AI Strategy Follows the Constraints of the Business

AI Strategy Follows Business Constraints

AI strategy begins with the business that must adopt it. Market position, customer promises, installed systems, data condition, rights, regulation, distribution, capital, workflow ownership, review capacity, and failure consequence determine which model capabilities can become useful work.

The same model can support an existing process in one company and enable a new product in another. The difference is not necessarily ambition. It is the constraint system.

flowchart TD
    A["Model capability"] --> B["Business position and customer job"]
    B --> C["Installed systems, data, and rights"]
    C --> D["Regulation, controls, and reviewer authority"]
    D --> E["Workflow and economics"]
    E --> F["Bounded AI strategy"]

Capability is the outside boundary

A model announcement can show that a task is newly possible. It cannot determine whether an organization should perform that task, with which data, for which customer, under which authority, or at what cost.

E004 uses Gemini 1.5 and Sora as dated examples. Long context expanded the amount of material that could enter one operation. Generated video expanded the space for visual iteration.

Neither announcement supplied a company-specific adoption plan.

Market position changes the starting point

An established company often has customers, distribution, contracts, licenses, capital, data, operations, and systems that already produce value.

Changing one process can affect several obligations. A customer-facing AI feature may depend on policy administration, billing, claims, identity, consent, records, complaints, accessibility, security, and service recovery.

A new entrant may begin with a narrower problem and fewer internal dependencies. It can design the interface, data capture, and workflow together.

The entrant also lacks many incumbent assets. It may need distribution, regulated authority, capacity, history, operational expertise, trust, and economics that survive real claims or service volume.

Insurance provides a useful comparison

Christopher Holland and Anil Kavuri's 2023 HICSS paper compares incumbent insurers and new entrants through product, process, and value-chain innovation.

In the cases studied, incumbents generally applied AI to defend and enhance existing positions. Entrants used it to build new products and customer experiences.

The paper does not say that every incumbent is cautious, every entrant is innovative, or one group will outperform the other.

It provides a map for asking where innovation is occurring and what the business is trying to protect or redesign.

Installed systems turn simple interfaces into structural work

The recovered E004 transcript describes a first-person experience with several acquired policy systems that stored data differently.

A dashboard request could require mapping fields, reconciling definitions, resolving history, and deciding which record was authoritative. A model could help with parts of that work. It could not remove the source-system and governance problem.

This pattern appears beyond insurance. A company with several customer records, product catalogs, contracts, or identity systems must decide which state the AI workflow is allowed to use and change.

The interface may be new. The obligations underneath it are not.

Data value and data readiness are different

An incumbent can possess decades of information and still struggle to use it for a new task.

Data may be distributed across formats, systems, acquisitions, jurisdictions, retention rules, contracts, and quality levels. The organization must establish authority, rights, lineage, access, meaning, freshness, and permitted use.

An entrant can capture cleaner data for a narrower workflow. It may lack longitudinal evidence, representativeness, claims development, rare events, and the operational history needed to evaluate the model.

More data and cleaner data are not the same advantage.

Regulation shapes architecture and evidence

NAIC's current artificial-intelligence topic record describes insurer uses across multiple functions and continuing work on governance, potentially high-risk models, third-party data and models, and examination evidence.

The NAIC Model Bulletin reminds insurers that consumer-impacting actions supported by AI remain subject to applicable insurance laws. It sets expectations for a written AI-systems program and risk-based governance where a jurisdiction adopts the bulletin.

The bulletin is not itself a model law or regulation. Applicable requirements depend on the state, entity, line, data, decision, and use case.

This still demonstrates why governance cannot be added after the interface launches.

Economics choose what survives

An AI use case can work technically and fail economically.

The business must account for data preparation, integration, model and infrastructure cost, review, exceptions, security, compliance, support, monitoring, correction, vendor dependency, and exit.

An incumbent may accept a modest process improvement across a large installed operation. An entrant may need the feature to create enough differentiation or efficiency to support acquisition and risk economics.

The strategy should state which economic result the workflow is expected to change and how the organization will know.

Start with the constraint map

Define the customer job and business position. Map the authoritative systems, data rights, operating obligations, reviewer capacity, failure consequence, integration path, economics, and exit.

Then select a bounded task.

NIST's Generative AI Profile is a cross-sectoral companion to the AI Risk Management Framework for incorporating trustworthiness considerations into design, development, use, and evaluation. It does not supply one strategy, but it reinforces the need to manage the system across its lifecycle.

The model belongs inside that strategy, not above it.

Strategy can change as the business changes

Incumbent and entrant are not permanent technical identities.

An incumbent can create a separately governed product, acquire a new system, partner, or rebuild a bounded workflow. An entrant can accumulate acquisitions, exceptions, old models, contracts, and customer promises that make later change difficult.

Review the constraint map when the business, product, systems, regulation, data, model, or economics changes.

Read [[What Gemini 1.5 and Sora Revealed About AI Adoption]] for the dated origin, [[Compare Incumbent Insurer and Insurtech AI Strategy]] for the insurance comparison, and [[Start AI Adoption With Bounded Tasks and Review]] for the first-use method.

This essay was developed with AI assistance from the recovered E004 transcript and the linked HICSS, NAIC, and NIST records. It does not classify or recommend a named company, insurer, insurtech, model, vendor, or product. Enterprise, insurance, governance, legal, economic, editorial, accessibility, and founder review remain required. Publication is unauthorized.