How to Start Using Generative AI in Your Business

The fastest way to start using generative AI in your business is to pick one repetitive, text-heavy workflow, give the team that owns it a business-grade AI assistant, measure the result against a baseline, and only then decide whether to scale, build something custom, or stop. Most companies skip the measuring and redesigning part. That's why individual employees feel faster while the P&L doesn't move.

The tools have changed a lot since this guide was first published in 2023. GPT-4, the model in the original title, is gone from ChatGPT, and today's options include OpenAI's GPT-6, Anthropic's Claude 5.5 and Google's Gemini 3 families. Below you'll find a five-step way to start, real examples from retail, banking, healthcare and proptech, the risks to plan for, and how to decide between buying, building and bringing in a partner.

Why Do Most Companies Still Struggle to Get Value from Generative AI?

Most companies struggle because they hand out AI tools to individuals without redesigning the workflows around them. The time each person saves gets spread across small tasks, and nobody captures it as lower cost or higher output.

McKinsey's State of AI in 2026 survey shows the gap clearly: 80% of respondents say AI has improved their individual productivity, but only 37% can attribute any EBIT impact to it, about the same share as a year earlier. Only around 6% are "high performers" who attribute at least 5% of EBIT to AI.

The chain is simple: tools without workflow change lead to scattered time savings, which lead to pilots that can't prove their value and get cut at budget time. The fix isn't a better model. It's choosing a workflow where the savings show up in a number someone already tracks, such as cost per support ticket, time to close a loan file, or hours spent on documentation.

How Do You Start Using Generative AI in Your Business? 5 Steps

You start with a narrow, measurable pilot and expand only on evidence. This sequence works for a 20-person startup and for a department in a large enterprise.

1. Pick one workflow. Look for work that is frequent, text-heavy and easy to check: answering support tickets, drafting proposals, summarizing contracts, writing meeting notes. Avoid starting with anything customer-facing and high-risk.

2. Choose a business-grade tool and set a usage policy. Business and enterprise tiers of the major assistants exclude your data from model training by default and give you admin controls. Personal accounts don't, so tell employees which tools they may use and what data they may never paste in.

3. Measure the baseline. Write down how long the task takes and how often it needs rework today. Without this number, the pilot ends in opinions.

4. Pilot with a small team and a human in the loop. Four to eight weeks is usually enough to see whether quality holds. Someone accountable reviews every output that leaves the building.

5. Decide whether to scale, adjust, build or stop. If the off-the-shelf tool works, roll it out. If it fails because it can't reach your data or systems, that's the point to consider a custom build.

Which Generative AI Models Can Businesses Use in 2026?

Businesses can choose from three leading closed-model families and several open-weight ones. The specific versions change every few months, so pick a provider based on security terms, pricing and where your data lives, not on last month's benchmark.

Models also change quickly. OpenAI launched GPT-6 Sol and Luna in September 2026 at half the API price of the previous generation, and it retired GPT-4o and several older models from ChatGPT in February 2026. If you build on an API, keep prompts, evaluations and integrations independent of a single model so you can switch when a cheaper or better one ships.

Open-weight models make sense when data can't leave your infrastructure or when volume makes per-token pricing expensive. They also make you responsible for hosting, updates and security, which is a real cost for a small team.

How Are Companies in Retail, Banking, Healthcare and PropTech Using Generative AI?

The strongest results so far come from narrow workflows with clear owners: content at scale, knowledge retrieval, documentation and new distribution channels. One example per industry shows what each one required.

How Is Generative AI Used in Retail?

Retailers use generative AI to produce product content at a scale that was never affordable by hand. CarMax used Azure OpenAI Service to turn about 100,000 customer reviews into roughly 5,000 review summaries, one for each make, model and year it sells. By Microsoft's account, the company estimated that the manual process would have taken about 11 years, and the AI-assisted one took a few months.

A CarMax staff member reviews each summary before it goes live, so the model drafts and people stay accountable for brand voice and accuracy.

How Is Generative AI Used in Banking and Finance?

In banking, the most proven use case is internal knowledge retrieval, not customer-facing chatbots. Morgan Stanley gave its financial advisors an assistant that answers questions from roughly 100,000 internal research reports and documents. By June 2024, 98% of advisor teams had adopted it, and the bank followed up with a tool that drafts client meeting notes.

The architecture behind this is retrieval-augmented generation (RAG): the model answers from your approved documents instead of from its general training. It requires clean, permissioned content and a provider contract that keeps proprietary data from being retained.

In finance, compliance sets the limits. Under the EU AI Act, AI used for credit scoring counts as high-risk, and after the 2026 Digital Omnibus amendments those obligations apply from December 2, 2027. Plan for documentation and human oversight from day one.

How Is Generative AI Used in Healthcare?

In healthcare, the clearest example so far is ambient clinical documentation: an AI scribe listens to the visit, with the patient's consent, and drafts the clinical note for the physician to review. At The Permanente Medical Group, about 7,260 physicians used AI scribes across some 2.5 million patient encounters, saving an estimated 15,791 hours of documentation time in just over a year.

What made it acceptable to clinicians and patients was the review step and the consent process, not the model alone. Any similar product has to be designed around HIPAA, GDPR or local health-data rules from the start.

How Is Generative AI Used in PropTech?

In proptech, generative AI is changing where buyers search. In October 2025, Zillow launched an app inside ChatGPT that lets users search listings, view photos and maps, and connect with agents from a conversation.

It also shows the trade-off. Zillow stressed that it keeps control of the experience to protect listing data and stay compliant with MLS display rules, and some in the industry questioned whether that's enough. If your product depends on licensed data, check those agreements before you put it inside a third-party AI assistant.

Which Everyday Tasks Can Employees Hand Over to Generative AI Today?

Employees can hand over first drafts of most routine reading and writing work: summaries, notes, descriptions and research tables. These tasks need no custom development, only a business-grade assistant and a habit of checking the output.

Every major assistant now reads images, which makes image descriptions and alt text an easy first task. Marketers, SEO specialists and accessibility specialists can upload a batch of images, get draft descriptions in seconds, and spend their time editing instead of writing from scratch.

Long reports, contracts and PDFs, including their charts and tables, can now be summarized in one go. The useful habit is to ask specific questions ("which clauses differ from our standard terms?") rather than asking for a generic summary, and to verify any figure before you reuse it.

Meeting notes are now built into Zoom, Microsoft Teams and Google Meet, and messier inputs work too.

AI features in spreadsheets and the "deep research" modes of the major assistants can also fill research tables: a column of company descriptions, industries or other public information for a list of names.

Treat the output as a researcher's first pass. Models still state wrong facts confidently, so spot-check before anything goes to a client.

What Are the Risks of Using Generative AI in Business?

The main risks are wrong answers, data leaks, compliance gaps and costs that grow faster than value. None of them is a reason to wait, but each needs an owner before you scale.

• Accuracy: models can produce plausible but false statements. Anything that reaches a customer, regulator or contract needs human review, and high-stakes workflows need automated evaluations, not spot checks.

• Data exposure: the bigger risk is often "shadow AI," meaning employees pasting client data into personal accounts because no approved tool exists. Giving people a sanctioned tool is the most effective control.

• Compliance: GDPR, HIPAA and sector rules already apply to AI. The EU AI Act adds obligations for high-risk uses such as recruitment and credit scoring, with deadlines now set for December 2027 and August 2028.

• Cost: seat licenses are predictable; API usage isn't. About 20% of McKinsey's 2026 respondents said AI operating costs limit their use. Set budgets and monitor token spend per workflow.

• Generic output: off-the-shelf assistants don't know your products, customers or processes. That limit is usually what pushes companies toward custom builds.

Should You Buy, Build or Partner for Generative AI?

Buy first, build when you hit a wall, and bring in a partner when the build is beyond your team's experience. An MIT NANDA study reported by Fortune found that 95% of companies saw no measurable P&L return from generative AI pilots, and that tools bought from specialized vendors or built with partners succeeded roughly three times as often as purely internal builds.

Building no longer means training your own model. Model APIs from OpenAI, Anthropic and Google are self-serve and billed per token, so a custom solution today usually means connecting an existing model to your data and tools, then testing it properly. The hard parts are data quality, permissions, evaluation and integration with legacy systems.

What Does It Take to Make Generative AI Pay Off Long Term?

Generative AI pays off when it becomes part of how work is done: specific workflows, clear owners, measured outcomes and an architecture that survives the next model release. Companies that treat it as a tool rollout get faster individuals and flat results.

If you've proven the value with off-the-shelf tools and need AI that works with your own data and systems, Monterail can help. We design and build AI-powered products and integrations, including in regulated industries such as fintech and healthtech, and we'll tell you honestly when an off-the-shelf tool is the better choice. Get in touch with our team to talk through your first use case.