Given IBM consulting’s genAI rollout, should CIOs rethink consulting strategies for 2025?

When IBM this week introduced its genAI program for consulting, it didn’t reveal any meaningful differentiators when comparing its offerings to what every consulting firm, and enterprise, is already doing with genAI. But it did talk about cost implications, and implied that CIOs would be better off using the vendor’s systems, even with the Big Blue markup.

Analysts pointed to the argument as a key reason why CIOs need to put in place strict criteria and questions to help them determine, on a case-by-case basis, when it makes sense to pay for a partner’s AI systems and when it doesn’t.

Mohamad Ali, SVP and head of IBM Consulting, told reporters in a media briefing that the scale that IBM is trying to provide could deliver efficiencies that enterprises might not be able to replicate.

“We have 160,000 consultants, and our aim is to give each one of them ten [AI] agents and assistants to help them do their job,” Ali said. “So if you think about that, that’s 1.6 million digital workers that we will have that will supercharge our consultants at that scale.”

Ted Schadler, a Forrester VP/principal analyst, said that that comment nicely encapsulates IBM’s AI consulting argument.

“This is about augmenting expertise and throughput, both valuable to a CIO. This allows a consultancy to price output differently — less time for the same output,” Schadler said. “Buried in [Ali’s comments] is that the consultants build the digital assistants, the agents. This is bottom-up, led by the practitioners doing the work. That’s important as a pipeline for innovation in tooling, IP collection and deployment, and putting knowledge to work.”

Ali also argued that this genAI effort allows IBM to deliver better outsourcing, as opposed to strict consulting, to enterprises. 

“We’ve been augmenting the people who run these processes with genAI based digital workers. And an example is a large industrial company where we’re handling their finance processes, including cash collections,” Ali said. “And so we built a family of genAI assistants to help the cash collection agent collect cash faster.” And the effort succeeded, he noted — productivity increased.  “But even more important than that, they were able to reduce the number of days it took to collect that cash.”

Schadler noted that this could be a critical area for CIOs to explore.

“I believe that managed business services-as-software — aka BPO — is a big driver of consulting revenues going forward. It creates stickiness if nothing else. There is a cost arbitrage and a quality play that can be built into a well-structured managed services deal,” Schadler said. “As a CIO or LOB [line of business] owner, it’s a bet that the partner can do a better job than you can internally. This is just as true in code development as it is in invoice reconciliation or claims processing. The arbitrage moves from labor to genAI-powered automation.”

But Schadler stressed that it doesn’t always make sense to use a consultant’s tools, and that CIOs must craft the questions they need to ask themselves to figure that out.

For starters, Schadler argued that most enterprises would rather leverage their own genAI systems, given that they have already paid for them. 

From an enterprise CIO’s perspective, Schadler said, “I would rather you use mine. I want to maybe use yours but I want to know the pricing implications first. How about if we use your (consultant) platform but you do not charge me for the delivery of that?”

Schadler said that not every CIO has focused on the implications of genAI consultant offerings, but that they need to do so right away. “Should you rethink your service provider strategies when dealing with genAI produced output? The answer is yes, you should absolutely do so.”

In some cases, he said, the best strategy is to backburner the genAI issues and instead focus on pure ROI. “Have high incentives on the part of your partner to just get what you need to get done,” as cheaply, quickly, efficiently and cost-effectively as possible, Schadler said. If the partner can best do that using its own AI systems, that’s fine. But let the numbers and the deliverables dictate, or at least strongly influence, that decision.

“Maybe that means not using a generalist when you need a specialist. It could mean the difference between delivering high expertise and high repeatability,” Schadler said. “It’s not about the code generator as much as it is about empowering their people.”

Just asking the right questions internally is a powerful start. “Do I get more benefits as the CIO from using my own system and making everyone else use it? I don’t think we know that yet,” Schadler said. “I want to know what models and what knowledge graphs and what fine-tuning you are using in delivering this work. Critically, I want to see the tooling. Seeing the tooling, to me, is just like seeing the resumés of proposed team members.”

Another critical factor is data management and mechanisms to prevent data leakage. If the enterprise’s team is providing information that goes into the consultant’s genAI systems, what mechanisms are in place to prevent it from being seen by other teams? Other clients? 

What happens if the consultant suffers a data breach and your sensitive data finds its way to the dark web? What do your contracts say about consequences? How does this impact both your compliance and cybersecurity processes?

“We need to know where the lines of responsibility for data lie,” Schadler said.

Another IBM executive at that media briefing was Dario Gil, SVP and director of research. Gil noted some changes in how the industry has to view AI mechanisms.

“Two years ago, I think we were all in an industry where there are only certain types of models that were going to meet all use cases,” Gil said. “If you look at the actual use cases that people need to solve in the real world, that assumption is extraordinarily costly. And revisiting that assumption is a huge unlock and the opportunity to scale.”

Schadler agreed with Gil’s point. “We believe that firms will have hundreds of models in deployment — as many models as applications. That runs counter to the primary belief system in place today,” Schadler said. “So model operations, model deployment, model quality, model costs, model security, become critical checkboxes in any sourcing decision.”

Ali argued that these same kinds of changes will also have to impact consulting businesses, pointing out that IBM has dozens of models available, and it’s important to choose not only the most cost effective one for a given task, but to understand the implications around intellectual property.

“We’re integrated with Adobe Firefly. We’re also integrated with Dall-E. So if you want to generate an image, and you want to do it for extraordinarily low cost, you can use Dall-E, but you have to use it in a way where you know the intellectual property concerns don’t exist. If there are intellectual property concerns, then it routes to Firefly, because Firefly has intellectual property protection,” Ali said. “So this is a very good question, being able to select the right models, select the right assistants and use models, is actually a complex thing. We’re solving it at IBM, but it’s going to be a thing that needs to be solved more broadly.”