Scalable Weekly · October 26–31, 2026

This week’s connecting thread is economic: AI is changing the cost of growth, the cost of mistakes and the cost of winning a customer. Those are the three questions through which we will read the news, rather than ranking funding rounds.

Guidewire wants to acquire Hyperexponential, which supports insurance pricing and underwriting decisions. Ravical aims to help accounting firms grow without proportional increases in headcount. Circuit Mind targets errors and revisions in electronics design. Pipedrive illustrates the other side: AI also changes how customers discover and choose software.

Our thesis: AI’s value lies in changing the economics of an activity, not simply adding an intelligent feature. A promise is not an outcome, however. Throughout this issue, we separate adoption, claimed benefits and returns still to be demonstrated.

The guiding question: what actually changes — revenue, cost, risk or capacity — and what evidence supports it?

IN ONE SENTENCE

AI creates value by changing customers’ economics; it can destroy value by changing the cost of reaching them.

What matters this week

Hyperexponential — The value of decisions

Guidewire has signed an agreement to acquire Hyperexponential for $1.7 billion in cash, subject to adjustments, plus up to $400 million tied to future recurring revenue targets. As of October 31, closing remains pending.

The London company brings models, data and rules into insurance risk assessment and pricing. Our interpretation: Guidewire wants a stronger position where software helps determine which risks an insurer accepts and at what price.

The economic lever is decision quality, alongside speed. The agreed price signals strategic interest; it does not, by itself, demonstrate better insurance outcomes.

Ravical — The cost of growth

Belgium’s Ravical announced a $17 million seed round, led by MMC Ventures, and reports more than 100 accounting firms using its platform across Europe. It aims to turn leading professionals’ expertise and procedures into a shared capability across the firm.

The economic lever is the relationship between clients served and staff required. The test is not how many answers an assistant generates, but how much additional work the firm delivers without increasing review, rework or risk.

Founders should measure the fully loaded cost per completed engagement: preparation time saved is not yet margin earned.

Circuit Mind — The cost of mistakes

Circuit Mind announced $25 million in financing, comprising an $18.5 million Series A and a previously unannounced seed round. Its platform combines AI and physics-based models to design and verify electronic circuits.

The economic lever is the cost of discovering errors late. A quickly generated schematic offers little benefit if it requires another prototype; useful verification can avoid wasted components and delivery delays.

The company reports substantial time savings, which we do not treat as universal results. The commercial question is concrete: how many projects reach production with fewer revisions, without compromising reliability?

Pipedrive — The cost of finding customers

Pipedrive announced a major restructuring. In the CEO’s letter reported by ERR, free web traffic has fallen and paid traffic has become more expensive, while small businesses increasingly ask AI for software recommendations.

The economic lever is customer acquisition cost. Management says this is not simply about replacing people with AI: customer expectations and product discovery are changing.

This is the counterweight to the other stories. Even a company that develops products faster can become less commercially efficient. Measuring internal productivity alone misses half the problem: the cost of turning a product into sales.

Under the Hood

Hyperexponential — A model matters when it reaches the decision

An insurer needs more than a faster way to read a submission. It must decide whether to accept the risk, on what terms and with what effect on its portfolio. A quicker response can help win a contract; the wrong price can damage its economics.

That tension makes Hyperexponential this week’s central case. Work efficiency and the economic quality of an outcome are different things. The platform seeks to connect them. Our analysis concerns the value of that connection, not a claim that AI has already solved insurance pricing.

Product: from model to quote

The pricing platform connects actuarial models with underwriters’ work, including versioning, controls and traceability. The operational challenge is making risk logic usable when an actual submission arrives.

Consider a simplified example: a submission contains incomplete information about a business. The team must enrich it, apply the appropriate model, check conditions and determine whether approval is required. Automating initial document reading addresses only one stage. A decision requires consistency between data, calculation and accountability.

For the customer, the useful comparison is therefore not “AI versus spreadsheet.” It is the cost of taking a model change all the way to a reliable quote without multiplying reconciliations and manual checks. That is a more demanding product requirement than generating a plausible answer.

Adoption: concrete but bounded evidence

In December 2025, Aviva reported 23 models live on hx and an expanded partnership following early testing. In June 2026, Allianz Commercial reported 13 pricing tools delivered in 13 weeks. These are corporate announcements naming identifiable customers, not fresh results from this week.

They demonstrate implementation and the ability to bring tools into daily work. They do not isolate an improvement in the ratio of claims to premiums, or loss ratio. That outcome also depends on risk selection, market conditions and claims development.

A buyer should separate two tests: does the process work better? And, across comparable periods and portfolios, does it produce economically better decisions? Evidence for the first does not substitute for evidence for the second.

Reaching the income statement requires an explicit bridge: freed capacity put to use, additional profitable contracts or lower operating costs, net of software, integration and review. A faster quote that changes none of those measures remains an operational improvement, not yet a demonstrated economic return.

AI: more prepared work, not infallible judgment

With hyperoperator, introduced in July, hx proposes an agent that prepares submissions within the insurer’s pricing, risk-appetite and authority controls. The announcement envisaged a controlled customer rollout; it does not establish that broad autonomous operation is already deployed.

The economic opportunity is to handle more submissions without proportional growth in administrative work. But a system that prepares more files can also move the bottleneck to review. Faster intake is useful only if downstream teams can absorb the additional workload without weakening decision quality.

The decisive measures would include completed submissions, exceptions, review time and correction costs. Counting automated actions alone does not establish that a team has gained useful capacity or that its total operating cost has fallen.

Defensibility: integration and context

Our hypothesis is that defensibility extends beyond the AI model. A customer with connected models, rules and processes must consider the cost of rebuilding and validating that combination before switching suppliers.

That can support commercial continuity, but creates an obligation. Excessively customized implementations would make growth resemble a services business. Evaluating the economics requires project duration, support costs, configuration reuse and margins. The sources reviewed do not allow us to reconstruct those measures.

The data advantage also needs boundaries. Integration into workflows does not demonstrate a right to reuse information across insurers. Access to customer context can be valuable without creating a freely transferable data asset. A credible assessment should distinguish access, permission and actual learning benefits rather than treating them as equivalent.

Distribution: why Guidewire matters

hx was already a Guidewire technology partner. Guidewire reports serving more than 570 insurers in 44 countries. The combination could offer the specialist product broader distribution and reduce some commercial friction.

That does not mean automatic conversion of the installed base. Budgets, integrations and operating priorities still have to be won. Execution includes making adoption repeatable without disrupting the customer’s ongoing work or turning every deployment into a bespoke transformation project.

For European founders, the distinction matters: owning specialist expertise and owning the channel that sells it are different advantages. An industrial agreement can bring them closer together, but does not eliminate the work required to turn them into revenue. Potential reach is not booked business.

Price: existing value and growth still to deliver

The SEC filing separates base consideration from the component contingent on recurring revenue after closing. Part of the purchase consideration therefore remains tied to future execution, rather than being an unconditional payment for the business as it stands.

The more than $75 billion in annual premiums cited in Guidewire’s announcement concerns underwriting decisions supported by the platform. It is not hx revenue, nor premiums priced entirely by AI. Without adequate revenue data, we do not infer a reliable acquisition multiple.

The conclusion returns to our opening thesis: potential value comes from connecting expertise to an economic decision. Durable value requires useful capacity, sustainable costs and customers who renew. The announced acquisition recognizes that opportunity; it does not certify that every outcome has already been achieved.

Key Lessons

• Sell a measurable economic change. Define cost per completed engagement or project, including human review, corrections and implementation. Time saved at the first stage is not yet margin.

• Measure the customer and your commercial channel. More capacity for users does not guarantee efficient growth for the vendor: track acquisition costs and conversion alongside productivity.

• Separate three kinds of evidence. Production usage, operational improvement and economic return are different milestones. Establish each before using them to justify pricing and expansion.

WHAT TO LISTEN / READ

• EP73 — Hyperexponential: why would Guidewire pay up to $2.1 billion for the brain of underwriting? — The central case: models, adoption and the limits of available evidence. Audio in Italian.

• Daily News — October 28: Hyperexponential, Circuit Mind, Easol — Insurance decisions, electronics design and experience software. Audio in Italian.

• Daily News — October 29: Braid, Ravical, MuRay — Professional expertise and growth capacity. Audio in Italian.

• Daily News — October 26: Miro, Nous, adsilico — Software consolidation, household agents and medical simulation. Audio in Italian.

• Daily News — October 27: Machenta, Atlas Bio, Agora Labs — Robot learning, virtual trials and clinical data. Audio in Italian.

• Daily News — October 30: Fractile, Argotec, Nscale — The week’s other frontiers: chips, satellites and AI infrastructure. Audio in Italian.