Editor’s note: I first published this piece on the Coveo blog in April 2018, when the pressing question for IT leaders was what to do about the sunset of the Google Search Appliance. The technology has changed. The habits have not. This update keeps the original structure, replaces the dated framing, adds one habit, and grounds the argument in research that did not exist in 2018.
In 2018 the pressure on CIOs was to build something with chatbots. In 2026 it is to build something with agents. The pitch has changed but the anxiety is the same: leaders know they need to innovate, they are not sure how to start, and they underestimate the degree to which search is the foundation for whatever they build next.
What has changed is the evidence. Eight years ago I argued from experience that innovation fails for organizational reasons rather than technical ones. Today that claim is measurable.
Gartner predicts that over 40 percent of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls. S&P Global Market Intelligence found that the share of companies abandoning most of their AI initiatives before production rose from 17 percent to 42 percent in a single year, with the average organization scrapping 46 percent of its proofs of concept. McKinsey’s State of AI 2025 reports that 88 percent of organizations now use AI in at least one function, yet only about 6 percent qualify as high performers attributing meaningful EBIT impact to it. A preliminary report from MIT’s Project NANDA, covered in Fortune, put the share of enterprise generative AI pilots with no measurable P&L impact at roughly 95 percent.
Read those together and a pattern emerges. None of them blames the models. All of them point at the organization.
So how do you build a foundation for innovation? You still need a playbook, and you still need to confront the myths that hold you back.
Habit 1: Bust the Myth of the Lone Genius. Build a Team.
In 2018 the myth was that innovation required one visionary to make it happen. The 2026 version is the myth of the lone model: pick the right foundation model, wire it up, and transformation follows.
The research says otherwise. Every failure cause named by Gartner for agentic projects is organizational. McKinsey’s high performers are distinguished by senior leadership ownership and by the willingness to redesign how work gets done, not by which vendor they chose.
Innovation is a verb. It is an active process, and it’s a team sport. Build a team that understands both the technology and the business outcomes, and make sure it includes the people who own the data, the people who own the workflow, and the people who will measure the result. That team translates what users say they want into what the outcome should be, rather than waiting on one person for the complete vision.
Habit 2: Use an Innovation Playbook.
Successful teams follow a simple formula. They define the outcomes they want. They plan effort around measurable results that prove the outcome. They standardize the process so it can be repeated.
Habits are simple, repeatable actions, and your playbook’s primary job is to outline the habits of your team: roles, communication structure, and scope. Every playbook will differ by organization and will change as you complete more projects. Rely on best practices first, then adjust.
The best practices I listed in 2018 hold up, with some updates:
- Don’t repeat yourself. Reduce user missteps and rework.
- Define the business outcome and what success looks like before deciding what to implement. “We need agents!” is not an outcome. McKinsey found high performers are nearly three times as likely as others to fundamentally redesign workflows when deploying AI, which is another way of saying they start from the work, not the tool.
- Don’t get lost in the hype cycle. Gartner estimates that of the thousands of vendors claiming agentic capabilities, only about 130 offer the real thing. Have a process for evaluating new opportunities before the demo arrives.
- Be open and honest about limitations and risks. Generative systems fail in new ways: confident wrong answers, gaps in coverage, stale grounding data. Someone on the team needs to be the “no” person.
- Create a culture where failure is acceptable and instrumented. Innovation requires failure, not perfection, but only if you can measure what failed.
Habit 3: Commit to Quick Wins.
Very few large “big bang” projects deliver on their promises. That was true of systems integration in 2018 and it is true of AI programs today. Gartner’s first-named cause of cancellation is escalating cost, and S&P Global’s data shows organizations discarding nearly half their proofs of concept. Both are what a big-bang roadmap looks like from the inside.
Change your thinking from big-bang projects to a longer roadmap with quick wins built in. Deliver small milestones focused on business impact and user experience. Provide value at every step. In search terms, that means connecting one more high-value source, fixing one ranking problem that support tickets keep surfacing, or shipping a grounded answer for one top intent. Walk before you run.
Habit 4: Test the Innovation Playbook with Search.
Search remains the best place to test your playbook, and the reason is stronger than it was in 2018. Every copilot, every retrieval-augmented assistant, and every agent is a search system underneath. The model can only reason over what retrieval hands it.
The most common problem I encounter in organizations is still an acute deficiency in information retrieval. The McKinsey Global Institute estimated that workers whose jobs are mostly communication and coordination spend nearly 20 percent of the workweek looking for internal information or tracking down colleagues who can help. And Gartner predicts that through 2026, organizations will abandon 60 percent of AI projects unsupported by AI-ready data, with 63 percent of organizations reporting they either lack or are unsure they have the right data management practices for AI. Fixing retrieval is the AI-readiness work.
The three tenets I recommended in 2018 have aged into requirements.
Search should be where users are. Making users go to a separate page to find an answer was a recipe for failure in 2018. Now the “user” is often an agent embedded in a workflow, and the same rule applies: retrieval has to be available where the work happens, as both active query and passive suggestion based on context.
Task for your playbook: Talk to users. Map where in their process better information would produce better outcomes. Put those in a backlog, roadmap the releases, and deliver the high-value ones first.
Search is about interfaces. Building search was never about fancy design. It was about data normalization and connecting systems. Grounding a language model is the same job with higher stakes, because the model will fill gaps with confident guesses.
Task for your playbook: Understand where the data users need actually lives. Work with the teams who own it to bring it into search, with the metadata and permissions that make it trustworthy.
Search is relevance. It’s personal. “I can’t find anything” still means “I can’t find what I need to finish my task.” General relevance matters less than relevance to a specific user, at a particular time, in a particular context. The difference now is that the interaction is conversational, and context accumulates across a session.
Task for your playbook: Model the workflows of high-performing users so the system can suggest the next best action given their context.
Habit 5: Instrument Before You Scale.
This habit was implicit in 2018. It needs to be explicit now.
“Unclear business value” appears in Gartner’s cancellation reasons because most teams cannot show value; they never established a baseline. Before scaling anything, know your current relevance metrics, build an evaluation set from real queries, and decide what you will measure after launch. Precision and rank get you started; ranked measures such as NDCG tell you whether the top results are actually the right ones. Monitoring is not a phase two item. It is what makes the quick wins in Habit 3 provable and the failures in Habit 2 survivable.
Getting Started
Deliver innovation in small, meaningful pieces. That approach facilitates adoption, reduces friction, and produces evidence. Get your organization into the business of shipping exceptional features on a foundation of good retrieval, and the larger innovation problem takes care of itself. It did in 2018. The research now says why.