Research and ICP Skills for AI Agents

Purpose

Produce a focused one-page brief that gets a rep ready for a discovery call in five minutes.

Inputs

• Company name and domain
• The contact name and title
• Your product and the problem it solves

Steps

1. Run Serper queries for recent company news, funding, leadership changes, and product launches from the last twelve months.
2. Pull the company homepage and about page to capture positioning and target market in their own words.
3. Search the contact by name for recent posts, interviews, or role changes that show current priorities.
4. Use an LLM to summarize the account in five bullets covering what they do, who they sell to, and recent moves.
5. Have the LLM map two likely pains to your product and draft three discovery questions tied to them.
6. List one credible reason this account fits your ICP and one risk that could disqualify them.

Output

A one-page brief with company summary, contact context, two mapped pains, three discovery questions, and a fit-and-risk note.

Purpose

Build a clear negative profile so you can suppress accounts that look reachable but never convert.

Inputs

• Closed lost deals with loss reasons from the last 12 to 18 months
• Won deals for contrast
• Firmographic fields (size, industry, region, tech stack)

Steps

1. Pull closed lost and won records into one table with matching fields.
2. Group losses by reason and tag which ones are structural (bad fit) versus timing or execution.
3. Isolate the structural losses and find shared attributes across them.
4. Compare those attributes against the won set to confirm they rarely appear in wins.
5. Draft each disqualifier as a testable filter rule with a plain reason line.
6. Rank rules by how much wasted rep time each one removes.
7. Note any rule that risks cutting real deals and flag it for review.

Output

A ranked anti-ICP list where each row has a filter rule, the evidence behind it, the estimated accounts removed, and a confidence note for the RevOps team to apply as suppression.

Purpose

Map every person who influences a purchase at a target account so outreach covers the whole committee, not one contact.

Inputs

• Target account name and domain
• Your product and the department it serves
• Deal size band to gauge committee complexity

Steps

1. In LinkedIn Sales Navigator, list people at the account in the buying department plus adjacent functions like finance and IT.
2. Export them into Clay with title, seniority, and tenure.
3. Use an LLM to assign each person a committee role of economic buyer, champion, user, or blocker based on title and department.
4. Flag the most likely champion and the most likely blocker with a short reason each.
5. Enrich verified emails for the champion, economic buyer, and one user.
6. Draft a tailored one line angle for each role that speaks to their specific concern.

Output

A committee map for the account listing each person, their assigned role, a champion and blocker flag, verified emails, and a per-role outreach angle.

Purpose

Turn a messy list of churned and lost accounts into clean loss reason segments so you can exclude bad-fit patterns from your ICP.

Inputs

• CSV of churned or closed-lost accounts with firmographics
• Any notes, call summaries, or reason fields you have
• Your current ICP definition

Steps

1. Load the export and drop rows with no usable reason text or notes.
2. Have the LLM read each account and assign one primary loss reason from a fixed list: price, missing feature, wrong team size, low usage, champion left, competitor win.
3. Add a secondary tag for any firmographic pattern, such as industry or employee band.
4. Group accounts by reason and count each segment.
5. Flag segments where more than 30 percent share a firmographic trait as candidate exclusion filters.
6. Draft one plain sentence per exclusion filter explaining why to skip those accounts.

Output

A tagged table plus a short list of ICP exclusion rules ranked by how many churned accounts each would have blocked.

Purpose

Score current customer accounts by public expansion signals so account managers know where upsell effort pays off.

Inputs

• List of current customer accounts with domains
• Product usage or seat count per account if available
• Definition of what a good expansion looks like (more seats, new team, new use case)

Steps

1. For each account, search recent news, job posts, and press for growth markers like new hires, new office, funding, or new product lines.
2. Pull headcount trend by comparing current LinkedIn or public counts against any stored baseline.
3. Tag each signal with a category: team growth, budget event, new use case, or leadership change.
4. Weight signals against the expansion definition and combine into a 0 to 100 score.
5. Note the single strongest signal per account as a talking point.
6. Sort accounts high to low and flag any with stale or missing data for manual review.

Output

A ranked table of accounts with scores, signal categories, one talking point each, and a data quality flag.

Purpose

Build a ranked account list from recent funding rounds and estimate when new budget is likely to unlock for your offer.

Inputs

• Your ICP definition (industry, size range, region)
• Target round types (seed, Series A, Series B)
• Offer category the funding would fund
• Lookback window in weeks

Steps

1. Query Serper for funding announcements matching your ICP and round types inside the lookback window.
2. Parse each result for company name, round size, round stage, investors, and announcement date.
3. Filter out companies outside your size range or region and drop duplicates by domain.
4. For each remaining account, infer likely budget-release timing based on round stage and typical post-raise hiring or tooling patterns.
5. Score each account on fit and timing, then assign an outreach week that lands inside the estimated budget window.
6. Draft a one-line trigger note referencing the specific round and investor for personalization.

Output

A ranked table of funded accounts with domain, round details, fit score, suggested outreach week, and a ready-to-use trigger line.

Purpose

Derive a data-backed ICP by finding the shared traits of your best closed-won accounts.

Inputs

• A list of closed-won accounts, ideally with deal size and retention
• A list of closed-lost or churned accounts for contrast

Steps

1. Load both lists into Clay and enrich every account with industry, employee count, revenue band, region, and tech stack.
2. Tag each row as won or lost so you can compare the two groups.
3. Compute the distribution of each trait across the won group and note the tightest clusters.
4. Compare those clusters against the lost group to find traits that separate winners from losers.
5. Use an LLM to write the resulting ICP as firmographic ranges plus two or three qualifying signals.
6. Draft three negative filters that describe accounts to exclude.

Output

A written ICP with concrete firmographic ranges, positive signals, and exclusion rules you can paste directly into list-building filters.

Purpose

Estimate the likely contract renewal window for each target account so outreach lands 60 to 90 days before they re-evaluate their current vendor.

Inputs

• Account list with company name and domain
• Name of the incumbent tool or category you displace
• Known signals per account (adoption date, case study date, press release)

Steps

1. For each account, search public sources for the earliest mention of the incumbent tool (case study, review site, job post, press release).
2. Extract the adoption date or first-mention date and note the source.
3. Infer the standard contract term for that category (12 or 24 or 36 months) and add it to the adoption date.
4. Project forward in whole terms to find the next renewal date that falls after today.
5. Flag accounts whose estimated renewal is inside the next 120 days as priority.
6. Assign a confidence level based on how direct the source date was.
7. Sort output by soonest estimated renewal.

Output

A table with account, estimated renewal date, confidence, source link, and a suggested outreach start date.

Purpose

Build a list of new accounts that resemble your best customers by matching on observable, non-obvious shared traits.

Inputs

• List of 15 to 30 top customers by revenue or retention
• Firmographic and tech data source (Clay enrichment)
• Any LLM for pattern extraction

Steps

1. Enrich each top customer with industry, headcount, tech stack, funding stage, and department size.
2. Feed the enriched set to the LLM and ask it to name 5 to 8 traits these accounts share more than the general market.
3. Rank those traits by how strongly they cluster, keeping only the top 4 signals.
4. Convert each kept trait into a concrete filter or search query for Clay.
5. Run the filters to pull a raw list of matching accounts not already customers.
6. Score each new account by how many of the 4 signals it hits, from 0 to 4.
7. Sort descending and flag any account scoring 3 or higher for outreach.

Output

A scored account list with the matched signals per row and a short note explaining the shared pattern.

Purpose

Find hidden account clusters where you win at high rates but rarely target, so you can redirect outbound effort.

Inputs

• CRM export of closed-won and closed-lost deals
• Firmographic fields per account (industry, size, region, tech, department)
• Deal value and sales cycle length columns

Steps

1. Load all closed deals and confirm each row has firmographic fields plus outcome and value.
2. Group accounts by combinations of two or three attributes, for example industry plus headcount band.
3. For each group calculate win rate, average deal value, and average cycle length.
4. Flag groups with above median win rate but low total volume of attempts.
5. Cross check flagged groups against your current active target list to confirm they are under pursued.
6. Rank the silent segments by win rate times potential volume.
7. Draft one sentence of reasoning for why each segment likely converts well.

Output

A ranked table of undervalued segments with win rate, deal value, current coverage, and a short rationale for each.

Purpose

Estimate your total addressable market and break it into segments ranked by how easy each is to win.

Inputs

• Your ICP firmographic ranges
• Target regions
• Any known win rate or notes by segment

Steps

1. Translate the ICP into structured filters and run a Clay company search to count matching accounts per region.
2. Break the total count into segments by industry and size band and record the count for each.
3. Enrich a sample of each segment to confirm the filters return real matches and adjust if noise is high.
4. Use an LLM to score each segment on reachability, pain intensity, and expected deal size from one to five.
5. Rank segments by combined score and note the top three to attack first.
6. Write a one line outbound angle for each priority segment.

Output

A TAM figure, a table of segments with account counts and scores, and a ranked shortlist with an angle for each priority segment.