News Skalata is now Tall Order
About Team Portfolio Founders Investors AI Ops Writing Contact Pitch us
AI Ops

How we use AI

We use AI to look at every deal in more depth, so our partners can spend more time with founders.

How a deal moves Last updated 9 October 2026

In short

Every deal that reaches us is analysed in full as soon as it arrives. The AI reads the deck and the website, looks into the founders and the market, scores the company against our written criteria, and writes the memo a partner reads.

A partner decides whether a deal proceeds to a discovery call, and the investment committee decides whether we invest. The AI can recommend a pass, but it can’t decline a founder on its own.

How a deal moves

Every deal takes the same path, however it came in. Hover over a step for a summary, or click it for the detail.

IntakeIntakeDeals come in through the pitch form on this site, a monitored deck inbox, a deck or link posted in our deal channel, the founder chatbot, and the Sourcing Signals agent, which finds founders before they apply. Every deal becomes one standard record, with missing fields marked as missing.
Deck readDoc IntelligencePulls the text off the deck and the website with Firecrawl and extracts the problem, product, traction, funding, sector, location, key metrics and any competitors the deck names. If neither has readable text, the run stops here.
GatePre-screening gateChecks the deal against our investment mandate: stage (pre-seed to Series A), round size (under $20M), an Australian connection, and whether it’s a technology business rather than direct-to-consumer or e-commerce. A recommended pass goes to a partner. A warning sign, like a solo founder, no revenue yet, a crowded market or a valuation ahead of the stage, becomes a question for the founder and lowers the score, but never ends a deal on its own.
FoundersFounder IntelResearches each founder’s career with Renidly and web search, and checks it against what the deck says.
MarketMarket TrendsSizes the market independently with Perplexity and Exa, and compares it with the market size the deck claims.
ContactsEnrichmentFinds the founders’ contact details through FullEnrich.
CompetitorsCompetitive IntelFinds competitors and their recent funding from its own searches, starting once the market research is done.
ComparablesComparablesEvery finished analysis is stored in Supabase as a list of numbers describing the company. Synthesis A pulls the five most similar past deals from it. Our accelerator cohorts and the deals we’d already written up went in before the first live run, and every new analysis adds to it.
Synthesis ASynthesis ADrafts the memo: an overview, key metrics, a rating on every criterion, grouped by category, the red flags and warning signs, a recommendation of proceed, pass or more info, and an interview agenda. Scored against the closest past deals, with citations.
Synthesis BSynthesis BRe-reads the draft against the original documents and every agent’s output, and flags any claim the sources don’t support before the memo is saved.
AI committeeAI committeeFour agents argue the deal from opposed positions. The sceptic hunts for the reason to pass, the optimist for the upside, the futurist asks where the world goes if it works, and the operator checks whether it can be built. They return where they agree, the dissent that didn’t resolve, and a next step.
MemoMeeting briefingWrites the memo into the company’s record in Attio, as a Company Profile and Briefing Notes with its score, and posts it to Slack. Its format depends on the stage: discovery call, interview, or the investment recommendation report for committee.
PartnerPartnerReads the memo and decides whether the deal proceeds to a discovery call or goes on watch.
ReviewPartner reviewA recommended pass lands in Slack. If the partner agrees, it’s declined. If they disagree, the run picks up where it stopped.
DeclinedDeclinedA partner writes the reason, and the decline is recorded in Attio.
Re-analysisRe-analysisWhen new material lands, like a call, an email thread or a partner’s own concerns, it works out which steps the new information affects and re-runs only those. Founder research is reused.
Due diligenceDue diligenceThe discovery call, the founder interview, reference calls, and technical and financial checks. What comes out of each goes back through re-analysis, and the final report goes to the investment committee.
WatchWatchNot yet. We note what would change our mind and come back to it.
Investment committeeInvestment committeeDecides whether we invest, working from the investment recommendation report.
InvestedInvestedTerms are agreed and the money is wired. The analysis stays attached to the company’s record.
CRMCRM recordEvery decision and the reasoning behind it are written back to Attio, whether we invested, watched or declined.

Agents

Each agent has one job, and they run in this order, set in code. On re-analysis, fixed rules pick which steps run again. Letting a model choose the order would cost more and make runs less consistent. Open an agent for what it does, the tools it uses and its edge cases.

Doc Intelligence

Reads the deck and the website: problem, product, traction, funding, sector and the competitors the deck names.

Firecrawl

Pre-screening gate

Checks the deal against our mandate, from stage and round size to location and business type, and recommends pass or proceed.

Founder Intel

Researches each founder’s career and checks it against the deck.

Renidly

Market Trends

Sizes the market independently, with growth and trends, and checks the deck’s number against it.

PerplexityExa

Competitive Intel

Finds competitors and their recent funding, after the market research is done.

PerplexityFirecrawl

Synthesis A

Drafts the memo, rates every criterion, pulls the closest past deals, and recommends proceed, pass or more info.

pgvectorSupabase

Synthesis B

Checks the draft against every source and flags any claim the sources don’t support.

AI committee

A sceptic, an optimist, a futurist and an operator argue the deal and return agreement, dissent and a next step.

Meeting briefing

Writes the memo into Attio and Slack, formatted for the stage the deal is at.

AttioSlack

New context

On re-analysis, pulls the facts out of new call transcripts, notes, emails and files.

AttioNotion

Enrichment

Finds the founders’ contact details.

FullEnrich

What we score

We score every company against the same criteria. Synthesis A rates every criterion strong, adequate or weak. Code then turns the ratings into a score from 1.0 to 3.0, takes a set amount off for each warning sign the evidence didn’t clear up, and maps the result to proceed, pass or more info.

How the score is worked out, and what it decides →

Domain depth

How they found the problem, whether they know how their buyers actually buy, and what they know about the industry that only comes from working in it.

Founder-idea advantage

What edge this founder has on this exact problem, and whether it’s hard to copy or something anyone could have.

Execution evidence

How fast they ship, from idea to product to first customer, and how closely customer feedback drives the product.

Technical capability

Whether the technical work is done in-house, what the team has built before, who they’ve convinced to join, and whether they run the company on AI too.

Perseverance and grit

Whether they’ve pushed through hard things before, including moments when quitting would have made sense.

Coachability

Whether they know what they don’t know, name their biggest risks unprompted, and can point to decisions they changed on advice.

Problem validation

Evidence the pain is real: customers who can describe it, trials that convert to paid, inbound demand and low churn.

Market size

A bottom-up count of real customers at realistic contract sizes. For AI companies, the labour spend they can replace, not only existing software spend.

Timing and adoption

Where buyers sit on the adoption curve, and whether sales conversations sound like “not ready” or “we’ve been looking for this”.

Market insight

Where the founder thinks the market is heading, why now, and whether that view comes from experience or from reports.

Category creation

The change it’s taking advantage of, like a new technology, regulation or behaviour, what it believes about the market that most people get wrong, and whether it changes how the category works or competes inside it.

Defensibility

What the product knows after 1,000 customers that a new entrant couldn’t copy, and how deeply it’s built into work the customer can’t afford to get wrong.

Product differentiation

Why customers say it’s better, and whether that’s more than price, and whether the innovation is in the data and workflow or only in the model call.

Distribution

Whether the company can reach customers in a way that doesn’t depend on its AI.

Market dynamics

Reliance on any one model provider, how hard it is for customers to leave, and how many tools could copy the core idea within a year.

Moats at scale

The barriers it can build as it grows: regulation, distribution lock-in, platform effects, trust, and unit costs that fall with volume.

Traction

Revenue, paying customers, gross profit, pipeline, customer concentration, and how much it cost to get there.

Product-market fit

Retention, usage, and whether customers sound enthusiastic or just polite.

Buyer value

The measurable results customers report, how fast they arrive, and whether the return on investment holds up when the customer’s finance team checks it.

Revenue expansion

Whether existing customers spend more over time, and whether pricing grows with their usage.

Go-to-market

A specific ideal customer, the first 10 to 20 named targets, acquisition cost against lifetime value, and which channels repeat.

Revenue model

How the company charges, what each new customer costs to serve (AI inference included), and how predictable the revenue is.

Data advantage

Proprietary data that gets better with use and can’t be replicated.

Model dependency

How much the company relies on one foundation model, and whether it survives a model change.

Position against foundation models

Whether it builds where foundation models won’t go, or competes with what they’re getting better at.

Workflow integration

How deeply the product sits in the customer’s work, and what removing it would take.

Platform ambition

Whether it stands alone or rides on another platform, and how a first product grows into a platform within 18 months.

Fund return potential

The three to five things that must be true for this company to return the fund, and whether the entry price makes sense for its stage and traction.

Ownership and dilution

What our stake looks like after the Series A, the Series B and at exit.

Exit paths

Comparable acquisitions and listings in the space and region, with best, base and downside cases.

Stack

n8n moves data in and out, LangGraph runs the analysis, LangSmith records every run, and OpenRouter connects to the models.

Running the analysis

What takes a deal in, runs the analysis and writes the results out.

n8n

Deals reach us through our pitch form, the founder chatbot, the deck inbox, our deal channel in Slack and the Sourcing Signals agent. Each one sends a webhook to n8n, which turns it into one standard record for the analysis. When the analysis finishes, n8n writes the results to Attio and Supabase and posts them to Slack. It holds the logins for those systems, and retries anything that fails. We use n8n because that plumbing, logging in to other systems, retrying and slowing down when a service asks, comes ready-made, and watching a run move step by step through its canvas taught us more about what we’d built than any log. None of the analysis happens there.

LangGraph

Runs the analysis itself, as code. It sets what runs when: Founder Intel and Market Trends run at the same time, and Competitive Intel waits for the market work. It saves progress as it goes, so a run can stop for a partner at the gate and carry on from the same point, and a re-analysis re-runs only the agents the new information affects. n8n can call a model several times in a row, but it can’t keep track of where an analysis is up to or take a different path depending on a result, which is why the analysis lives here.

OpenRouter

One connection to every model provider. Each agent’s model is a setting, so moving an agent to a different model, or a different provider, needs no code change, and every call’s cost is logged against the run. Today every agent runs on Anthropic’s models.

Research

The tools the agents use to look things up beyond the deck. Perplexity and Exa are in an A/B test for market research.

Firecrawl

Turns a company’s deck and website, and competitors’ websites, into plain text the agents can read.

Perplexity A/B test

Answers market questions from the live web, with a source for every figure.

Exa A/B test

Finds companies and articles from a plain description, like “export compliance software for seafood”.

Renidly

Pulls each founder’s roles, employers and dates, and their post history, from their public LinkedIn profile.

FullEnrich

Finds founders’ contact details. It checks a series of data providers in turn and stops at the first good answer.

Alexandria Testing

Firecrawl’s library of data sources for AI agents, including professional profiles, research papers and code. We’re testing it as a research source for the agents.

Tracing

Every run is traced, and every change is tested before it ships.

LangSmith

Traces every run as one tree under the deal it belongs to: each agent in order, the prompt it used, the tools it called, what came back, how long it took and what it cost. When an output is wrong, we can find the step and the prompt behind it in a couple of clicks. Every agent’s prompt lives here too, versioned, with a production and a testing version, and is loaded when a run starts. Before a change goes live, it runs across our test set of thousands of past deals and is scored side by side with the live version. It doesn’t ship if it does worse on any group of deals, even when the average improves.

Storage

Where the criteria, every analysis and every file are stored.

Supabase

Our database. It holds the scoring criteria and screening rules, and every analysis, file and source the AI produces.

pgvector

How we find similar past deals. Each finished analysis is stored as a list of numbers describing the company, and the closest matches become its comparables.

Where the team works

The tools the team works in every day.

Attio

Our CRM. Each company’s record, with the AI’s analysis, its score and the decision we made.

Notion

Where partners keep the criteria, screening rules, thesis and mandate. Each deal also gets its own workspace for our notes, assumptions, diligence questions and the investment memo, and those notes feed back into re-analysis.

Slack

Where each finished analysis is posted, and where a partner agrees or disagrees when the gate recommends a pass. Posting a deck in our deal channel starts a run.

Email

Resend

Sends our email: sign-in links for our tools, login codes and notifications to the team.

Models in use

New models come out every few weeks. We run each one against our test sets, past deals where we already know the right answer, and score it on the same checks: whether its claims are supported, how sound its reasoning is, what it pulls out of a deck, and how it applies our mandate. We switch when a model does measurably better.
ModelWhat it doesWhy this model
Claude Opus 5.5Synthesis A drafts the memo and Synthesis B checks it against every source.The hardest reasoning in a run: long, multi-source analysis, where it has stayed the most consistent on our deals.
Claude Sonnet 5.5Reads the deck, runs the gate and the research agents, writes the memo into Attio, and runs the founder chatbot.Mostly reading, searching and summarising at volume, where speed and cost matter more than deep reasoning.
Perplexity SonarAnswers the market research questions from the live web.Every answer comes with its sources, so each market figure can be checked.
OpenAI text-embedding-3-smallTurns each finished analysis into numbers, so similar past deals can be found.It runs on every deal, and a small embedding model is enough for matching.

Who does what

The AI’s work all happens before a person forms a view, and it takes about five minutes. The decisions all happen after.

The AI does this

  • Reads the deck, the website, the data room and anything that arrives later.
  • Screens the deal against our mandate, and can recommend a decline but can’t send one.
  • Researches the founders, the market and the competitors.
  • Scores the company against our criteria and drafts the memo.

A person does this

  • Whether to proceed. A partner reads the memo and decides whether the deal goes to a discovery call.
  • Every investment decision. The investment committee makes all of them.
  • Every no. When the gate recommends a decline, a partner agrees or overrides it in Slack before it counts.
  • The written reason on a no. A partner writes it. If it reads like a model wrote it, tell us. That’s a fair complaint and we’ll fix it.
  • Every reference call. A partner is on the phone. Software may transcribe or summarise the notes afterwards.
  • The conclusion in the memo. The AI writes the first draft. What the memo concludes is a partner’s call.

What the team sees

The analysis lands in Slack and Attio, where a partner makes every decision. Shown here with Coldline, our made-up test company.

In Slack

Every recommendation posts to #deal-review, our deal channel, where a partner acts on it.

The AI recommends Proceed

The recommendation, what the company does and a summary, linked to the record.

deal-review6 members
MessagesAdd canvasFiles & links
Tall OrderAPP10:04 AM
AI Recommendation: Proceed
Company:Coldline
Recommendation:Proceed
Supabase Deal ID:6513270e-269e-4d37-b2a7-4de452e6b438
LangSmith Run ID:9531985d-5d9d-49f8-9818-e811892f902b
Attio Company ID:36f675cc-81e7-4ef5-a8e2-5d940ed90475
What they do:Develops and operates AI software that automates export compliance documentation and evidence packs (cold-chain evidence, certificates) for chilled and live food exporters, charging per consignment cleared.
Summary:Coldline is a rare case where the founder is the user: a former QA manager who signed every export certificate at a seafood exporter is now automating that exact job. Three paying exporters, $180k ARR and a no-edit rate rising from 18% to 62% show the product works in production, but this is still a beachhead. The open questions are commercial and structural: how large the market is beyond AU seafood, who carries liability when a drafted certificate is wrong, and what the true cost to serve is.
Open in Attio

The gate recommends a pass

We removed the round slide, so the only raise in the deck was a future $25M Series A. The gate read that as this round. The partner disagrees, the run carries on, and the error is logged to the test set so the gate doesn’t repeat it.

deal-review6 members
MessagesAdd canvasFiles & links
Tall OrderAPP9:42 AM
AI Recommendation: Pass
Company:Coldline
Recommendation:Pass
Supabase Deal ID:6513270e-269e-4d37-b2a7-4de452e6b438
LangSmith Run ID:d23f0824-128b-4f33-8c5c-7fd0a6a3a450
Attio Company ID:36f675cc-81e7-4ef5-a8e2-5d940ed90475
Failed rule:Round size
What they do:Develops and operates AI software that automates export compliance documentation and evidence packs (cold-chain evidence, certificates) for chilled and live food exporters, charging per consignment cleared.
Why it failed the gate:The deck’s only mention of a raise is a $25M Series A planned for 2028, so the gate read this as a $25M round, which is over our $20M limit. Stage, location, and business model all fit our mandate.
AgreeDisagree
SH2 replies Today at 9:52 AM
Tall OrderAPP9:51 AM
Sam Henderson disagreed: “This is a $1.8M seed. The Series A slide is their plan for 2028, not this round.” The run picked up from the gate, and the research agents are running.
Logged against the deal with the partner’s name.
Tall OrderAPP9:52 AM
New error logged to the test setThe test set is every deal the gate has got wrong. Whenever we change the gate, we check the new version against it, and it only goes live if it gets every one right, this deal included.: the gate mixed up a future raise with this round.

In Attio

Attio is our CRM. The run fills in Coldline’s record, adds it to the founder pipeline and saves the memo as a note beside the founder’s enquiry.

Companies / ColdlineAsk Attio
OverviewNotes 2ActivityEmails 0Team 2Tasks 0Files
Highlights
AI Recommendation
Proceed
Analysis Stage
Discovery Brief
Notes 2
Activity
Tall Order AI added Company to Founder Pipeline12 hours ago
Tall Order AI changed 18 attributes in Founder Pipeline11 hours ago

Click a highlight to read it in full, or a note to open it.

Coldline

Coldline Analysis - Pre-Discovery

Coldline

Pipeline Stage: pre_discovery | Sources: Web enquiry, pitch deck, website (coldline.com.au), Founder Intelligence (114 citations), Market Trends research, Competitive Intelligence research, 1 comparable deal. Named Competitors were missing from the deck.

Investment Overview

Coldline automates export compliance for chilled and live food exporters, starting with Australian seafood. CEO Tess Halloran is a former QA manager who signed every export certificate at a seafood exporter. She is now automating that exact job. The AI sits where it is hardest to commoditise: physical-world sensor and photo data, destination-specific rules, and a liability-bearing workflow that a human still signs. Three paying exporters, $180k ARR and a no-edit rate rising from 18% to 62% show the product works in production, but this is still a beachhead. Customers are few and concentrated, and one is the founder’s former employer. No new logo is disclosed since Aug 2025, and the pipeline math does not reconcile. The open questions are commercial and structural: how large the market is beyond AU seafood, who carries liability when a drafted certificate is wrong, and what the true cost to serve is. The founders’ backgrounds check out against their LinkedIn profiles, the Coldline team page and Tess’s public talks. The company is raising a $1.8M seed.

Recommended next step: Progress

Key Metrics

  • ARR $180k (about $15k MRR), $80 per consignment, about 2,250 consignments a year
  • 3 paying exporters, about $60k average per exporter
  • $2m qualified pipeline across 11 exporters
  • Raise: $1.8M seed, valuation unknown; $350k pre-seed (June 2024)
  • Team: 5 named (2 founders, ML engineer, software engineer, field ops lead)

Assessment

Founder & Team Strong

Tess Halloran spent about 11 years in QA at Eyre Bay Seafood Co., the last five as QA Manager, and holds HACCP Lead Auditor certification. CTO Minh Tran spent six years building farm sensor networks at Paddockwise. Both backgrounds match their LinkedIn profiles and the website’s team page. Tess spoke on the AI in export compliance panel at the Southern Seafood Export Forum (Hobart, Nov 2025) and on the From the Wharf podcast (Oct 2025), and the 2023 agent she built for her own QA team is described on the Coldline blog. The product is GA with three paying exporters. Caveats: neither founder has prior founder experience, the equity split is unknown, and the first customer is Tess’s former employer.

Market & Problem Adequate

The problem is real: missing cold-chain evidence gets a consignment held, and the manual process takes about 90 minutes per consignment. Market sizing is the weak point. The deck estimates a $300B TAM (global cold-chain logistics), a $4B SAM and a $160M SOM, with no stated methodology. Low-confidence independent research suggests AU food compliance/traceability software is roughly AUD 0.2 to 0.35B. A software-only proxy gives about AUD 14 to 32M (IndexBox). Adjacent categories are growing at around 10 to 12% CAGR (Grand View Research). At about $60k per exporter, $50M ARR needs roughly 800+ exporters.

Competitive Advantage Adequate

Monitoring incumbents such as Sensitech, Controlant, Cooltrax, ELPRO and Berlinger own the logger layer but do not automate export documentation. Controlant is pharma-focused despite a $35M raise (Controlant). The closest Australian peers, FreshChain Systems and RegSoft, are grant-supported and early (Dept of Agriculture). Coldline’s difference is executing the consignment end to end. The deck names no competitors, and there is no partnership or customer comparison yet.

AI Depth Strong

The edge is in data and workflow, not the model call. There are five layers: a per-consignment agent, a regulator-notice LLM, a vision model fine-tuned on 2,600 labelled consignments, a time-series excursion forecaster and a holds agent. Every release is regression-tested against those 2,600 consignments. The work is regulated and physical-world, and a human signs. The vision model is described in detail on Coldline’s engineering blog (Sep 2026): it reads carton labels, ice coverage and seals from phone photos, and every release must match or beat the current model on the 2,600-consignment test set. Caveats: we haven’t seen a demo, the provider split is unknown, and the dataset is modest.

Traction & Validation Adequate

Three paying exporters: Eyre Bay (Jan 2025), Southgate Abalone (May 2025) and Coral Coast Prawns (Aug 2025), all three named on the website. One trade article (May 2025) says Coldline works with four exporters. The deck and website say three, and at the time it was two, so the deck’s figure is kept and the conflict noted. ARR of $180k clears the seed "$10K+ MRR" anchor. Churn, cohort and growth data are not disclosed. The $2m pipeline across 11 exporters implies about $180k per prospect, about 3x the current average. No customer-reported outcomes are given, and NDR is not reported.

Business Model Adequate

Usage-based pricing of $80 per consignment, charged on release, with no hardware and onboarding in weeks. Gross margin, inference cost and human labour per consignment are modelled, not proven. Model-specific risks are liability on a regulated certificate against a small fee (Tess said on the From the Wharf podcast that who carries the risk on a wrong certificate isn’t settled in contracts yet), pricing possibly anchored to labour rather than loss avoided, and seasonal, trade-exposed volumes. Fund return is Insufficient Data because valuation is unknown.

Red & Soft Flags to Address in Discovery

No soft flags were raised at screening. Key risks from the analysis:

  • Market ceiling: at about $60k per exporter, $50M ARR needs roughly 800+ exporters, which likely requires expansion beyond AU/NZ seafood into meat, dairy and horticulture and offshore exporter bases that is not yet evidenced. The deck’s $300B TAM and $4B SAM have no methodology.
  • Liability exposure: Coldline drafts regulated export certificates, and an error that causes a hold could destroy cargo worth far more than the $80 fee. Liability terms and insurance are still unresolved (a stated use of funds).
  • Customer concentration and possible founder-favour: only three customers, the first being the CEO’s former employer. The share of the 2,250 consignments from Eyre Bay is unknown.
  • Commercial velocity: no new customer disclosed since Aug 2025, and the end-2027 target of 20 exporters (about $1M ARR) is modest for a seed round, with v1 built in 2023.
  • Pipeline credibility: $2m across 11 exporters implies about $180k per prospect, about 3x the current average, which is unexplained.
  • Cost-to-serve opacity: there is no gross margin or inference cost data, and a dedicated field ops role may indicate human services embedded in delivery.
  • Volume exposure: per-consignment revenue is tied to seasonal seafood export volumes and to destination market-access shocks.
  • Source conflict: a May 2025 trade article reports four exporters against the deck’s three. The deck’s figure is kept, and the conflict is worth one question.

Recommendation: Progress

The strongest elements are a founder who lived the exact problem and an AI architecture whose edge is in proprietary sensor and photo data and a liability-bearing workflow rather than in prompts, backed by real paying customers. The discovery call must resolve the commercial questions this rating rests on: the unexplained pipeline math and slow new-logo pace, liability allocation, true cost to serve, and whether a credible path exists beyond AU seafood to support a venture-scale exit.

Discovery Agenda

Questions:

  1. Your $2m qualified pipeline across 11 exporters implies about $180k each, roughly three times your current $60k average. How is that figure built, and which exporter do you expect to sign first?
  2. When Coldline drafts a certificate that the QA manager signs and a consignment is then held or rejected because of an error in Coldline’s evidence pack, how is liability allocated in your three current contracts, and what are you trying to land in the standard terms?
  3. Walking through one consignment end to end, how much human time from your team (including Leah in Port Lincoln) goes into it today, and what does that make your gross margin per $80 consignment?
  4. Eyre Bay was your employer for 11 years. What did winning Southgate and Coral Coast look like with no prior relationship, and what has slowed new exporter signings since August 2025?
  5. For Coldline to be worth $200M+, it has to move beyond Australian seafood. How do you sequence the move into other chilled and live food categories versus offshore exporter markets like Chile, and what in the data layer carries over versus has to be rebuilt?
  6. What does the holds agent know today about clearing a hold at a specific port that a well-funded entrant couldn’t learn in its first year, and how many resolved holds has it learned from?

Founder Intel: Tess spent about 11 years signing export shipments at Eyre Bay Seafood Co. and studied Food Science at the University of Adelaide. On From the Wharf she called liability "the question we get first, every time". Ask what it was like being the person who signed every shipment, and how the agent she built for her own team became Coldline.

Market Intel: The closest Australian peers, FreshChain Systems and RegSoft, are grant-supported and early, with no visible commercial traction (Dept of Agriculture).

What We Still Need

  • Gross margin, inference cost per consignment and the human time inside delivery.
  • The basis of the $2m pipeline and how far along each of the 11 exporters is.
  • Liability terms in the three current contracts.
  • Revenue split across the three exporters, and Eyre Bay’s share of the 2,250 consignments.
  • Co-founder equity split, valuation, round structure and cap table.
  • How many resolved holds the holds agent has learned from.
Coldline

Coldline Inbound Enquiry

Coldline

Submitted through tallorder.vc/pitch. Deck attached and saved to the company record.

Your name
Tess Halloran
Email
tess@coldline.com.au
LinkedIn profile
linkedin.com/in/tess-halloran-coldline
Company
Coldline
Website
coldline.com.au
Where are you based?
South Australia
Why you?
I spent 11 years at Eyre Bay Seafood Co. in Port Lincoln, the last five as QA manager, signing every export certificate. The paperwork isn’t the hard part. The hard part is that every importing country sets its own evidence rules and changes them, and you find out when a consignment is held at the other end with live product on board. In 2023 I built an agent that drafted our certificates from the logger exports, and my team used it every day. Coldline is that agent rebuilt properly with Minh, my co-founder, and it now runs for three exporters.
Round size
$1.5M to $2M
Stage
Revenue, raising to scale it
Deck
coldline-deck.pdf (14 slides)
How did you find us?
LinkedIn
Privacy policy
Accepted

After the first meeting

New material, like a call transcript or a data room file, re-runs only the agents it affects. The memo says where each new number came from, such as “Tess, on the call”.

deal-review6 members
MessagesAdd canvasFiles & links
Tall OrderAPP2:14 PM
Coldline memo updated: four changes and one new flag.
Company:Coldline
Recommendation:Proceed (unchanged)
New context:Call transcript, 6 emails, cashflow.xlsx, deal lead’s note
Re-ran:Founder Intel, Competitive Intel, Synthesis A and B, AI committee
Biggest change:ARR is now $210k, up from $180k when the deck came in. Founder-stated on the call.
New flag:Runway is 7 months, at $45k a month.
Open in Attio
What came inA 46-minute call transcript, six emails with the Coldline team, cashflow.xlsx from the data room, and the deal lead’s note.
What ran againFounder Intel, Competitive Intel, Synthesis A and B, and the AI committee. Doc Intelligence and Market Trends were skipped.
What changedFour claims, including ARR up to $210k, and one new flag.

Click the note to open the updated memo.

Coldline

Coldline Analysis - Post-Discovery

Coldline

Pipeline Stage: post_discovery | New context: first-meeting call transcript (46 min), six emails with the Coldline team, cashflow.xlsx from the data room, deal lead’s note. Re-ran: Founder Intel, Competitive Intel, Synthesis A and B, AI committee. Skipped: Doc Intelligence, Market Trends.

What Changed

  • Burn and runway. Previously not disclosed. Burn is $45k a month and runway is 7 months (cashflow.xlsx).
  • Customer concentration. Eyre Bay is on a three-year contract, which softens the concentration flag. Founder-stated on the call; the contract hasn’t been seen yet.
  • Competitors. Tess says a US competitor is piloting in New Zealand. Added to the competitor map, founder-stated.
  • ARR. Now $210k, up from $180k when the deck was submitted (Tess, on the call). Founder-stated; the latest monthly revenue export will confirm it.

New Flag

  • Runway. 7 months at $45k a month. The round needs to close in time for the new hires to land before the money runs low.

Flags Kept

  • Liability, raised by the deal lead: “Liability terms still bother me.” Still no standard terms or insurance for a rejection the agent causes.

Customer Evidence

Southgate Abalone’s founder, quoted in an email from Tess: the agent “caught an excursion that would have cost us a shipment.”

Recommendation: Progress

Unchanged. The first meeting answered the runway question and strengthened the customer case. The open items for due diligence are the latest revenue figures, the Eyre Bay contract and liability terms.

What it learns from

The AI keeps no memory of its own. What makes it better over time sits in our databases, and survives every rebuild of the code.

Scoring criteria

Built from years of accelerator cohorts and portfolio companies we followed from application form to outcome.

Past deals

Every finished analysis is saved, and new deals are matched against the most similar ones. Our accelerator cohorts and earlier write-ups went in before the first live deal, so there were past deals to compare with from day one.

Outcomes

Each deal is logged as a win, a loss or a pass. As bets resolve, we recalibrate the scoring against what happened, and when a partner overrides the AI we record who turned out to be right.

The criteria had years of history from the start. Outcome tracking is newer, and improves as more deals resolve.

Roadmap

Everything above runs on every deal today. None of these runs on deals yet.

Launching Q4 2026

Founder chatbot In progress

Founders can apply by talking it through instead of filling in a form. The conversation goes into the same analysis as every other deal.

Launching Q4 2026

Sourcing Signals v2 In progress

Finds founders before they apply or announce anything, from new company and domain registrations, GitHub activity, patents and research grants, people leaving scale-ups, and accelerator cohorts. Firecrawl Alexandria’s profile and code data is one of the planned sources.

Launching Q4 2026

Deal tracking In progress

Keeps watching deals we passed on or put on watch, and flags one for another look when something happens that could change our view.

Launching Q1 2027

Technical due diligence Scoping

Checks on the technology, for deals that get past a first meeting.

Launching Q1 2027

Financial due diligence Scoping

Checks on the financials, for deals that get past a first meeting.

Changelog

DateTypeChange
5 Oct 2026ImprovedPre-screening calibration. Revised the pre-screening prompt against deals the team flagged, so fewer clear mismatches come back as More Info.
5 Oct 2026Bug fixesGeography pass reason. Deals that met our location rule were sometimes given International as their pass reason anyway. That reason now only goes on deals that actually fail the location rule.
5 Oct 2026ImprovedReport language. Simplified the language in re-analysis notes and investment recommendation reports, and now check them with our Humanizer agent.

Full changelog →