By Gagan Sethi, CEO, Ebix Technologies
For the last two years, almost every technology company has been trying to answer the same question: How do we become an AI company?
We all have done our part and this is what I have learned – Giving employees access to AI tools does not make a company AI-first. Adding a chatbot to an existing product does not make a company AI-first either. Even for the advanced ones deploying dozens of AI agents across engineering, customer service, finance and operations can still leave the underlying company largely unchanged.
An AI-first company is one that begins to change how work itself gets done.
McKinsey’s 2026 research found that companies generating the greatest value from AI were far more likely to have fundamentally redesigned workflows rather than simply adding AI to existing processes. Nearly three-quarters of its AI high performers reported redesigning workflows because of AI, compared with only about a quarter of other organizations.
Build an AI culture
I increasingly think of an AI-first technology company as a laboratory. A good laboratory does not begin with a predetermined answer. It creates an environment where ideas can be tested quickly, measured and either discarded or developed further.
The same model should work inside a technology company.
A business leader should be able to identify a problem on Monday, work with a small technology team to turn the idea into a working experience quickly, put it in front of users and decide whether it deserves further investment. The distance between an idea and something people can actually use should become dramatically shorter.
This is very different from the traditional enterprise technology process, where an idea becomes a requirements document, moves through several committees, enters a development queue and may not become visible software for months.
AI changes that equation where Natural language can increasingly become the starting point for requirements. Working interfaces can be created before full specifications are written. Code can be generated and reviewed faster. Test cases can be created directly from business requirements. Documentation can be produced alongside development rather than after it.
The goal is not to eliminate engineering discipline but to rethink SDLC.
BCG’s research on technology organizations shows how quickly this is changing. Two-thirds of surveyed companies are already using AI within software development, with organizations reporting improvements across code generation, testing, review and release automation. Legacy modernization is also emerging as a major AI use case.
That is why I believe the software development lifecycle itself should be one of the first enterprise processes to be substantially redesigned around AI.
Start with workflows
One mistake I see repeatedly is beginning the AI discussion with technology.
Which model should we use? Should we build an agent? Should we use one vendor or another?
Those questions matter later. The first question should be much simpler: Where does work get stuck?
Look for processes where people repeatedly read documents, extract information, compare rules, move data between systems, prepare the same analysis, wait for approvals or reconstruct information that already exists somewhere else in the organization.
Those are usually better starting points than searching for an abstract “AI use case.”
For instance, in one of our domains of insurance, consider the path from a submission arriving at an underwriter to a decision being made. Information may need to be extracted from multiple documents, checked against underwriting rules, compared with historical information, summarized and routed for additional review.
AI does not need to replace the underwriter to dramatically change that workflow.
It can perform much of the assembly around the decision while leaving judgment and accountability with the professional making it.
The same applies to claims, lending, healthcare administration, compliance and many other document-heavy industries. Deloitte’s 2026 insurance outlook points to underwriting assistance, fraud detection and all types of claims handling as areas where insurers are increasingly moving toward practical deployment.
This is where I believe the largest enterprise gains will come from. Do not try to make individual tasks slightly faster, but try removing unnecessary steps altogether between them.
Use AI to understand the technology you already have
There is another part of the AI opportunity that receives much less attention. Large enterprises do not operate on clean technology stacks of greenfield projects. They operate on applications developed over decades.
Some contain millions of lines of code. Yet these systems still process transactions that cannot simply stop while a company rebuilds everything from scratch.
Until recently, modernizing those platforms was painfully expensive because teams first had to understand them.
AI is starting to change that. It can help engineers analyze old applications, document undocumented code, trace dependencies, identify obsolete components, generate migration plans and assist with moving business logic onto modern technology.
IBM has described technical debt as one of the most complex problems facing CIOs and CTOs and sees AI-assisted modernization as an increasingly important way of addressing accumulated application complexity.
This may ultimately become one of AI’s most valuable enterprise applications.
The glamorous story always is creating something entirely new, but for an AI first organization, the important story is making its twenty-year-old systems understandable enough to upgrade safely.
Domain knowledge becomes more important
As AI models become broadly available, access to a model will become less of a competitive advantage.
Domain knowledge will become more important.
Again, consider insurance. A general model can understand what an insurance policy is. That does not mean it understands how a particular carrier evaluates a risk, how a broker compares products, which information matters during review, how regulations differ or what exceptions have accumulated through years of operating experience.
The same applies to healthcare, banking, wealth management and virtually every regulated industry.
The real value sits in the combination of technology and institutional knowledge.
That includes structured data, historical decisions, workflows, policies, rules, documents and, importantly, the domain knowledge held by experienced employees.
An AI-first company therefore needs to think about its data and integrations very differently. The objective is no longer simply to store information. It is to make the company’s knowledge understandable and usable across AI-enabled workflows.
McKinsey has identified data readiness as one of the principal constraints companies encounter when attempting to move AI from pilots into enterprise use.
This is why I would argue that an AI strategy for any organization without a data strategy and domain knowledge will eventually stall.
Human review is key for AI automation
There is also a tendency to assume that the ultimate measure of AI maturity is how many humans can be removed from a process.
I disagree. In many industries, particularly regulated ones, human judgment is exactly what allows more of the surrounding workflow to be automated safely.
The objective should be to determine where human judgment creates value and where human effort is merely being consumed.
For instance, continuing with my insurnace example, an underwriter making a risk decision creates value. An underwriter spending twenty minutes copying information from five documents into another system does not.
A financial advisor discussing a family’s priorities creates value. An advisor spending an hour assembling information already stored across several systems does not.
The design principle should therefore be straightforward: automate the assembly and preserve the judgment.
Governance needs to follow the same principle. Identity, permissions, traceability, human confirmation, monitoring and auditability should be designed into the system itself.
In my view, companies that treat governance as part of architecture will ultimately move faster in the AI world than companies that treat it as a compliance paperwork.
Measure business results
Finally, CEOs need to change how AI progress is measured. I am already becoming less interested in how many AI tools my organization has deployed, how many prompts employees run or how many models have been tested.
I want to know what changed.
Did a product move from concept to market faster?
Did software release cycles shorten?
Did processing times decline?
Did fewer documents require manual processing?
Did the cost of maintaining an old application fall?
Did customers receive a better experience?
Did revenue improve?
If the answer to those questions is still no, your organization has implemented AI without becoming AI-first.
The companies that ultimately benefit most from AI will not necessarily be those that chose the best models and bought licenses for their employees first. Models will change. Vendors will change. The technology available eighteen months from now will almost certainly be different from what companies are buying today.
The advantage belongs to companies that learn how to experiment quickly , make their knowledge usable, modernize what they already own, put governance into the architecture and measure everything against business outcomes.
That is what building an AI-first company really means. In summary – AI is not just a technology project even for a technology company. It is a different thought process and a different way of running the company.