plan is clear but the execution keeps slipping because you don’t have the hours in the week, that’s the actual signal to bring someone in, not a failure on your part. The tools tell you what to do. They don’t do the doing.
These are the frameworks I use inside every engagement. Run it yourself using the guide above, or I can run it for you.
Book 20 minutes →