Finnish accounting firms in 2031 — where the productivity goes when nobody passes it on

What the industry looks like in 2026

Finland's accounting service industry covers roughly 4,000–4,500 firms, employs 12,867 people, and turned over €1.48 billion in 2024¹. Around 800 firms hold authorised-membership credentials through Taloushallintoliitto². Top-5 combined share sits at around 10% — concentrated enough to set tone, fragmented enough that no single firm dictates pricing.

The supply side sorts into eight structural positions. Three listed groups (T1) anchor visibility: Talenom, Aallon Group, Administer. A PE-roll-up tier (T2) consolidates aggressively; Norvestor's Rantalainen alone has completed roughly 120 acquisitions over seven years, ~2,000 employees across Finland and Sweden³. Mid-tier specialists (T3) like Greenstep run cloud-native at 14–20% EBITDA. Small independents (T4), the 800–1,200 firms with 5–20 employees, are where the owner-CEO conversation is most acute. The toiminimi tier (T5) sits in the long tail. Platforms (T6/T7) — Procountor under KKR, Visma's Netvisor, Fennoa — sell into every other tier and increasingly to end-clients directly. In-house corporate finance departments (T8) sit on the demand side as a partial substitute.

flowchart TD
    Client["End-client: Finnish SME or corporate"]
    Client --> T1["T1 Listed groups
Talenom · Aallon · Administer"]
    Client --> T2["T2 PE-roll-ups
Aspia · Azets · Rantalainen"]
    Client --> T3["T3 Mid-tier specialists
Greenstep class"]
    Client --> T4["T4 Small independents
5–20 employees"]
    Client --> T5["T5 Toiminimi
long tail"]
    Client -.SME-direct.-> T67["T6/T7 Platforms
Procountor · Netvisor · Fennoa"]
    T1 --> T67
    T2 --> T67
    T3 --> T67
    T4 --> T67
    T5 --> T67
    T8["T8 In-house corporate finance
(demand-side substitute)"] -.substitutes for.-> Client
    classDef client fill:#0C1B3D,stroke:#0C1B3D,color:#FFFFFF;
    classDef firm fill:#F7F7F5,stroke:#0A7D70,color:#1F2D3D;
    classDef platform fill:#F7F7F5,stroke:#1858B8,color:#1F2D3D;
    classDef inhouse fill:#F7F7F5,stroke:#E07A5F,color:#1F2D3D;
    class Client client;
    class T1,T2,T3,T4,T5 firm;
    class T67 platform;
    class T8 inhouse;
  

The structural decisions of 2024–2026 priced the divide between service and software in the open. KKR paid more than €1 billion for the software portfolio that came out of the pre-split Accountor (the brand now split into Aspia for T2 services and KKR's software estate of Procountor, eTasku, Mepco, Apix, Isolta)⁴. The implied revenue multiple is at least ~7.5× on the €132M software-line revenue⁴. Talenom's own Easor demerger completed in Q1 2026: €20M+ ARR, 15,000 SME clients on the software side, 67.7% EBITDA margin in the most recent disclosed quarter⁵. Two separate boards, on >€1bn-equivalent value, decided the software inside the firm was worth more outside it. That is the structural setting the next five years play out against.

The boundary between service firm and software firm hasn't blurred — it's been priced and crossed.

The capability stratum — what AI can actually do for accounting work by 2031

In this piece we distinguish three strata of AI in accounting work: assistant tier (S-Assistant — AI working alongside the bookkeeper), agent tier (S-Agentic — agent handles the workflow, human checks at milestones), and autonomous tier (S-Autonomous — AI runs continuous accounting, human checks only in exception cases). The distinction keeps their economics, failure modes, and regulatory implications separate.

Accounting work splits into three AI-capability strata. S-Assistant is human-in-the-loop AI: categorisation suggestions, OCR (receipt text recognition), anomaly flags, auto-filled journal entries with a bookkeeper checking each one. By 2026 it is table-stakes. Taloushallintoliitto itself ships a paid product covering tax-authority guidance, KILA opinions, accounting law and payroll; the spring 2025 trial drew roughly 1,400 testers from a base of ~800 member organisations⁶. When an industry association ships its own software, the technology is no longer a differentiator.

S-Agentic is the live frontier in 2026. An agent handles a workflow end-to-end (month-end close, ALV period, payroll preparation, tilinpäätös first draft) and a bookkeeper reviews at milestones rather than at every line. Rantalainen's May 2025 partnership with Helsinki-domiciled Taxxa.ai is the canonical Finnish production-scale deployment: a six-person vendor team, more than 250 firms across six countries, knowledge base covering 30,000+ pages of Finnish tax and payroll documentation⁷. The economic signal underneath the deployment matters as much: three years ago the product would have required 30+ engineers and a $20–40M Series A; in 2026 it runs on six people and bootstrap funding. That cost collapse is what makes the "subscribe" path economic for a T4 firm.

S-Autonomous (AI runs continuous accounting with human intervention as exception only, audit-grade evidence trail) does not yet exist in production at scale. Vic.ai's "Autonomous Finance" is branding around a roadmap with named agents (VicAgents, VicPay, Vendor Portal launched June 2025), not a deployed continuous-close service⁸. The realistic horizon is narrow-domain S-Autonomous at vendors with operational discipline by 2028; broad-domain at the optimistic end by 2030–2031, beyond 2031 at the realistic end.

The strata are not different intensities of one wave. They have different failure modes, governance footprints, talent implications and pricing models; the foresight only works when they are kept separate.

Per-tier outlook to 2031

Each tier captures the productivity gain differently because each holds a different position in the value chain.

T1 listed groups bifurcate. Talenom plus Easor, demerged Q1 2026, is the prototype of the separated path: service business at a service multiple, software at a software multiple. Aallon Group and Administer remain the live counter-examples requiring explanation. Aallon Group builds growth through acquisitions (four in H1 2025, around 1,000 Finnish firms identified as potential targets⁹) without a separate quantified software thesis. Administer holds to a bundled model in which the eFina platform, services and HR are sold together. The 2031 question for both is whether they unbundle on the Talenom-Easor pattern or articulate a durable-bundle thesis. A third independent unbundling event by end-2028 would settle the structural reading.

T2 PE-roll-ups continue compounding and start deploying agentic AI at scale. Aspia, the rebranded former Accountor services business, runs about 3,100 experts across six countries⁴. Azets Insight Finland turned over €88.8M with 789 employees in 2024¹⁰. Rantalainen × Taxxa.ai is the canonical specialist-vendor partnership pattern. The captured gain depends on post-acquisition integration discipline. That is where T2 economics live.

T3 mid-tier specialists are the most scenario-resilient tier. Greenstep — €66.9M revenue FY ending May 2024, 613 employees at year-end, ISO/IEC 27001:2022 certified April 2024, around 800 staff by 2025 across eight countries with 3,000+ clients¹¹ — has the service-mix breadth that protects it. Advisory and CFO-as-service work compresses less than 20% under mature S-Agentic; firms that already weight toward it gain share from compressing transactional incumbents.

T4 small independents (5–20 employees) bifurcate sharply. The undifferentiated mid-band T4 firm — no vertical specialism, no customer intimacy, no AI-native posture — faces a Base scenario of 75–90 revenue in 2031 (2026 = 100), 4–10% EBITDA, headcount at 50–65. The differentiated T4 firm holds one of three durable positions:

• regional customer intimacy on a long-term book;
• vertical specialism in an underserved niche;
• AI-native rebuild on platform-embedded S-Agentic features layered with disciplined workflow redesign.

It runs 90–105 revenue, 8–14% EBITDA, 60–80 headcount. Wulff-Yhtiöt's Tiliteema acquisition in 2025 (~€1M revenue, 9 staff, 230 customers) is the cleanest contemporary anchor for the structural-acquisition path absorbing the first kind¹². Aggregate T4 firm count shrinks; the residual is more profitable.

T5 toiminimi shrinks materially. Solo bookkeepers face client-acquisition-cost asymmetry against platforms that market self-serve directly. Holvi — Finnish FSA payment-institution licence, sold by BBVA in 2024/2025 to Keru Fintech Investments, banking-plus-accounting hybrid for SMEs and solopreneurs¹³ — is the consumer-facing route end-clients can take without a tilitoimisto at all. Retiring bookkeepers without successors compounds the run-off.

T6/T7 platforms originate the capability and distribute it asymmetrically. Procountor under KKR runs more than 30,000 Finnish company users and over 1,400 Nordic accounting firms; Procountor Tili (embedded business banking with Alisa Bank, April 2024) is the first visible direction-of-travel toward end-client-facing pricing¹⁴. Visma Solutions' Netvisor line held ~26% Finnish accounting-software market share in 2024; Visma Solutions group reported €159.7M revenue and 40.3% EBITDA margin in 2025¹⁵. Fennoa — €23.7M revenue FY 2025, +34% YoY, 950+ tilitoimisto customers, 85 employees — is the Finnish platform with a published commitment to sell only to accounting firms¹⁶. The platform tier divides on exactly this axis: operator-loyal platforms transmit the gain to the tilitoimisto; end-client-facing platforms hold the option to capture it themselves.

Per-tier composite trajectory under the Base scenario. T6/T7 figures are platform-firm unit economics, not service-line revenue.

Same productivity wave, four different outcomes — gated by who owns the agentic layer and who sells the end-client relationship.

The economics — why no price drop yet, and what changes that

The load-bearing question is not whether AI productivity is real. It is. The question is why it does not yet appear as price compression in the published hinnastot — and what specifically would force the change.

The answer separates three quantities the conversation usually collapses.

Codifiable work is compressing 70–80% in labour hours at firms running deployments at scale. Productivity is being captured at firm level, as margin and as growth. The visible-but-modest industry-aggregate wedge reflects two things: only the leading T1 firms are at peak compression yet, and the gain is absorbed into headcount reduction rather than output expansion. There is no contradiction; there is a tier-asymmetric deployment phase, with the rest of the industry still 2–4 years behind.

The umbrella holds for five reasons. S-Assistant is table-stakes, so no firm can attack on cost. End-client switching costs (data migration, KLT-relationship continuity, year-end timing) suppress price discovery on the demand side. Productivity is captured in headcount reduction rather than output expansion, so industry capacity does not over-shoot demand into a price war. Top-5 share is around 10%, too low for any single firm to break the equilibrium and capture compensating volume. No firm has the incentive to break ranks.

It does not hold forever. Two triggers operate at different timescales. The structural trigger is ViDA — adopted 11 March 2025, mandatory cross-border e-invoicing from 1 July 2030¹⁷. Finland already runs at roughly 90% structured e-invoicing adoption through Tieke, Maventa, Apix and Procountor operator networks plus the 2020 public-sector law¹⁸. When ALV reporting becomes near-real-time, the discrete monthly close that the ALV-period billing line is anchored to dissolves; pricing for that line resets on the other side of the regulation. The observational trigger is the first publicly visible SME-direct platform pricing event below median tilitoimisto fees — Procountor Tili-style banking-plus-accounting bundles, or a Visma direct route, or a new entrant. The date is unknown; the indicator is watchable from 2026.

flowchart LR
    A["Traditional
(pre-2022)
100% labour"] --> B["Partially automated
(S-Assistant, 2024–2026)
50–80% remaining
by work category"]
    B --> C["Mature S-Agentic
(2028–2030)
20–55% remaining
on codifiable core"]
    C --> D["Firm-level cost base
50–60% compression cap
(18–25% non-automatable floor)"]
    classDef regime fill:#F7F7F5,stroke:#0A7D70,color:#1F2D3D;
    classDef ceiling fill:#0C1B3D,stroke:#0C1B3D,color:#FFFFFF;
    class A,B,C regime;
    class D ceiling;
  

The cost-base math underneath sets the ceiling. Codifiable accounting work — tositekäsittely, ALV calculation and reporting, payroll, tilinpäätös first draft, reconciliation, anomaly detection, veroilmoitukset — compresses 70–80% in labour hours at mature S-Agentic. Firm-level cost compression caps at 50–60%, not 75%. The non-automatable floor sits at 18–25% of the current cost base and does not yield meaningfully:

• KLT continuing education and signature-grade responsibility on tilinpäätös;
• KILA interpretation;
• the customer relationship itself;
• audit-response preparation;
• money laundering act reporting obligation (AML);
• complex payroll judgement on terminations, parental leave and pensions;
• ongoing cyber-security upkeep;
• the management of the firm itself.

The two numbers must coexist. Quoting 70–80% as the firm-level reality turns a real productivity story into an unattainable margin promise.

A proprietary AI build at Talenom scale under 2022 reference classes would have anchored €8–20M over three years. The 2026 figure, with rented LLM infrastructure and 2–3× integration-engineering productivity, is €3–8M over 18–30 months. The proprietary IP that remains is the accounting domain layer and Finnish compliance, not the model.

The umbrella holds while productivity gain stays inside the firm. It breaks when a tier-specific first-mover or a regulatory dissolution moment forces the reset.

The regulatory layer

The default narrative on accounting AI in Europe is that regulation is the headwind. For the codifiable core, it is closer to the accelerator.

The EU AI Act entered into force 1 August 2024 and becomes fully applicable on 2 August 2026¹⁹. Annex III high-risk classifications bite narrowly on accounting. Pure transaction classification, VAT reporting and bookkeeping fall outside. Only AI influencing employment decisions (§4) or evaluating creditworthiness of natural persons (§5(b)) tips into high-risk. The boundary runs between payroll calculation (in scope of automation, out of scope of Annex III) and payroll-tied advisory affecting hiring or termination; between corporate-counterparty creditworthiness assessment and creditworthiness assessment of ammatinharjoittajat and yksityiset elinkeinonharjoittajat. The narrowness is the news.

GDPR Article 22 continues to constrain fully automated individual decisions with significant effect on natural persons. EDPB Opinion 26/2025 and the Digital Omnibus joint EDPB/EDPS opinion signal a shift toward a safeguards-based reading²⁰. The meaningful-human-review test sets a real design constraint (sampling rate, exception sensitivity, escalation paths must be documented) but is achievable under disciplined workflow design. Kirjanpitolaki 605/2024 lifted size-class thresholds roughly 25% for fiscal years starting 1 January 2024, moving firms out of stricter regimes²¹.

KILA has issued no AI-specific opinion through May 2026²². The silence carries two readings simultaneously. KILA's view is most likely that existing electronic-bookkeeping principles apply — no regulatory vacuum requires new guidance. The practical effect is nonetheless to leave S-Autonomous deployments in the codifiable core without explicit professional-standards cover. For tilitoimistot serving audit-obligated clients, the binding floor is therefore the auditor's ISA evidence requirement, not KILA. A separate Finnish-specific evidence standard for AI-prepared accounting work is likely to emerge in 2027–2029. ViDA is the load-bearing regulatory event of the horizon; mandatory cross-border e-invoicing arrives 1 July 2030¹⁷. The Finnish national transposition is not yet published; the structural date holds regardless.

Regulation in the codifiable core is mostly a calendar, not a constraint — and the calendar runs toward more structured data, not less.

The watch-list — what to monitor between now and 2028

Ten indicators, each tier-portable and falsifiable. V2 grades them.

Strategic options for owner-CEOs

Four postures are durable inside this horizon. Each is a real choice, not a presentation.

Platform-rider. Subscribe to a platform-tier vendor's agentic features (Procountor, Netvisor, Fennoa) and redesign internal workflow around them. The path most T4 firms will end up on whether they choose it or not. The productivity gain is shared with the vendor; the moat is whatever is left after the platform takes its share: relationship, niche expertise, regional density. Workable for firms that have one of those three.

AI-augmented operator. Build the workflow discipline that captures the productivity gain at firm level: disciplined sampling, exception architectures, internal LLMOps (operational management of large language models), customer-segment-specific configuration over platform-shipped features. Realistic for T2, T3 and the more ambitious T4 firms. The Greenstep cloud-native model is the live exemplar. Capital cost sits in the workflow redesign, not the model.

Vertically-integrated platform. Build or own the platform layer, not just the service layer. The path Talenom walked into Easor before unbundling; the path Aallon Group and Administer still occupy as live counter-examples. Capital-intensive (€3–8M over 18–30 months for proprietary build), risky if the software multiple does not survive the build, and increasingly the exception requiring explanation by 2031.

AI-vendor partnership. Pair early with one Finnish vertical-AI specialist on the Rantalainen × Taxxa.ai pattern. Contribute proprietary Finnish-tax-and-payroll labelled data; receive preferential capability, co-development access, or equity in exchange. Available to T2 firms with operational scale to be a meaningful data partner, to T3 firms with sharp vertical specialism, and selectively to the most disciplined T4 firms in narrow niches. The data-network-effect moat survives the entry-cost collapse. That is what makes partnership a strategically different posture from subscription.

The wrong posture for any tier is none of these. Continued vertical integration without articulating the durable-bundle thesis, transactional pricing on transactional work without a non-transactional layer, owner-led generalist work without specialism — these positions compress in any scenario.

The strategic question is which posture is congruent with the firm's existing client book, capital position, and the owner's appetite for either capability investment or partner dependence.

Appendix — a fifth option for owners weighing exit. The four postures are operator choices. A board chair facing fiduciary duty surfaces a fifth: owner-exit ahead of structural repricing. While the pricing umbrella still protects service-multiples — before either ViDA 2030 or the first SME-direct platform event resets them — the structural option to sell exists. The window is the umbrella's remaining life.

Open questions

Seven open questions matter most. We name them so the reader holds them consciously.

Pricing-umbrella first-mover identity and arrival year. The first observable break can come from one of three sources:

• a platform pricing SME-direct;
• a PE-backed T2 deploying AI productivity to attack share;
• or ViDA's structural 2030 reset.

We expect at least one to fire within the horizon; we name no ordering. Whether broad-domain S-Autonomous reaches production at any Finnish tier by 2031 — narrow-domain at >100-client scale is plausible by 2028; broad-domain is the optimistic-end 2030–2031 versus realistic-end beyond-2031 split V2 will resolve. KKR's exit timeline and exit-buyer for the Procountor portfolio is the single highest-information uncertainty determining 2028–2031 platform-tier shape — a trade sale collapses the multi-platform market; secondary PE preserves the coopetition equilibrium; an IPO repeats the Easor pattern.

T8 in-house substitution rate and direction. The scale of in-housing on the T8 side is decisive for T4 survival: whether mid-market and SME-band demand softens by 5% or 25% over five years. No Finnish CFO survey anchors the number. Whether KILA breaks its silence on AI-generated accounting work — a positive opinion accelerates S-Autonomous adoption among audit-obligated portfolios, a restrictive one delays it. Whether Aallon Group and Administer unbundle or hold integrated postures through 2031 — both are live counter-examples; a third unbundling event by end-2028 confirms the structural reading. Whether the 2025 −0.2% revenue decline was purely macro or partially early AI-driven price compression — settled by 2026 macro recovery and revenue trajectory, held as macro per Jari Seppä's attribution at publication time.

Calibration note. Historical analogues — ERP transitions from 1996, the ATK-aalto of 1995–2010 — suggest that timing claims in industry foresight tend to run 12–24 months optimistic. The dates in this piece may follow that pattern. The V2 update in late 2026 will grade against actual observation, which is what the Industry Horizon series is built for.