"A consultant is someone who saves the client almost enough to pay his fee" - Arnold H Glasgow

I wanted to do a review of what consultancies where saying about AI in 2025 and its trajectory and both critique it but also complime it so others could more easily get a set of different views.

What I have found is disparate and confusing but only because AI is so all encompassing at the moment a moment I believe we are destined to fail to meet today but have full faith we will surpass and then some in the long term.

I wanted to write this all down as it forces me to read a variety of different opinions on AI. I supply as a resource to others as it's very link centric.

I wanted to put this about not as a opinion piece but that people might have access to the variety of opinions coming out of consultancies in one place.

recommendations 

I cannot help but have a opinion. I recommend listening to these.

Eviden

https://eviden.com/publications/tech-radar/artificial-intelligence/ 

Boston Consulting Group

https://web-assets.bcg.com/b6/13/feb499c545b5baba750741f657e6/gen-ai-increases-productivity-and-expands-capabilities.pdf 

Bain and company

https://www.bain.com/insights/topics/ai/ 

The World Economic Forum

https://www.weforum.org/publications/blueprint-for-intelligent-economies/ 

And with some very limited reservations listed below McKinsey 

My Analysis of the analysis

I have tried to grade them on my own interpretation of who I think is making exaggerations. I try to describe that view point and why I made that decision.

Eviden

I really like this tech radar from Eviden. Yes I used to work for them. I think it should detract from the fact it's super simple and visual; it probably fits on to a power point slide.

tech-radar

Not once does it mention AI agents. I like that it has physics informed AI on the list.

A thing I have always enjoyed with Eviden and Atos is they do actual case studies and print customer stories. I think actual examples of how AI can be used today beats predictions about what AI might be tomorrow.

Example:

enhancing-fraud-detection-with-ai-and-digital-data-at-esure 

That is why there top of this list.

A criticism I have of all the others is in a sense they write long reports like a university produced white paper. I like the visualisation and simple language.

I think it has greater impact knowing time and place, the who where and when from the case studies of their own work.

it's more meaningful.

Boston Consulting Group

I really liked Boston Consulting group. It had a front of house article that summarised everything, a methodology page and then links to the individual reports on their findings.

The gist of their reporting is that gen AI makes knowledge workers more productive that it does this by lifting everyone's skill level to some degree.

50% better at data cleaning. 13% better at predictive analytics. 20% better at statistical analysis. All on some really interesting and tidy graphs.

Source:gen-ai-increases-productivity-and-expands-capabilities.pdf 

I liked how they admitted there was a problem of knowing if these skills where maintained or would be learned; or lost.

Then I checked there section on AI agents.It's a trigger for me when people start telling me AI agents have agency, they are goal oriented. There going to take your job both because I think they are wrong but I think it's instant evidence someone is trying to shock for clout.

ai-agents 

Says AI agents are just AI using tools. that's my own position. They do hype it a bit saying that AI is autonomous mentioning planning in AI etc. I think this will be true in the longterm and disagree in the short but; that seems on point. My problem with agents is most are a wrapper over a LLM currently and sold as a whole worker out of the box.

I still think this is an over the horizon capability for AI to do repeatedly and a adaptable way because that would imply a causal model (see last weeks post) and no one is good enough to build that amount of models and put them everywhere and in some sense what a lot of companies are doing is implementing a large language model at each of those tasks.

I would criticise this because it's like a lot of people talking about agents they do not mention their stack. A LLM powered agent would work fine in the lab but would fail in production making it dangerous. This is true because a non causal model would work fine until it found something it failed on and then would be a bit broken.

So I think most agents are not causal models and are just LLMs being sold as agents.

Boston Consulting Group Conclusion

im now subscribed to their newsletters. I like how simple the graphs are.

Bain and Company

source:bain

I liked Bain its website is just a series of articles on specific points. A lot of it is from other sources Forbest etc from which I found this refreshing take.

"But for all the noise surrounding generative AI, it’s easy to forget something fundamental: Generative AI–related transformation is still just change. It follows the same human patterns leaders have navigated for decades. The difference now? The velocity, the visibility, and the stakes.

Executives and investors are confronting a dual challenge: They must apply time-tested change principles and wrestle with unfamiliar questions about trust, experimentation, and scaling. Understanding this duality is the key to leading effectively in the AI era."

In some ways that is a bit lazy. I don't think so if someone has said your view better why not just sign post to it. Reference it, give credit. I like access to a AI reports and views and this is just a source of different opinions.

I like Bain in the same way I like Apple News. It has lots of different sources all in one place.

If you scroll down their website they then have dedicated reports on what their consultancy believe is the way forward on AI.

Examples are

this:understanding-the-five-types-of-ai-consumers

which is a very data driven look at the types of people who use AI and how consumers will react. It shows that people are using it more but that there is a stubborn population of users who look unlikely to use it. This is just a bit bigger than 20% of users. It says to me that it is not a great idea to see everything as a nail for AI to hammer in 20% of consumers is a large enough demographic that no business can afford to lose.

parsing-how-winners-use-ai-commercial-excellence-agenda-2025

They are showing that all companies are now deploying AI and about 30% have scaled two into production and most companies (62%) have scaled at least one into production. It would seem to suggest it's the move to production and not the test phase that is hard.

Its a great article splitting up the company traits of those that are winning versus those that are losing the race and it's in terms of financial gain. As we move on you will find im often quit critical at the absence of financial information. This has that detail many do not.

Gartner Reprint

I can't get actual Gartner... it costs money... so assume this is just all I got. It's important because everyone below references Gartner. If you know the history of Gartner you know why.

Source: gartner

"Agentic AI will introduce a goal-driven digital workforce that autonomously makes plans and takes actions — an extension of the workforce that doesn’t need vacations or other benefits. This research describes the opportunities and dangers of agentic AI for IT leaders, and explains how to prepare."

Opportunities

• Agentic AI gives AI new levels of agency (the ability to select what actions to take for achieving particular outcomes). It’ll provide a significant opportunity for performancegains that will increase over time as the systems evolve to more effectively achieve their goals.
• Agentic systems will change the future of decision making. They can quickly analyze complex datasets, identify patterns and act. This will avoid labor-intensive data modeling, lead to better problem solving, reduce time to action and enable new concepts of scale.
• Agentic AI will dramatically upskill workers and teams, enabling them to manage complicated processes, projects and initiatives through natural language. However, the orchestration and governance of autonomously acting software entities require advanced tools and strict guardrails.

Recommendations

• Integrate agentic AI into your strategic planning. Define the levels of agency you’ll allow in each environment and workflow.
• Design agentic AI solutions to connectdisparate applications and data, while prioritizing improved user experience and efficiency. Map decisions and actions that can be automated between previously siloed data and processes.
• Establish clear implementation and operational guardrails for agentic AI,including legal and ethical guidelines on autonomy, liability, security and data privacy. Ensure that agentic AI identity, security and monitoring capabilities are robust enough to protect your organization, employees and customers.

Strategic Planning Assumptions

By 2028, 33% of enterprise software applications will include agentic AI, up from less than 1% in 2024.By 2028, AI agent machine customers will replace 20% of the interactions at human-readable digital storefronts.By 2028, at least 15% of day-to-day work decisions will be made autonomously through agentic AI, up from zero percent in 2024.

Critcisms of Gartner

I can't read the full report as I'd have to pay but I find places later on where they quote Gartner I want to be able to reference what they did say; so this is the best I could find.

They make predictions and put percentages to those predictions and I think this is a bad idea because it's a high surface area to be wrong about.

Particularly I am doubtful about AI agents doing financial transactions. I feel a website that I just click and press the buttons is as fast as telling a AI and risking it doing something wrong. Though I could see that change with my groceries as it's a chore but surely I only need to geberate and input that list once 

Also if we start building AI agents buying stuff everywhere I feel you can also just build APIs and that would cost less. Whenever people say AI is going to take X amount of work I often fail to believe them where a cheaper traditional approach works here it's APIs offer more security and selenium web scraping will do it cheaper and with less compute. Sure I might have the AI write the selenium code for me after analysing the website but again I do that once. Also more broadly searching for goods on a website is a very easy experience im not sure why an AI would be better as people don't use hugely fuzzy search terms.

Score 3/10 I like Gartner I wanted to get their report and the summary I find I can't agree with. I do not want predictions about AI what I think is critical is what do CEOs do now?

But also it's Gartner... they make mistakes but they get more right than me who you going to believe.

Mckinsey

Source:the-state-of-ai-how-organizations-are-rewiring-to-capture-value_final.pdf

A key message I notices across all of this is you need to build a new organisation, put a senior leader in charge. Afterwards your going to be completely changing the org chart anyway.

McKinsey I think might be the source of this as it says that one of the biggest predictors of AI and it's successful use is CEO oversight.

A ongoing criticism I have is no one tells you about sales or actual cost cuttings.

"Twenty-one percent of respondents reporting gen AI use by their organizations say their organizations have fundamentally redesigned at least some workflows."

There is no statement on the impact on that workflow. It feels like if I reword it what it actually said only 21% of whom said they use AI actually did and remodelled at least some workflows. It just feels that data is missing and where it exists it amounts to someone else is using AI.

McKinsey says: "Twenty-eight percent of respondents whose organizations use AI report that their CEO is responsible for
overseeing AI governance." you see what I mean it doesn't tell you anything about outcomes."

A theme from McKinsey was the problem of AI safety and that companies aren't following identified provesses. "Less than one-third of respondents report that their organizations are following most of the 12 adoption and scaling practices for gen AI."

What it reports is most companies appear to be setting up a PMO for project management. I feel this seems the right idea in one sense but it suggests that the best way to manage AI is project management based ie business and not data or analytic driven. I know not all PMOs are project support functions but it lept out at me that the AI things made of data who are processing data are not being watched by people skilled in data; they are potentially skilled in PRINCE2.

McKinsey say "Twenty-seven percent of respondents say employees at their organizations review all content created by gen AI before it is used, and a similar share says that 20 percent or less of gen- AI-produced content is checked." This really shocked me bearing in mind we have not been using AI content for a long time it made me seriously worry if this was filling databases with rubbish.

It reports respondents are most concerned about security and inaccuracies.

It states for some reason that the number of data visualisation jobs has declined and for some reason they are singled out (which includes my job). I did google searches asked an AI and checked my LinkedIn all appear untrue.

A key take away I found interesting is in the organisations using the AI it's the CEOs that report using it more than any respondent and I wondered if there was a implicit bias that because this AI is useful to me it must be universally useful.

Criticism of McKinsey

I also tried to not like McKinsey report. I found it charming and data driven nonetheless. 

I liked McKinsey report more than others. I think the emphasis on transformation means that it avoids making any wild predictions.

I feel it's guilty of creating this idea that the best way to implement AI is a PMO. Now there are IT PMOs but these are typically related to software development and deployment across the org. I feel that AI as a trend has been unkind to developers and I worry there is a missed opportunity as the process and team dynamics are not explored.

If it is concluded that AI is best implemented by a Programme Management Office. I would assume this office is transistory building AI projects to then hand off the asset. But this is a assumption on my part and I think it underlies the problem here what happens to the AI and the AI team? Do we manage the AI by IT? If they're so damm agentic HR (I'm partially joking)? Does the AI get a AIR team (Artificial Resources anyone?) that the PMO transitions into? If it transitions into a AIR what skills does the AIR need is it data analysis skills? If you create a AIR isn't that just a RPA automation team with a funny name? Won't that AI use go closer and closer to just plain business as usual operations if so are we back to dismantling this PMO when finished?

You get my point... it's sort of suggested that AI is its own special skills without anyone saying what those skills are. This bugs me I feel thar a lot of AI might be worth being a second and not first mover on to let other organisations make these initial mistakes and work back from the successes and avoid the pitfalls.

I was really thankful that McKinsey had not excessively mentioned agents and agentic AI. I found McKinsey later in my searches I was so sick of it by then.

I feel some of their graphs are chart junk being sort of the accounting style graphs where they try to show you they are good at using data visualisation software but could be easier if was just a bar chart. When speaking in my domain mentioned data I could quickly denounce as untrue and a mistake and I have a policy of extending my distrust to everything you say if your wrong about something I do for a living but it wasn't a stupid mistake it's likely a mistake.

AI and automation trends 2025 UI path 

source: https://start.uipath.com/rs/995-XLT-886/images/ai-and-automation-trends-2025-ebook.pdf 

• AI gains capability to think plan and act on its own.
• tech providers create a environment where people, robots and AI work together in harmony.
• Agents get to work on long tail automation projects.
• Job sharing with the machine the great work relocation begins.
• built in AI: lifts companies from the trough of disillusionment. (Companies will struggle to capture value by themselves but tech companies will take up the gap).
• From RAG to riches: new AI tame the data deluge.
• Regulation escalation: world powers act to reign in AI.

Those are the headlines they quote Gartner and IDC to tell you Agentic AI is here.

Both Gartner and IDC costs money. Like see below... I'm not paying that. There is no data it's not saying Agentic AI is here and made Company X (not the X I mean a theoretical company X). Then they tell you they will all work together you'll want lots of them. Then they tell you they'll work with the people.

They quote their own CEO on this. I am sure they are unbiased.

They also quote McKinsey.

“The value that agents can
unlock comes from their
potential to automate a long
tail of complex use cases...
that have historically been
difficult to address in a cost-
or time-efficient manner.”9
-McKinsey & Co., “Why Agents Are the Next Frontier of Generative AI,” 2024

The examples they then give for "Agentic AI" has no rhyme or reason. I have already written an article on why I do not think anyone really knows what agentic or agents in AI means. If you look at the historical definition what they are selling is not agents.

"A study from OpenAI estimates that AI could take on half the work of almost 20% all workers.10 McKinsey estimates that, by 2030, 30% of all work hours will be performed by machines, not people.11 And these studies were completed BEFORE agentic AI had fully emerged."

I think this is a really strange quote half of 20% is 10%. Also in some sense all the work is currently done by people on laptops.

"The C-suite is on the front lines of this change, challenged to usher their companies into a dimly lit— but fast-approaching—future state. They’ll be assisted by a burgeoning rank of consultants and operations designers specialized in conceiving of new AI operating models, managing massive change, and creating and implementing cross-enterprise agentic systems.
Human resources will have to retrain and upskill tens of thousands of employees in using new AI tools and
partnering effectively with agents. They’ll need to find new workers with the right combination of technological skills and core capabilities in critical thinking, problem solving, and creativity. They’ll have to rethink hiring plans and redo evaluation and reward systems."

Its up to you C suite you and your consultants god bless you. The bit that I find odd about this is the entire tag line is you need us to tell you how to rethink your business and afterwards your going to need to rethink the entire business and HR because they just won't have the skills. The hidden message is everyone in your business is kind of unable to understand new technology your gonna need us to rebuild this whole thing for these AI that have just appeared this year. This is despite the fact the term AI agent and agentic AI goes back to the 1990s.

Then there's a lot of stuff that looks like RPA with the term RPA replaced with AI.

I work in data I could just print the entire page on RAG to riches about data. It basically says data is scattered around the business 

"According to a Gartner study, almost half of digital workers say they struggle to find the data they need to effectively perform their jobs. The average employee wastes almost 3.5 hours per week just dealing with information burden. And 38% of workers say they must work extra hours to get on top of it.24"

The thing is here is data sits in tables you have search tools to get that data. The only way you'd need AI to get that data is if you did not know where it was. If you don't know where it is it's probably for security reasons and me in data doesn't want to give it to you. AI is not going to help you find data you do not know exist. No database administrator is going to open up their definitions to everyone for security reasons and where they do it's called a data dictionary.

I can also tell you why I don't think any of this will work from the below.

"Retrieval augmented generation (RAG) improves GenAI models’ performance by giving them access
to real-world data while they generate responses, elevating GenAI “from a neat parlor trick to a business advantage.”28 For example, RAG has helped a global consulting firm cut the time consultants spend searching for information by 40%, saving $5 million annually.29 RAGs are projected to chalk up a 44.7% CAGR, 2024 to 2030, hitting $9 billion by the end of the decade.30"

The emphasise was this was useful for consultants therefore it is useful for everyone. When consultants get called in you have about a week for discovery of data and a AI tool rushing through and summarising and analysing is a game changer.

Me as a data analyst I know my databases. I know my pipelines I do not need to search for anything. It's there where I want it because I put it there. I know who can access what because I vetted them and gave them access. I would not deploy something to do data analyst using GPUs and large amount of compute to summarise the data for them because if I've done my job I know every day they ask for the same thing and then I just deploy the minimum access they need that gives the company high security letting only these people access that data.

They are literally saying here hey you should send your data all of it across the internet through a API have it processed by a GPU and linear algebra to search your entire database structure to do a query. That is crazy...

UIpath Critcisms

I am not going to bother... I just thought it quoted Gartner on there most outlandish predictions. Quoted McKinsey to make it sound like when they where talking that they explicitly meant the agents that Gartner thought could happen by 2028 where happening now...

Explicitly this was not what I read when I read those documents.

I had a feeling with McKinsey that ok some of there data I can find data from other sources on the amount of data visualisation jobs. That's not a big thing. With UI path I mean I just don't agree on the impacts on the jobs I have most experience with.

If you don't believe someone on the stuff your most experienced with how do you believe them on anything else?

When speaking in my domain it mentioned data and related subjects and I could quickly denounce as untrue and a mistake and I have a policy of extending my distrust to everything you say if your wrong about something I do for a living but it was a stupid mistake and I think it's unlikely a honest mistake.

0/10 cannot recommend

Moving on.

IDC

How much are you charging for telling me AI is cool?

That much? Nah mate...