Public developments and original analysis · Announcement date: · Sources
Menlo Ventures' 2026 enterprise AI report, released on November 3, contains a revealing gap between planned and reported spending. Survey respondents' average AI spending reached 4.7 times the previous year's level, against a planned 1.9 times. The median was 2.2 times, compared with a planned 1.8 times. Spending exceeded expectations, but the difference between average and median is a warning against letting one headline number determine every company's next budget.
Morning Consult surveyed nearly 400 U.S. enterprise decision-makers in September for the report. Menlo is also an investor in AI companies, and the research has a particular sample and perspective. Its findings can inform questions. They cannot establish a purchasing benchmark for a Chinese industry or show what return an individual company should expect from adoption.
The useful question is what the additional expense purchased: more work, better results, or experiments whose purpose remains unsettled. All three can coexist, and each calls for a different budget decision. Cutting every growing bill can remove valuable work. Increasing every allocation with growing usage can keep funding output that nobody needs.
How estimates miss the volume of work
A software estimate based on employee seats has a relatively understandable starting point: the number of accounts multiplied by the price per account. Usage-based AI services introduce other variables, including task frequency, material length, tool calls and revisions. The same people can expand the scope of their work without adding accounts. Fixed-price products and metered services need separate estimates rather than a single purchasing assumption.
Take a sales team preparing account briefs. In this hypothetical budgeting example, the initial request covers the customer's website and existing contact records. It later expands to competitor comparisons, tenders, leadership changes and several language editions. Each addition sounds useful, but together they define a different research assignment. A bill exceeding the original estimate may reflect an unmade decision about research depth rather than a rise in model prices.
That distinction should shape the agreement with a provider. Define the questions each standard brief must answer, its permitted sources and a stopping point. Sales can request deeper work according to the importance of an opportunity. A routine account may need enough background for the next conversation; a major opportunity may justify substantial research. An open-ended assignment conceals the difference in effort and makes turnaround times harder to explain.
Costs may also reflect work being repeated: missing information leads to more searching, an interrupted task begins again, or a generated brief goes unused. These are reasons to record what happened, not to invent a general waste percentage from a bill. A manager needs records tied to particular tasks and feedback from the people expected to use the results.
Deciding which work merits more funding
Begin the next allocation discussion with deliverables that have reached their intended users. For an account brief, did sales use the suggested questions? Did the research identify missing information? Which sections needed rewriting, and which were accurate but irrelevant to the upcoming meeting? Comparing those observations with the work invested gives a basis for expanding a task category or reducing a research step.
Time estimates need care. Saving an employee an hour of research does not automatically reduce payroll. Using the time for more customer conversations may be valuable, but that change must actually happen. Revenue attribution is harder still: product, price and the wider sales process also influence a sale. A budget discussion can recognize improvements in time and quality while keeping these different outcomes distinct.
Mokyun's digital employee proposals should meet the same standard. Its published positioning around customer information preparation and knowledge retrieval can support an agreement to deliver a sourced account brief, with sales identifying which questions helped the next meeting. Actual materials, authorized access, running costs and human review need confirmation in the project environment. A product description explains the intended work; a real task's effort and use require their own records.
Sometimes the best change is to ask AI to do less. If the account background has not materially changed for several weeks, retaining the confirmed brief and adding new information may be more sensible than rebuilding it daily. A task that puts a few known values into a defined format may be adequately served by a spreadsheet or conventional automation. Such choices can make a project more sustainable by reserving expensive processing for work that requires understanding and judgment.
Separate exploration from routine use
Experiments need room for discovery. Otherwise teams have little chance to find useful applications they did not anticipate. Their expense should still be identifiable and discussed separately from routine work that already delivers consistently. An experiment without a usable result can end with its findings preserved. It does not need to become a permanent service to justify the initial purchase.
Routine work needs an explainable cost range. A budget owner can plan for expected volume, required quality and exceptions, with room for an unusually important assignment. A limit that is too tight can stop useful work near completion. An unrestricted allowance can let additional research accumulate without a deliberate decision. The operating arrangement should let an owner see a departure from the estimate and choose whether to continue, narrow the scope or stop.
Menlo's figures offer a timely reason to revisit spending assumptions. The stronger argument for an increase is an established business use and an understanding of the effort needed to deliver it. Industry spending provides background. A company's own work and results should determine where its next allocation goes.
Sources
- Menlo Ventures: 2026 The State of Enterprise AI
Source published: · Accessed:
Mokyun Editorial · neo@mokyun.com
