Common Pitfalls in AI Adoption: Missing the 'Three Questions'
AI adoption has entered a phase where the competition is no longer about 'PoC' but about 'habituation in the workplace'.
At the HubSpot event 'GROW EUROPE 2025', OpenAI GTM Lead Orian English stated that European companies' use of AI has shifted over the past year from the 'testing' phase to the 'embedding in the organization' phase. The background to this is the explosive spread of individual use, followed by the acceleration of corporate adoption. The focus of practical work is converging on what to solve, who will drive it, and how to measure it, rather than searching for flashy use cases.
1. AI adoption has 'accelerated'—the reality shown by the numbers
Orian noted that 'ChatGPT weekly active users have quadrupled in a year,' suggesting that adoption at the consumer level has changed the 'prerequisites.' In fact, OpenAI CEO Sam Altman has stated that ChatGPT has reached 800 million weekly active users.
The momentum is also strong on the corporate side, with OpenAI announcing that 'ChatGPT Enterprise seat counts are up about 9x year-over-year.'
In other words, AI is becoming an infrastructure for company-wide productivity rather than a 'toy' for a few advanced departments—this is the biggest tectonic shift.
2. The 'three questions' you must ask before adoption
2-1. What problem are you solving (a vitamin or a painkiller?)
Define 'what you want to achieve' first—Orian emphasizes that whether it is revenue (top line) or efficiency (bottom line), if you leave the goal vague, the ROI will become a matter of 'vibe'.
The point is to break it down not because 'it looks convenient,' but 'to move this specific metric'.
2-2. Who is the executive sponsor?
'The absence of a sponsor skyrockets the probability of failure.' With only the enthusiasm of the front lines, authority, budget, and rule-setting will not keep up, leading to a stall. The ideal is for the C-suite, or at least a senior leader with strong authority, to lead with the attitude of 'I use it myself'.
2-3. What is the definition of 'Good' and how is it measured?
'What is success, and how do you measure it?' It can be quantitative (man-hour reduction, revenue contribution) or qualitative (satisfaction, creativity, NPS). However, if you don't set KPIs at the beginning, you won't be able to explain it later.
As Orian says, while AI has high perceived value, it tends to get caught up in 'like or dislike' arguments if there is no measurement design.
3. The misunderstanding of 'AI Ready'—don't wait for the perfect foundation
Orian says that narrowly defining AI readiness as 'a state where a massive data foundation has been built' is a brake in itself. The following remark is impressive:
'Good today is better than perfect tomorrow.'
3-1. Start with low risk and high impact
For example, start with areas that have no customer touchpoints and relatively low confidentiality, such as 'internal FAQs' or 'internal document search.' Orian introduced an example where they built a system within OpenAI that allows access to information scattered across Notion/Google Docs via chat, and usage spread more than expected.
What is important here is that the first winning move is not a 'grand AI project,' but the reduction of friction in daily tasks like 'searching, summarizing, and answering'.
4. Strong issues in Europe: sovereignty, privacy, and control
In Europe, conversations like 'Is it safe? Who manages the data?' tend to linger for a long time—this point is realistic. The countermeasure is not a matter of spirit, but design.
4-1. Assume that 'corporate data will not be used for training'
OpenAI explicitly states that organizational data from Enterprise/Business/Edu plans is not used for training by default.
(Conversely, you should definitely inventory what happens with which plan and settings before adoption.)
4-2. Entering an era where you can choose the 'storage location' with data residency
OpenAI provides data residency (in-region storage) for Europe, making it easier to comply with sovereignty requirements.
Furthermore, they have announced the expansion of data residency to the UK as well.
4-3. 'Inheritance of permissions' is the key to using connectors
Orian touches on connections (connectors) with tools like HubSpot and Slack, stating that 'access rights on the existing tool side are also reflected on the ChatGPT side.' Official documentation for HubSpot also explains that it respects HubSpot user permissions.
In other words, AI governance is not just an 'AI issue,' but an extension of ID management, permission design, and logging.
5. The turning point is 'AI literacy investment'—everyone speaks the language
Finally, Orian says that the key to the ideal state in 2026 (a world where company-wide adoption has progressed and results are being achieved) is 'AI literacy'.
'Leaders invest, employees speak the 'language of AI,' and build muscle as a habit.' Ultimately, what determines success or failure is not the selection of tools, but whether the way they are used in the field is shared as a 'standard model'.
Conclusion: The shortest path is to 'start small, measure, and expand'
Rather than flashy automation, design in this order: 1) problems to solve, 2) sponsors, 3) KPIs, 4) low-risk starting points, 5) permissions and data sovereignty—and create a 'start using it' environment. AI adoption is not a technology project, but an operation of organizational transformation.