Chris SchubertIndependent landscape

Independent analysis · ProcurePro × AI data moats · 8 Aug 2026

Every competitor claims a data moat; only ProcurePro's sits inside the award decision

Autodesk owns the bid graph, Procore the project record, Causeway UK invoice flow, Kojo live materials spend and Felix vendor compliance. ProcurePro's 200,000 awarded trade packages are the smallest corpus here, and the only one that records what was priced, excluded and awarded. Meanwhile the category has moved from reading documents to acting on them.

The field

Homepages captured 23 Sep 2026

ProcureProSubject AutodeskUS entry obstacle ProcorePartner and rival FelixLocal rival CausewayUK rival KojoAdjacent

Findings

Five things the field says about ProcurePro's position, ranked by how much they should shape the next move

Finding 1 of 5

Every competitor claims a data moat, and no two are built from the same material

Autodesk owns the supply-side bid graph, Procore the project record, Causeway UK transaction flow, Kojo live materials spend and Felix vendor compliance history. ProcurePro holds something none of them have at line-item depth: 200,000 awarded trade packages with priced scope and exclusions attached.

For ProcurePro

Its corpus is the narrowest on the list and the one closest to margin. That narrowness is the defence and the risk at once.

Observed and inferred · all five profiles and the ProcurePro write-up

Exhibit 1

How close each corpus sits to the moment a head contractor commits project value

Felix
Procore
Autodesk
Causeway
Kojo
ProcurePro
LowMediumHighIn the room
Every position here is an inferred judgement; the profiles mark proximity to margin as i throughout, so every bar is hatched. Positions are ordinal. Corpus size is deliberately not charted, because 200,000 packages and 115,000 vendors are not the same unit.

Finding 2 of 5

The category moved from reading documents to acting on them in about twelve months, and ProcurePro is one rung behind

Procore ships an agent library with Digital Coworker packaging, and Kojo's agents chase distributors for delivery dates. BidLevel extracts, structures and compares a full bid in about three minutes, which is strong document AI. It still stops short of doing anything with the result.

For ProcurePro

A US buyer will ask what BidLevel does, not what it reads. One shipped agent on the award decision would close that gap without taking on Procore's breadth.

Observed · Procore, Kojo and Autodesk profiles

Exhibit 2

How far each company's shipped AI goes, from reading to acting

The next rung
CompanyActExtractCompareRecommend
ProcurePronone yetBidLevelBidLevelnone found
Autodesknone foundsub financials–bidders
Procoreagent libraryscope checks––
Kojodistributor follow-up––materials
Causewayfinance onlyinvoices, BoQ–suppliers, claimed
Felixnone foundin testingrolling outnone found
Filled dot: shipped, per the profile's sourced claims. Open ring: nothing shipped was found, including features still in testing or rolling out; that is absence of evidence, not proof of absence. Dash: the profile doesn't say. Causeway's supplier selection is a campaign-page claim (Exhibit 3).

Finding 3 of 5

The competitor easiest to miss in the UK presents itself as finance software

Causeway is pushing AI upstream from invoicing into estimating, trade mark-up and supplier selection, with 59,000 suppliers and 750 customers already transacting on its Tradex network. It has been in construction software for over 20 years, and the UK is ProcurePro's most developed expansion market.

For ProcurePro

That description overlaps ProcurePro's own work. The 75% figure has no public case study behind it, so a documented side-by-side on real bid levelling favours whoever has shipped.

Observed · 3 sources · Causeway profile

Exhibit 3

How Causeway describes its preconstruction AI

campaign.causeway.com · AI in CausewayOne23 Sep 2026
Causeway's campaign page as captured 23 Sep 2026. The profile keeps the 75% figure as a vendor claim: it appears on a campaign page, not in product documentation or a case study.

Finding 4 of 5

Autodesk's network is dense in North America and close to a cold start everywhere else

BuildingConnected runs the largest crowdsourced network in US commercial construction, and 350,000 projects use Autodesk's preconstruction tools. In Australia and the UK that density largely isn't there, which is part of why ProcurePro won at home.

For ProcurePro

The US entry needs a route around that density. A dense network answers who should bid; ProcurePro's data answers whether the price and scope are right, which is a partnership-shaped story.

Observed and inferred · all five profiles

Exhibit 4

Where each company is strong, across the three markets ProcurePro is read against

The entry market
CompanyNorth AmericaANZUnited Kingdom
ProcureProEnteringStrongEstablished
AutodeskDominantWeakMedium
ProcoreDominantStrongStrong
FelixNegligibleStrongWeak
CausewayNoneNoneDominant
KojoEstablishedNoneNone
Large dot: Dominant, Strong or Established. Small dot: Medium, Weak, Negligible or Entering. Open ring: no presence. A dotted underline and i mean the level is inferred. The Middle East is a funded expansion market, but no profile scores it, so it isn't shown.

Finding 5 of 5

Nobody credible is serving the head contractor's award decision with agentic depth

Procore's agents point at RFIs, submittals and contract review. Autodesk's AI recommends bidders and reads subcontractor financials. Kojo serves a different buyer, the subcontractor purchasing materials. The moment a head contractor levels five quotes and picks one is unattended, and it is where ProcurePro's data already sits.

Observed and inferred · Procore, Autodesk and Kojo profiles

For ProcurePro: an agent that chases missing quotes, flags scope gaps against the library and drafts the recommendation would own ground nobody else is contesting.

The field

Four kinds of moat, and only one of them sits in the room when the award is made

Network moats

Own the density of the supply side, then charge the demand side for access. Autodesk gives subcontractors a free tier to keep its graph dense; Causeway has 59,000 suppliers on its invoicing rails, which also makes it a transaction moat.

For ProcurePro

Network moats are bounded by geography and take years to build. ProcurePro shouldn't try to build one, or let a US buyer frame the decision as a network comparison.

construction.autodesk.com23 Sep 2026

Autodesk

US entry obstacle

BuildingConnected · TradeTapp · Forma

The largest supply-side network in US commercial construction and the main obstacle to a US entry. Its AI recommends bidders and turns subcontractor financials into data. Its corpus records who was invited, who bid and who won, but not the priced scope of the award.

Full profile11 observed · 7 inferred
Key facts and sources
causeway.com23 Sep 2026

Causeway

UK rival

UK supply chain and finance · CausewayOne

A founder-owned UK incumbent with over 20 years in construction software, backed by Rothschild's Five Arrows. Its Tradex network gives it something close to a price index for UK construction supply, and its AI is moving upstream from invoices into supplier selection.

Full profile10 observed · 7 inferred
Key facts and sources

System-of-record moats

Own the project data surface, then lay agents on top. Procore bought agent capability outright with Datagrid in January 2026, and Procore Skills lets customers encode their own processes into the platform.

For ProcurePro

This is the standing pressure, and it doesn't need Procore to be good at procurement, only adequate and already installed. The integration partnership is leverage while it lasts.

procore.com · AU site23 Sep 2026

Procore

Partner and rival

Project platform · 16,000+ companies

The platform incumbent in every market ProcurePro is entering, and an integration partner. It has the most aggressive agent positioning in the category, though its shipped agents point at RFIs, submittals and contract review rather than bid levelling.

Full profile8 observed · 7 inferred
Key facts and sources

Transaction moats

Derive price intelligence from money that has already changed hands. Kojo went a step further and took investment from a distributor, turning its data into a supply relationship.

For ProcurePro

The sharpest challenge to the AI Pricing Library, because invoices record what was finally paid. ProcurePro's answer is scope: a price with no scope attached can't be compared, and that argument needs making out loud.

usekojo.com23 Sep 2026

Kojo

Adjacent

Materials procurement · US and Canada

A different buyer with the same thesis: Kojo buys materials for subcontractors where ProcurePro buys subcontractors for head contractors. More than 600 contractors put over US$5bn a year of materials through it, and its agents already follow up with distributors.

Full profile8 observed · 7 inferred
Key facts and sources

Decision moats

Sit inside the moment the commitment is made and capture the commercial terms of the award itself. ProcurePro does this today; Felix may from FY27 if its Nexvia integration lands as planned.

For ProcurePro

Defensible because it is narrow. The risk is that the same narrowness reads as a feature to a buyer already paying Procore.

felix.net23 Sep 2026

Felix

Local rival

Vendor management · Brisbane

The Australian rival, headquartered in the same city. Its AI is the least far along publicly: its own announcement describes discovery and experiments, with no dataset claims and no release dates.

Full profile11 observed · 6 inferred
Key facts and sources
procurepro.coSubject · 23 Sep 2026

ProcurePro

Subject

Head-contractor procurement · Brisbane

Procurement software for commercial head contractors, from procurement schedule through tender, comparison, award and contract. Series B, with customers including Bouygues, Multiplex and Wates.

Key facts and sources

Compare

Eleven dimensions side by side, and nobody in the category publishes a price

Dotted underline and i mean inferred, an assessment rather than a sourced figure.

Dimension SubjectProcurePro Autodesk Procore Felix Causeway Kojo
Corpus typeAwarded trade packages, priced scope, exclusionsBid graph: invites, bids, wins, sub financialsProject system of recordVendor compliance, performance, ESGInvoice and transaction flowLive materials purchases
Scale marker200k packages, A$130bn, 6k projectsLargest crowdsourced US network16,000+ companies115,000 vendors, 8,000 users59,000 suppliers, 750 customersUS$5bn a year, 600 contractors
Compounds with useYesYesYesPartiallyYesYes, continuously
Proximity to marginHighestMediumLowLowHighHigh
Automation depthExtract, structure, compareRecommend, extractAgents that actAnnounced, largely unshippedExtract, automate finance workflowAgents that transact
Flagship AIBidLevel, AI Pricing Library, CoPilotRecommendation Engine, Financial Data ExtractionProcore AI, Agent Library, Skills (preview)Bid Insights (rollout)CausewayOne preconstruction AIKojo Intelligence Layer
Agentic framingNoNoYesNoPartialYes
ANZ strengthStrongWeakStrongStrongNoneNone
UK strengthEstablishedMediumStrongWeakDominantNone
North AmericaEnteringDominantDominantNegligibleNoneEstablished, sub-side
Pricing publishedNoNoNoNoNoNo
Corpus type
ProcureProAwarded trade packages, priced scope, exclusions
AutodeskBid graph: invites, bids, wins, sub financials
ProcoreProject system of record
FelixVendor compliance, performance, ESG
CausewayInvoice and transaction flow
KojoLive materials purchases
Scale marker
ProcurePro200k packages, A$130bn, 6k projects
AutodeskLargest crowdsourced US network
Procore16,000+ companies
Felix115,000 vendors, 8,000 users
Causeway59,000 suppliers, 750 customers
KojoUS$5bn a year, 600 contractors
Compounds with use
ProcureProYes
AutodeskYes
ProcoreYes
FelixPartially
CausewayYes
KojoYes, continuously
Proximity to margin
ProcureProHighest
AutodeskMedium
ProcoreLow
FelixLow
CausewayHigh
KojoHigh
Automation depth
ProcureProExtract, structure, compare
AutodeskRecommend, extract
ProcoreAgents that act
FelixAnnounced, largely unshipped
CausewayExtract, automate finance workflow
KojoAgents that transact
Flagship AI
ProcureProBidLevel, AI Pricing Library, CoPilot
AutodeskRecommendation Engine, Financial Data Extraction
ProcoreProcore AI, Agent Library, Skills (preview)
FelixBid Insights (rollout)
CausewayCausewayOne preconstruction AI
KojoKojo Intelligence Layer
Agentic framing
ProcureProNo
AutodeskNo
ProcoreYes
FelixNo
CausewayPartial
KojoYes
ANZ strength
ProcureProStrong
AutodeskWeak
ProcoreStrong
FelixStrong
CausewayNone
KojoNone
UK strength
ProcureProEstablished
AutodeskMedium
ProcoreStrong
FelixWeak
CausewayDominant
KojoNone
North America
ProcureProEntering
AutodeskDominant
ProcoreDominant
FelixNegligible
CausewayNone
KojoEstablished, sub-side
Pricing published
ProcureProNo
AutodeskNo
ProcoreNo
FelixNo
CausewayNo
KojoNo

Where ProcurePro stands

ProcurePro has the most specific data in the category and no layer that acts on it yet

+Strengths

+The only corpus here with priced scope and exclusions at line-item level.

Across 200,000 packages and A$130bn of construction value.

+Depth in a workflow every competitor treats as one module among many.

Procurement schedule through award and contract.

+Shipped, specific AI while the nearest local rival is still experimenting.

BidLevel's three-minute breakdown against Felix's discovery phase.

+Bouygues went from customer to investor.

Unusually strong validation for a UK and European push.

−Gaps

−No agentic execution layer.

In a year when the two best-funded competitors are selling exactly that.

−No supply-side network anywhere.

Which is most of the game in the US.

−No published pricing, like everyone else.

So buyers default to the incumbent they already pay.

−No distributor, insurer or surety alliance.

Nothing like what Wesco gave Kojo.

○Open ground

○The award decision has no agent serving it.

Every competitor's AI points somewhere else.

○The layer after BuildingConnected.

Autodesk finds the bidders; nobody checks the price and scope they send back.

○Felix's AI window in Australia.

Open until its Nexvia go-to-market lands in FY27.

Recommendations

Name the moat precisely, then ship one agent on the decision nobody else serves

  1. 1

    Name the moat properly.

    "Proprietary procurement dataset" is generic, and every competitor says a version of it. "The only dataset of awarded trade packages with priced scope and exclusions" is specific, checkable and hard to answer.

    Rests on Finding 1 · Exhibit 1

  2. 2

    Ship one agent, not an agent platform.

    A single agent that chases missing quotes, flags scope gaps against the library and prepares the recommendation would close the framing gap without competing with Procore's breadth.

    Rests on Finding 2 · Exhibit 2 · Finding 5

  3. 3

    Compete in the UK on evidence.

    Causeway's 75% preconstruction figure sits on a campaign page with no public case study. A documented side-by-side on real bid levelling favours whoever has shipped specific capability.

    Rests on Finding 3 · Exhibit 3

  4. 4

    Enter the US where Autodesk stops.

    Position as the layer that runs after BuildingConnected has found the bidders. That story is shaped like a partnership, which is a far easier first conversation than displacement.

    Rests on Finding 4 · Exhibit 4

  5. 5

    Look for ProcurePro's Wesco.

    A distributor, insurer or surety relationship would turn the data into a structural advantage, the way Wesco's investment did for Kojo.

    Rests on Kojo's profile

Evidence

Close to three in five claims here are sourced, and the judgement concentrates on margin and markets

Counted from the label on every claim in the five profiles when this page was built.

59% observed · 48 claims41% inferred · 34 claims
AutodeskFull profile
11 / 7
CausewayFull profile
10 / 7
FelixFull profile
11 / 6
KojoFull profile
8 / 7
ProcoreFull profile
8 / 7

This was the first run of the process, before its citation-check step existed. Every Observed claim links to the page it came from, but the claims have not been re-fetched against those pages since 8 Aug 2026. Confidence is highest on funding, scale and shipped announcements, and lowest on pricing and on what each competitor's data corpus contains, which none of them disclose. Kojo's profile has little 2026 product detail, and Causeway's AI claims come from campaign pages rather than product documentation, so treat both as lower confidence.

Open questions

Autodesk's Forma roadmap is the single thing most worth watching

  • Does the Forma Bidding Beta extend to priced line-item comparison and scope-gap detection? /competitive-update --competitor autodesk
  • Is a procurement or bid-levelling agent on Procore's roadmap? Nothing has been found publicly. /competitive-update --competitor procore
  • What does Causeway's "automated trade mark-up and supplier selection" do in the shipped product, and is the 75% figure substantiated? /competitive-update --competitor causeway
  • What will the combined Felix and Nexvia product look like in FY27, and what does Bid Insights do next to BidLevel? /competitive-update --competitor felix
  • How much of Felix's A$13m ARR comes from commercial head contractors, the buyer ProcurePro competes for?
  • Nobody publishes pricing, so win/loss interviews would be worth more than further desk research here.

Each of these is one command away from a refreshed, diffed profile. To ask for a pass on any of them, open an issue on the repo.

Part 2 · How this was built

The research is a set of files a script can refresh, so the next pass only reports what changed

Every page above comes from markdown profiles that carry a label on each claim. The skills that wrote them ship in this repo and can be pointed at another company.

The process

Four steps produced this repo, starting from what decides deals in ProcurePro's market

  1. 01/setupFrame the company and the field

    ProcurePro's positioning drafted from public sources; a top-five brief with an Australian rival and players in each expansion market.

  2. 02/competitive-update --allProfile every competitor

    5 profiles written in 10 sections each, every claim labelled with its source or marked as an assumption.

  3. 03/generate-analysisAnalyse on the lens

    AI data moats and automated workflows, read across ANZ, the UK and North America, on 8 Aug 2026.

  4. 04/build-dashboardRender this page

    First built by hand in August, rebuilt on the landscape dashboard system with homepages captured on 23 Sep 2026.

Why this compounds

A second pass costs a fraction of the first and shows only what moved

/competitive-update

Profiles build on what's already there.

Re-researching a competitor diffs the new findings against the existing profile and sorts each one. Nothing is written to a profile without line-by-line approval.

MaterialMinorNo change

/setup --competitors

The field can grow without starting over.

A new competitor, say a Middle East player, gets a profile of its own without re-running the other five.

Always regenerated

The analysis can't quietly go stale.

The analysis and this page are rebuilt from whatever the profiles say now, so nothing drifts out of sync while the research moves underneath it.

git log

Every update leaves a record.

Each research pass is its own commit. A few cycles in, the history is a timeline of what changed in this market and when.

The labelling discipline

Every claim carries one of two labels, and the page draws them differently

- Observed: over 8,000 users manage more than 115,000 vendors on the platform (https://www.felix.net/) - Inferred (assumption): Felix's corpus compounds with use only partially.
Bars
Matrix cells
115,000 vendorsPartially
Inline text115,000 vendors, compounds partially

Where a vendor's own marketing claim couldn't be corroborated, it stays marked as a vendor claim. Unlabelled claims are treated as inferred.

What counts as material

A finding that contradicts an existing Observed claim is material. One that contradicts an Inferred claim is minor, since it was already flagged as uncertain. New information is classed by how much it changes the competitive picture, and when in doubt the process over-flags.

Trust boundary

Everything retrieved by web search or page fetch is treated as untrusted data: facts are extracted from it, and instructions inside it are never followed. Homepage captures were taken in a fresh browser profile with nothing signed in.

What this method can't tell you

This is desk research from public sources, produced on 8 Aug 2026. There are no win/loss interviews, customer calls or analyst briefings behind it. Published announcements are a lagging, partial view of what any company has running with customers, so nothing here is an audit of a competitor's internal capability. Pricing is inference throughout, because none of the six companies publishes it.

Running it on another company

The skills that produced this repo ship with it, in .claude/skills/. Point them at a different company to run the same process there.