Budget for AI without breaking the bank
The so-called ‘AI arms race’ has been going on long enough now for organizations to make intelligent decisions about its pros and cons, and how much they want to invest in AI based on expected returns. AI tools and services aren’t cheap, but how costly is the alternative of not investing?
On the one hand, failure to robustly embrace AI could be a market differentiator. On the other, AI investments often don’t deliver everything an organization hopes. The challenge becomes how to get the best return on investment (ROI) from third-party AI tools and services, without busting the budget.
“The honeymoon phase with AI is over. The sugar rush of Copilot is real, but it fades fast if you can’t show enterprise value,” explains Carsten Krause, global head of enterprise practice at energy management firm Schneider Electric.
AI investments that get positive customer reviews
With the large-scale, and wide-scale adoption of AI technologies over the past three years, definite trends have emerged that illustrate what organizations like and don’t like about investments so far. Recent studies offer a mixed bag of reviews.
AI vendor Atlas Synth AI indicates its customers have seen the best results from advanced predictive analytics, natural language processing for real-time reporting, and AI-enabled process automation that eliminates repetitive, low-value tasks.
“These tools allow our clients to respond to changing conditions faster, improve decision-making accuracy, and redirect talent to higher-impact work,” says Arthur Hall, partner and AI consultant at the firm.
Overall, organizations are most satisfied with AI tools that produce rapid, measurable impact. Examples include generative AI platforms for marketing content, AI-powered analytics that deliver real-time business insights, and AI-driven chatbots that improve customer service response times. These solutions are appealing because they have clear use cases, low adoption barriers, and produce visible ROI in weeks rather than months.
Satisfaction also grows when AI is embedded into existing workflows, such as Copilot drafting documents, ServiceNow predicting next steps, or Salesforce Einstein surfacing the right lead.
“Organizations are also finding success with Retrieval-Augmented Generation (RAG) models layered on enterprise knowledge. Suddenly AI isn’t hallucinating, it’s giving trusted answers that free people to act faster,” Krause explains.
Industry-specific capabilities also play an important role in AI satisfaction. For example, law firms are generally most satisfied with AI tools that directly enhance productivity and accuracy in core legal tasks, says Mark McCreary, partner and chief AI and information security officer at Fox Rothschild LLP, a national law firm. Tools for legal research, contract analysis, and document drafting – especially those integrated with large language models are widely praised – including platforms that assist with summarizing depositions, drafting motions, and reviewing contracts, which have shown time savings of up to 90%.
AI tools aiding in the legal practice
In terms of their own experience with AI tools, McCreary says Fox Rothschild had adopted a suite of AI tools that support legal research, drafting, and knowledge management. The firm is most satisfied with tools that integrate seamlessly into existing workflows, offer secure, enterprise-grade infrastructure, and provide meaningful time savings without compromising accuracy.
“We are more cautious with tools that require significant customization or lack robust governance features,” McCreary says. “Our focus remains on tools that align with our ethical obligations and deliver measurable value to our attorneys and clients.”
At Schneider Electric, the biggest wins come from AI in the flow of work—embedded in CRM, ERP, or ITSM where people already operate. Those succeed because they bend to human workflows, not the other way around, Krause says. The least effective are isolated projects that require employees to change behavior without a clear benefit. Adoption fails if humans don’t see a personal win.
AI investments that don’t fully impress customers
Conversely, firms are least satisfied with tools that lack transparency, produce unreliable outputs, or fail to integrate well with existing workflows, McCreary says. Concerns around AI hallucinations, data privacy, and ethical compliance remain top reasons for hesitation. Tools that overpromise and under-deliver, especially in niche or untested use cases, often fall short of expectations.
Organizations also tend to be least happy with tools that require significant infrastructure changes, large volumes of cleaned and labeled data, or specialized technical expertise to operate effectively.
“Frustration often comes from AI products that overpromise broad capabilities but fail to address specific business needs, leading to underutilization and stalled adoption,” Hall explains. “The lack of explainability in decision-making is another common complaint, as it can undermine trust in AI outputs.”
In terms of market offerings overall, Hall says Atlas Synth is less enthusiastic about highly niche AI applications that require long development cycles or lack interoperability with existing enterprise systems. While some of these tools can be groundbreaking, the time-to-value is often too long for fast-moving business environments. Additionally, tools that lack strong governance and audit capabilities are a concern, especially in regulated industries where transparency is non-negotiable.
Determining ROI of AI investments
Determining the ROI on artificial intelligence investments requires a combination of quantitative and qualitative analysis, McCreary says. Key metrics include:
- Time saved per task (e.g., reducing 16 hours of drafting to 4 hours … or minutes)
- Reduction in error rates or rework
- Improved client satisfaction and retention
- Increased capacity without increasing headcount
“ROI should also factor in indirect benefits, such as enhanced reputation, better compliance, and improved attorney morale,” McCreary says. Firms often recoup investments through improved efficiency and higher-value engagements, rather than direct client billing.
It is important to model AI ROI before investment, Hall explains. Start with baseline performance metrics to measure the current state, then project improvements based on hard cost savings – such as reduced labor or improved efficiency – and soft gains – such as better customer experience or faster decision-making. Implementation, training, and ongoing operational costs must be factored in to determine the net benefit.
The most effective organizations also include a post-deployment validation phase to compare actual results against projected ROI. This ensures continuous improvement and provides the data needed to make future AI investment decisions with greater accuracy.
Developing an investment strategy for AI tools
Once an organization decides it definitely wants to invest in, or increase investments in, AI, the process should be deliberate and systematic, Hall says. Key steps include:
Business mapping: Document core processes, decision points, and the key performance metrics tied to success.
Capability matching: Identify AI technologies that can either accelerate or optimize these processes, ensuring they meet technical and governance requirements.
Prioritization and proof of concept: Focus on quick-win use cases with high business value to validate ROI before full-scale deployment.
“This structured approach minimizes risk while ensuring AI investments directly advance strategic goals. The more tightly the AI capability is aligned with measurable business outcomes, the faster adoption and impact will occur.
In terms of expectations from any AI investment, must-haves are measurable operational efficiency improvements, reduced risk exposure, improved decision accuracy, and compliance alignment these directly affect an organization’s bottom line and strategic resilience.
As to nice-to-have benefits, “Emerging capabilities or experimental features, while interesting, don’t directly improve critical performance metrics. These can be worth exploring, but they should never displace investment in AI solutions that have proven, tangible business impact. The priority should always be sustainable value creation over novelty,” Hall says.
Budgets for AI investments vary widely and often by industry, but in general, Hall recommends the following investment levels: Year 1 should be 1% to 3% of operational budget, focused on foundational capabilities, workforce training, and small-scale pilots to validate AI impact; Year 2 should be 3% to 5% to expand successful pilots into broader business functions, integrate AI more deeply into workflows, and establish governance and risk frameworks; Year 3 should be 5% to 7% for full-scale enterprise adoption, cross-system integration, and next-generation innovations such as advanced machine learning models or autonomous decision systems.
Budget planning should be flexible, as AI adoption is iterative. Companies should fast-track high-ROI use cases, while scaling back or retiring underperforming initiativesd.
For large firms, McCreary recommends: Year One should be $500,000 to $2 million for pilot programs, vendor selection, and initial deployment; Year Two: $1 million to $5 million for scaling, integration, and training; Year Three: $2 million to $10 million for enterprise-wide adoption, governance, and innovation.
These figures include licensing, infrastructure, staffing – such as for data scientists – and change management. Many firms view these investments as part of long-term strategic transformation.
Investing for flexibility
Organizations pondering an AI investment should understand that the most successful AI initiatives are those that combine cutting-edge technology with robust governance, clear performance metrics, and a culture that embraces change. In response, Hall says that at Atlas Synth AI, the firm emphasizes that AI adoption is not a one-time project but an evolving business capability.
“The organizations that thrive with AI are the ones that view it as an enabler for strategic agility, not just automation,” Hall explains. “They invest in transparency, explainability, and ethics alongside technical performance, ensuring that AI-driven decisions are both effective and trusted. In the next few years, the competitive gap between AI leaders and laggards will widen significantly and those who act now, with purpose and discipline, will be in the strongest position to lead.”
