Small & medium business

Lots of firms, clear professional obligations, and a buyer who already pays about A$30 a seat for cloud AI. The most crowded space on this site.

At a glance

A big market that is thin at the top.

2.81M
Australian businesses. 65% employ nobody; only 68,325 have 20–199 staff.
ABS, June 2026 Verified
~3,700
Firms with 20–199 staff in professions with confidentiality duties
ABS class data, our cut Estimate
65%
Of SMEs not using AI cite distrust of AI decisions or wanting to keep human control
National AI Centre, 2026 Verified
48%
Of Australian workers admit using AI in ways that go against company policy
KPMG / Uni of Melbourne 2025 Verified
Who they are

Two rings of buyers.

Firms with 20–199 staff, by vertical
ABS businesses by industry class and employment size, June 2025. Dark bars: professional obligations. Light bars: commercial confidentiality.

Ring 1: obligation-heavy

  • Legal, accounting and tax, financial advice and broking, insurance broking, GP, specialist, dental, physio and allied health including psychology
  • ~3,730 firms with 20–199 staff, and ~25,300 with 5–19
  • 16,793 private law practices, 78% of them sole practitioners; ~7,135 accredited general practices; 15,065 financial advisers

Ring 2: confidentiality-heavy

  • Engineering, architecture, design, advertising, recruitment, market research, scientific research, management consulting
  • ~4,900 firms with 20–199 staff, and ~13,000 with 5–19
  • The pull is NDAs, client IP and tender data. No regulator is pushing.

Our cut of ABS data. The ideal buyer is a 20–199 person practice: it has a practice manager to sponsor it, a real document library, and partners who worry about privilege and PI claims.

Why they'd want Selma

Regulators say "not public tools". None of them say "on-premises".

Selma's edge here is certainty and a simple story ("it never leaves the building, and the Ledger proves it"). The law doesn't require it. Many firms will judge a no-training cloud tool good enough.

VerticalHookWhy
Tax agents and accountantsStrong, newTPB guidance (22 July 2026): entering client information into AI tools can be disclosure to a third party, which needs each client's permission. On-prem removes the consent step. ~4,100 accounting firms with 5–199 staff.
Law firms (litigation, family, criminal, IP)StrongJoint regulator statement (Dec 2024); NSW Supreme Court Practice Note SC Gen 23 on suppressed and subpoenaed material; court AI protocols in almost every jurisdiction; the US Heppner ruling that public AI use waived privilege.
Psychology and counsellingStrongHighly sensitive session content, always covered by the Privacy Act.
Allied health / NDIS providersMedium-strongOverlaps our pilots; report-heavy work.
GP and specialistsMediumClear obligations, but ~40% of GPs already use cheap AU-hosted cloud scribes. Selma fits back-office and governance, not scribing.
Advisers, brokersMedium-weakPrivacy and trust; no AI-specific rule found. Price-sensitive micro-licensees.
Engineering, architecture, agencies, consultingCommercial onlyContractual confidentiality; Copilot usually acceptable.

Shadow AI is the real buying trigger. A partner who knows staff are pasting client files into free ChatGPT has a live professional indemnity risk today. The pitch is less "adopt AI" and more "give people a sanctioned tool so they stop using the unsanctioned one".

The Privacy Act's small business exemption (turnover under A$3M) still applies outside health, so a privacy-law pitch has little force for many 5–19 person firms. Professional rules apply regardless.

Who else is in the room

Crowded at every level.

Cloud assistants

ProductA$/user/month25 users, 3 yearsNote
Microsoft 365 Copilot Business31.40 (26.91 promo)~A$28KUnder 300 seats; needs a base licence
ChatGPT Business~28~A$25KNo training on business data by default
Claude Team~30~A$27KNo training by default
Copilot Enterprise44.90~A$40KOver 300 seats
Perplexity Enterprise Pro~61~A$55K

Vertical AI

ProductPrice (estimates where unpublished)Note
Harvey~A$1,800–3,000 per seat/month, ~25 seat minimum EstimateTop-tier firms (Clayton Utz, Gilbert + Tobin) in an Australian environment
Legora~A$460–1,200 per seat/month Estimate
LexisNexis Protégé, Thomson Reuters CoCounsel~A$340–760 per seat/month EstimateNeed Lexis+ or Westlaw. A 10-lawyer firm pays ~A$137K–274K over 3 years.
LEAP, Smokeball, ClioBundledOwn the small-firm relationship: channel or threat
Xero JAXIncluded in A$78–143/month plansLedger-level AI is effectively free
Lyrebird, Heidi, NovoNoteFree tiers; A$30–160/monthScribing is crowded and cheap

Private and local AI

OfferPrice (A$)What you get
Ollama, LM Studio, AnythingLLM, Open WebUI, JanFreeThe runtime. No ingestion quality, access control, audit or support.
Sandlabs (Melbourne)From 2,500 + GSTDone-for-you private AI setup on the client's hardware. Law, accounting, health, brokers.
AI Compute Australia (Melbourne)6,500–12,500+On-prem inference appliances plus an Australian-hosted API
Nexus Tech, FlowWorksQuotedPrivate AI builds for health, legal and finance SMEs
Custom LLM (Melbourne)80K–150K upfront + 3K–8K/monthEnterprise on-prem builds
NVIDIA DGX Spark / ASUS GX10~9,000–10,800128GB memory, runs 70B+ models
Ryzen AI Max+ mini PCs (Framework, GMKtec, Beelink)~5,200–6,600128GB unified memory
Olares One~7,600Consumer "own your AI" box, RTX 5090 mobile
Go Abacus Go1, Zylon in a BoxNot publishedOverseas appliances for regulated enterprise

The spec sheet problem. For A$5K–10K a firm or its IT provider can buy a box with 128GB of model memory. Selma's 16GB card runs 8–14B models. Side by side on paper, Selma looks like the weaker machine. Nobody else visibly offers the Constitution, citations, audit ledger, measured accuracy and sector packs. Memory prices also rose sharply in 2026, so the A$3K build cost needs re-quoting.

What price could look like

A flat monthly fee, not a seat price.

A$100 per user per month is about three times Copilot Business and would fail here. One option explored: a packaged "Selma Practice" tier, sold through channels, with commissioning cut to 2–3 days using templated Constitutions per profession.

Three-year cost for a 25-person firm
A$, ex GST. Selma tiers are illustrative.
Illustrative tierForSetup incl. unitMonthly3 years
Practice 105–19 staff, one siteA$6,000A$400A$20,400
Practice 5020–50 staffA$9,500A$750A$36,500
Business 15050–150 staff, bigger hardwareA$15K–20KA$1,200A$58K–63K

Monthly fee would include monitoring, model updates, annual re-measurement and hardware swap. Profession packs (court practice notes, TPB and ATO rulings, RACGP and Ahpra standards) as add-ons at perhaps A$100–250 a month each.

Channels, because one-at-a-time doesn't pay

ChannelWhyWatch out for
Managed service providersHold the IT relationship and are being asked "what do we do about AI?"They can build a DIY box. Selma needs to offer margin, training and a partner portal.
Practice software (LEAP, Smokeball, Best Practice, Cliniko, Xero apps)Own the daily workflow and dataBuilding their own AI; may see Selma as a rival
Professional bodiesPublish the guidance that creates the hook; member benefitsSlow, careful about endorsement
PI insurers (LawCover, Avant, MDA National)Carry the cost of breaches; a premium discount would be powerfulLong cycles, prefer neutral guidance
IT distributors (Dicker Data, Ingram Micro)Reach thousands of resellersOnly once it truly self-installs
Tensions to talk about

For

  • Clear, current rules: TPB, court practice notes, Law Society guidance, OAIC, Ahpra
  • "Human at the Helm" speaks directly to the 65% who want control
  • Flat cost suits firms with many part-time and support staff, and overnight file review
  • Commercial margins could fund mission work: NFP and Pacific deployments
  • Health and social assistance is the fastest-growing industry (+6.7%)

Against

  • Brand drift. Selling boxes to mortgage brokers could make Selma read as another AI vendor, and unsettle NFP partners or DFAT. A sub-brand is one answer.
  • Price anchors at ~A$30 a seat and A$2,500 setups
  • Frontier models write better, and word travels fast in professional networks
  • "Copilot is already inside our tenancy"
  • Support load across many small sites with no in-house IT