Strategy AI agent · Startup scouting
Startup Scouting AI Agent
Winnow the inbound lists, the conference floors and the banker introductions down to one deduplicated shortlist, log each founder claim beside its source and date, and leave acceptance to a named scout.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Named scout
- Claims dated
- After launch
- Agent Care
What this agent does
Sources the company, never the acceptance
In
- 01
A company is sourced, and the thesis it was sourced against goes on the record beside it.
- 02
A claim is made in a pitch, and it is logged as a claim with its speaker and its day.
Reason
- 03
A company arrives twice, once off an accelerator list and once from a banker, and the two merge.
- 04
A legal name differs from the trading name, and both spellings hang off a single company record.
- 05
A search returns nothing, and the absence is written down as an absence rather than as weakness.
Decide
- 06
A company overlaps a live conversation elsewhere in the firm, and the clash goes to a human.
- 07
A shortlist ages, and each entry carries the day it was checked and what has gone unchecked since.
Out
- 08
A company raises or is acquired, and its entry is marked stale until a scout has looked again.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
Sourcing, merging and dating belong to the agent. It holds no view on valuation and recommends no investment; a named scout, never the agent, accepts a company onto the shortlist.
Example workflow
One company, channels to shortlist
- 1Company surfacesAccelerator cohorts, conference lists, banker introductions or an unsolicited founder email
- 2Sourcing context assembledThe company, the legal and trading names it answers to, the thesis it was sourced against, and the day it first appeared
- 3Claim evidence draftedThe company, the claims made about it, the material under them and completeness
- 4Controls appliedDuplicate checks, name-resolution checks, freshness checks and completeness confidence
No human action required
Stages 1 to 4 run unaided, and nothing is accepted at any of them — the agent is qualifying, and a scout comes in at the completeness gate.
Travels to the named scout, who accepts.
Adds a ventures analyst read first.
Ventures review
The record is held with its claims, its gaps and the channel each one arrived through.
Accept · Append evidence · Send to ventures review
Accepted — by the named scout
- 6Company and pipeline records updatedOnly where write access and records policy allow it
- 7Outcome evaluatedClaim provenance, duplicate resolution, analyst corrections and what review found
Corrections
Each analyst correction is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Accepting a company onto the shortlist.
- Deciding what a company is worth.
- Calling an unverified claim confirmed.
- Ranking one thesis above another.
Automation boundary
Agent acts unaided
- ✓Merge duplicate arrivals into one company record under both names.
- ✓Log every founder claim with the material behind it.
- ✓Note the thesis each sourced company answers, in the record itself.
- ✓Raise an overlap with a live conversation to its owner.
No company joins a shortlist except by a named scout, inside the agreed boundaries.
- Judging whether a company is investable.
- Telling a founder the process has ended.
- Setting the thesis the team scouts to.
- Changes to the shortlist or the thesis register.
Example output
One founder claim, annotated
Our partnership development agent takes over the moment a commercial relationship comes into view, with ownership, screening and commission rationale; this page stops at sourcing and qualification.
Sourcing record · single company
Illustrative example- Claim
- Traction stated in the pitch deck
- Recorded as
- Said by the founder, unverified and unmeasured
- Company
- Growth claim
- Evidence of record
- Pitch deck, 3 August 2026
- Confidence
- Held unaccepted
- Held for
- The ventures scout, by name
As received
Lifted from the deck and the founder call, and it carries only as much weight as they do.
What the record holds
- Pitch deck
- Founder call note
- Accelerator listing
Why no claim is confirmed here
Whether this claim holds up is a matter for measurement, not for a model.
What the score decides
Below the configured threshold a record picks up a ventures read before the scout sees it.
Action
AcceptAppend evidenceSend to ventures reviewValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every company
From the channel that carried it
Where the evidence is used
This is a corporate ventures and strategy tool, not an investment process: the agent holds no view on valuation, makes no investment recommendation and puts no company in rank order.
A pitch is not diligence
What a founder says arrives as a claim, and the agent keeps it as one — with whoever said it and the day it was said.
What was checked, and not found
Coverage thins sharply at the earliest stages. For a good many sourced companies no register entry, no funding note and no press mention could be retrieved, and each blank is written down as thin coverage rather than as a mark against the company.
Build an evidence trail
The company, the claim it made and the scout who logged it stay on the record.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Deal-flow channels
- Accelerators
- conferences
- inbound
- Cohort lists and introductions
Company and register data
- Company registers
- trading names
- Legal names and incorporation entries
Thesis and pipeline
- Ventures CRM
- thesis register
- Sourcing rationale and stage
Agent
Startup scouting evidence
- Reads the channels
- Merges the record
- Holds for the scout
Founder and adviser material
- Decks
- call notes
- investor updates
- Founder claims and their dates
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six forges between the model and the scout
Six forges worked in turn, the last the hottest. Whatever holds its shape is named in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a record may stand
- L3
- Gives the acceptance to a scout by name
- L4 – L5
- Keep the company and the claim behind it
- L6
- Holds the shortlist unissued when signals degrade
How Nestack evaluates it
Evaluate the whole assembly — not only the shortlist record that comes out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the shortlist a scout reads
- E1Final-output evaluationDid the record log what the material actually says?
- E2Step-level evaluationDid the agent read the right company, the right channel and the live name set?
- E3Tool evaluationDid it read and write the correct company record and the correct claim?
- E4Confidence calibrationDo low-confidence records actually attract more analyst corrections?
- E5Slice evaluationHow does performance change across specific sourcing channels?
- E6Business outcomeHow many records needed a correction before the scout accepted?
Floor — the shortlist the ventures team answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, each lodged at the point in the run where it starts.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- MN-03
Stale company read
The company entry is not the one the market now knows.
Stage gathers
The companies, the channels, the claims and the dates
02 · Reasoning
2 modes
- MN-04
Claim asserted, not sourced
A claim is carried forward with nothing under it.
- MN-06
Blank read as bad
Thin coverage is turned into a low score.
Stage proposes
The claims, their sources and completeness
03 · Tool / write
2 modes
- MN-02
Thin record passed forward
A record moves on without the ventures read.
- MN-05
One company counted twice
Two channels leave two entries for one business.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- MN-01
Accepted, provenance unrecorded
The entry shows a shortlist place but not what it rests on.
Stage returns
The shortlist a scout accepts and a board reads
05 · Change / Version
1 mode
- MN-07
Silent freshness drift
A company moves on while its entry keeps the old note.
Stage tracks
Model, prompt, sourcing rules and record fields
- Sev-1 · a company accepted on no source
- Sev-2 · one company counted as two
- Sev-3 · channel degrades, record held back
Affected slices
Unsolicited email absorbs the corrections
A thesis-level verification figure can read clean while unsolicited founder email carries most of the rework. Nestack reports the correction rate by sourcing channel, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Unsolicited founder email | 6.5% | 3.6× | Review | |
| Conference and event lists | 4.6% | 2.6× | Review | |
| Accelerator cohort lists | 2.8% | 1.6× | Watch | |
| Banker and adviser introductions | 1.5% | 0.8× | Normal |
Bar: correction-rate lift vs. banker introduction baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
What an unmeasured claim costs
A cycle ends when the unverified founder claim is a standing case. That suite is what the next shortlist issued is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Correction rate rises on unsolicited founder email.
- 02
Diagnose
The traction number in the deck that the company had never actually measured is read back until one cause remains.
- 03
Improve
Changes leave numbered, and the companies that forced them travel attached.
- 04
Verify
One company case still failing is enough to stop the whole release.
- 05
Learn
The case is kept, and the scouting rules are rewritten alongside it.
Learn → Detect
The return edge. The next company is measured against a suite one case longer.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, shortlist assembly, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Thesis-capture and acceptance-boundary discovery.Week 1
- 02Channel, register and founder sources.Week 1, Week 2
- 03Claim-to-source and duplicate-resolution mapping.Week 1, Week 2
- 04Sourced-company ingestion.Week 2
- 05Company, claim and thesis binding.Week 2, Week 3
- 06Completeness scoring and review routing.Week 3
- 07Scout acceptance workflow.Week 3, Week 4
- 08CRM and channel integration.Week 5
- 09Verification and duplication cases.Week 4, Week 5
- 10Guardrails and shortlist controls.Week 4
- 11Company-trail instrumentation.Week 5, Week 6
- 12Deployment, documentation and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne thesis, one year | ProductionProduction scouting workflow | AdvancedMultiple theses / regions |
|---|---|---|---|
| Introduced at Pilot | |||
| Record assembly to your thesis | ✓ | ✓ | ✓ |
| Named scout release | ✓ | ✓ | ✓ |
| Sourced-company baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by thesis | — | ✓ | ✓ |
| Scout sign-off workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Deal-flow-source integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-region sourcing | — | — | ✓ |
| Cross-thesis evidence packs | — | — | ✓ |
| Large company estates | — | — | ✓ |
| Multi-channel deduplication controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, company volume, approval controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01Your live theses and the scouts who own them
→Thesis mapping and channel capture
Week 102Representative decks, cohort lists and call notes
→Claim binding, duplicate logic and the sourcing baseline
Week 203Your standing scouting procedure and the scouts it names
→Thesis mapping, claim binding and the automation boundary
Week 104Access to relevant APIs, feeds or exports
→Channel, register and founder source assessment, then integration setup
Week 205Shortlists you would not want revisited
→Verification cases and the evaluation round
Week 406What no scouting note may guarantee
→Completeness scoring, review routing, guardrails and release controls
Week 307A named scout who accepts onto the shortlist
→Handover to the named scout, then pilot and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
None of this is rounded for neatness; the fifth band shows two phases because that is how they run.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Scouting workflow discovery, thesis mapping and the automation boundary
W2Source integration and the sourced-company baseline
W3Record assembly, duplicate logic and release controls
W4Evaluation suite, verification cases and failure-mode testing
W5Channel integration, pilot shortlists and targeted corrections
W6One scouting year run under the ventures lead, then Agent Care handover
- Reading the band
- Every bar is the width its own work earns, and week five doubles up because two phases sit inside it.
- At the end of W6
- When the company record validates, Agent Care picks the agent up.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Strategy AI agent
Build a startup scouting agent around the traction number nobody ever measured.
Show us one company and the claim you took it forward on. Not a valuation, not a recommendation — a dated note of where the company came from, which thesis it answers, what the founder said, and what could not be found at all.
- Typical delivery
- 4–6 weeks
- Deployment
- Your live theses
- Named scout
- Claims dated
- After launch
- Agent Care available
Nestack Agents · Startup scouting
AGT-CS-12 · Agent Care available after launch