10/30/2025
Do you think we are done talking about AI? No, it’s clear now that this technology didn’t’ show yet all its power for companies and individuals. AI is here, it’s unstoppable, and it’s creating measurable competitive advantage for firms that implement it correctly. How can we make this transformation a success?
As one of the leads of our solution analysis team, I’ve been coordinating our AI transformation over the past few months. When our company embraced the power of agentic AI, they partnered with a Silicon Valley firm to train some management leaders. I was part of the squad representing solution analysts, and from there, I led our AI adoption initiative with a core team. What follows isn’t theory, it’s what we experienced: the framework we built, the results we achieved, and the lessons we learned the hard way. I shared the key learnings from this experience with my colleagues; I share it for external audience through this article.
We started our AI journey by identifying where consultants spent the most time on low-strategic-value work. Here’s what we discovered:
AI excels at tasks that are repetitive yet complex – work that requires consistency, speed, and accuracy but not strategic judgment.
In software consultancy specifically, four areas consistently deliver strong ROI:
- Configuration work translating business requirements into technical setups with precision and speed
- Report development generating optimized SQL queries and navigating complex database structures
- Document management automating schema mapping, variable identification, and technical documentation
- Technical support handling repetitive troubleshooting patterns with consistent quality
These tasks consume enormous consultant time but add limited strategic value. That’s your opportunity. AI can handle the analytical heavy lifting while your consultants focus on client relationships, complex problem-solving, and strategic advisory work.
The firms winning with AI aren’t replacing expertise, they’re amplifying it. That’s exactly why we launched our AI initiative.
After months of experimentation, we developed a disciplined approach that produced reliable results. Here’s the four-step framework we followed:
Step 1: Analyse Your Work and Identify AI Opportunities
Before touching any AI technology, get full clarity about how your consultants spend their time. What’s repetitive but data-intensive? Which activities need accuracy more than creative judgment? Where does complexity come from information navigation rather than strategy? Don’t start with AI capabilities, start with operational bottlenecks where speed, consistency, and pattern recognition solve real problems.
Step 2: Rethink Tasks for AI Augmentation
Don’t ask: ”Can AI do this task?”
Ask: ”How can AI make this faster and more accurate while keeping human judgment central?”
For each opportunity, determine what AI automates fully, what it assists with, and what stays in human control. You’re not replicating workflows, you’re redesigning them for human-AI partnership, leveraging what each does best.
Step 3: Prepare Quality Documentation
AI agents perform exactly as well as the documentation you provide them. Exceptional documentation produces exceptional performance. Poor documentation produces hallucinations.
Documentation must be complete, precise, clearly structured, and unambiguous. In some cases, reformatting into markdown or structured hierarchies improves AI parsing accuracy. This foundation determines whether your AI implementation succeeds or fails.
Step 4: Create, Test, and Refine AI Agents
Design specialized AI agents focused on specific areas. Be precise in your instructions: exactly what the agent should do, what outputs you expect, clear boundaries, and explicit criteria for when human input is required.
Test with real implementation scenarios, real configuration challenges, genuine edge cases. Track accuracy rates, time savings, error types, and situations requiring human intervention.
Refine based on real-world testing: documentation gaps, prompt clarity, output formats, escalation criteria. This cycle: create, test, refine, repeats until the agent performs reliably in production scenarios.
When selecting your initial AI use case, focus on high-impact areas where success has been demonstrated across consultancy environments. I want to focus on three applications that have shown good results.
Application 1: SQL Query Generation
Report development is repetitive, follows logical patterns, and requires precision over strategy, perfect for AI.
Train an AI agent with database documentation and example reports. It analyses requirements and generates optimized queries, dramatically reducing development time and eliminating syntax errors. Bonus: consultants learn SQL best practices from AI explanations.
Limitation: Human review is mandatory. AI handles standard requirements well but struggles with highly customized configurations.
Application 2: Configuration Translation
Configuration work is time-intensive and detail-heavy. AI can translate business requirements into technical configurations with correct sequencing and dependencies.
Feed your agent comprehensive documentation about configuration options and implementation examples. For complex workflows, consultants receive validated blueprints in minutes, with AI identifying missing requirements early.
Limitation: Cannot handle multiple complex elements simultaneously. Work iteratively, feed requirements one feature at a time for better results.
Application 3: Data Mapping and Document Management
Data mapping is knowledge-intensive but repetitive, perfect for AI automation.
Train an agent with complete schema descriptions and mapping rules. AI automates most of the standard mapping work with dramatic error reduction, letting consultants focus on genuinely custom requirements.
Limitation: Client-specific customizations require human refinement. AI excels at standard patterns but needs expert guidance for unique requirements.
After months of implementation across multiple use cases, we identified patterns that consistently separated success from failure.
Foundation Requirements
Deep domain expertise is essential: documentation quality is determinative: complete, precise, clearly structured documentation (consider markdown for example for better AI parsing). Products evolve, so treat documentation as living and budget for ongoing maintenance.
Execution Discipline
Start small, one-use case, validated results, measurable ROI, then scale.
Design for iteration: breaking complex tasks into sequential steps produces better results than bulk processing.
Build explicit human review checkpoints: AI handles most standard scenarios but needs human intervention for strategic judgment.
Context-Specific Assessment
AI performance varies significantly by task type, data structure, and customization requirements. Assess each opportunity individually rather than assuming universal applicability.
Performance Management
Track concrete metrics from day one: time savings, error rates, consultant satisfaction, client impact.
Test with real scenarios, not hypotheticals, actual client requirements expose documentation gaps.
Create clear escalation paths defining when and how AI hands off to human experts.
Gather consultant feedback continuously, users know where AI helps and struggles.
Learning Mindset
As a learning tool, AI explains reasoning and offers multiple approaches, consultants build skills while accelerating delivery. Invest in ongoing refinement: continuous documentation updates and prompt optimization based on real-world performance patterns.
The difference between transformative AI implementations and disappointing ones comes down to these fundamentals. We learned this the hard way. Technology is ready. Execution discipline determines outcomes
The reality we didn’t expect
After months of implementation, the biggest surprise wasn’t what AI could do, but how it transformed the way our team works.
We expected time savings and got them. But we also got consultants excited about tackling complex SQL because they learn with every AI interaction. Juniors ramp up faster because the AI shows its reasoning. Seniors, freed from tedious tasks, now focus on strategic thinking.
Execution Discipline
Start small, one-use case, validated results, measurable ROI, then scale.
Design for iteration: breaking complex tasks into sequential steps produces better results than bulk processing.
Build explicit human review checkpoints: AI handles most standard scenarios but needs human intervention for strategic judgment.
The hard truth about implementation
It’s messier than it looks. Your first agent won’t meet expectations. Documentation gaps will surface. We’ve seen AI confidently produce wrong configurations due to vague documentation and spent hours refining prompts that should have worked.
But here’s the key: success with AI isn’t about the best tools it’s about doing the unglamorous work. Writing clear documentation. Testing real scenarios. Gathering feedback. Iterating constantly.
Where we are now
We’re still learning. Each week brings new insights and challenges. What we’ve learned so far is that AI works best when treated as a partner one that benefits from guidance, clear boundaries, and ongoing improvement, rather than as a quick fix for every problem.
The competitive advantage isn’t in having AI. It’s in implementing it with the discipline that produces reliable results.
Roua Ben Amara, Senior Manager
Share this article: