🌏 閱讀繁體中文版
Your Boss Just Forwarded You an Article
“Company XX saved 30% in costs using AI”
With a note: “Should we bring in a consultant to evaluate this?”
What you’re thinking:
- The last digital transformation project still isn’t finished…
- We haven’t even organized our data yet…
- How much do consultants cost? Is this just paying tuition for another lesson?
This article is for you—the one being asked for an opinion.
First, Some Numbers: Who’s Making Money in This AI Wave?
Before deciding whether to spend money, look at where the money flows.
Supply Side (Selling Shovels)
| Player | Revenue Performance | Source |
|---|---|---|
| Cloud Providers | AWS + Azure + GCP annual revenue $330B+, GenAI contributing half of growth | Holori [Archive] |
| Consulting Firms | Accenture GenAI revenue from $100M → $900M (9x growth in one year) | Accenture Earnings [Archive] |
Demand Side (Mining for Gold)
| Metric | Data | Source |
|---|---|---|
| AI Project Failure Rate | 80%+ (2x traditional IT projects) | RAND [Archive] |
| Abandoned After POC | 30% | Gartner [Archive] |
| Companies Abandoning Most AI Initiatives | 42% (vs 17% last year) | S&P Global [Archive] |
What Does This Mean?
Those selling shovels profit steadily. Those mining for gold bear the most risk.
- Cloud providers: You pay for what you use; project success or failure doesn’t affect them
- Consulting firms: They collect fees; project success or failure is your problem
- You (the client): If the project fails, you bear the cost
This isn’t saying you shouldn’t do AI, or that all consultants are scammers.
It’s saying: Before spending money, know your position at the table.
Self-Assessment: Do You Need an AI Consultant?
Answer these 5 questions. Score 1 point for each “Yes”:
1. Do You Have “Usable” Data?
Not just “have data”—“usable” data.
- Is your data organized, not scattered across systems?
- Is there basic quality control?
- Is someone responsible for maintaining it?
Why this matters: According to RAND research, the second leading cause of AI project failure is “lacking sufficient data to train an effective model.” Implementing AI without a data foundation is throwing money away.
2. Do You Have Internal Technical Staff to Interface?
- Do you have engineers who can understand what the consultant is doing?
- After the consultant leaves, is there someone to take over maintenance?
Why this matters: Consultants won’t stay forever. Without internal staff to take over, the end of the project is when system decay begins.
3. Does This Project Have a Clear Business Objective?
Not “implement AI”—“achieve what with AI.”
- Can you articulate specific KPIs?
- Could this goal be achieved without AI?
Why this matters: RAND research shows the primary cause of failure is “misunderstanding between business and technical teams about problem definition.” If you can’t clearly explain what problem you’re solving, consultants can’t save you.
4. Can Your Budget Last Until Results Appear?
- Gartner data: Average time from AI prototype to production is 8 months
- Simple RAG document search costs $750K+
- Custom model fine-tuning (like Llama) costs $5-6M
Why this matters: Many projects die from “budget exhausted before seeing results.” Budget must account for the possibility of failure.
5. Does Your Boss Have Patience?
- Can they accept 6-12 months before seeing initial results?
- Can they accept “this direction isn’t right, we need to adjust”?
Why this matters: S&P Global’s survey shows companies abandoning AI initiatives jumped from 17% to 42%, partly because “executives are impatient to see returns.” Impatient projects usually die fastest.
Scoring Results
0-2 Points: Don’t Rush, Build Foundations First
Your data, personnel, and objectives may not be ready. Hiring consultants now is probably paying tuition.
Recommendation: Spend 3-6 months organizing data, defining problems, and building internal awareness.
3-4 Points: Can Evaluate, But Be Cautious
You have some foundation, but there may be gaps. You can engage consultants for “assessment,” but don’t sign large implementation contracts yet.
Recommendation: Start with a limited POC (proof of concept), controlling budget and timeline.
5 Points: Ready to Engage Consultants
Foundations are in place. You can seriously evaluate consultant partnerships. But still watch for red flags in the next section.
Red Flags: Don’t Hire These Consultants
If any of these apply, consider talking to someone else:
1. Only Talks “AI Transformation,” No Specific Use Cases
“We’ll help you implement AI, unleash the power of your data, achieve digital transformation…”
Problem: If a consultant can’t articulate what specific problem AI can solve in your context, they’re probably just using templates.
Ask them: “Given our current situation, what specific scenario would you recommend starting with? Why?”
2. Guarantees Success, Quick Results
“We can go live in 3 months, guaranteed ROI.”
Problem: Gartner data says prototype to production averages 8 months, and 30% are abandoned after POC. Those guaranteeing success are either overly optimistic or inexperienced.
Ask them: “What percentage of your past projects didn’t meet expectations? What were the reasons?”
3. Unwilling to Discuss Data Quality
“We’ll deal with data later. Let’s look at the technical architecture first.”
Problem: Data quality is key to AI project success. According to Informatica’s survey, 43% of companies cite “data quality and readiness” as the biggest barrier to AI success. Consultants avoiding data discussions may just want to sell technology.
Ask them: “Having assessed our current data situation, how much time and cost do you estimate for preparation?”
4. No Failure Cases to Share
“All our projects have been successful.”
Problem: 80% of AI projects fail. If a consultant says they have no failure cases, either they haven’t done many projects or they’re not being honest.
Ask them: “Can you share a case that didn’t meet expectations? What did you learn?”
Green Flags: What Good Consultants Look Like
1. Asks About Your Problems First, Doesn’t Push Tools
Good consultants spend time understanding:
- What problem are you trying to solve?
- What have you tried? Why didn’t it work?
- What are your constraints? (Budget, staffing, timeline)
Not: “We have the latest LLM technology that can help you…”
2. Willing to Say “You Might Not Need AI”
Some problems don’t need AI to solve. Good consultants will honestly tell you:
“This problem can be solved with a rule-based system; you don’t need AI.” “Your data volume isn’t sufficient; AI isn’t cost-effective right now.”
Consultants willing to turn down money are usually more trustworthy.
3. Has a Complete POC → Production Path
Good consultants will tell you:
- What should POC validate? What counts as success?
- After POC success, how do you scale to production?
- What infrastructure and staffing does production require?
- Who’s responsible for long-term maintenance?
Not: “Let’s make a demo to show the boss first.”
4. Willing to Share Mistakes They’ve Made
Good consultants proactively say:
“We’ve done this type of project before. Here are some pitfalls to watch for…” “A previous client had a similar situation to yours. They ended up…”
Experienced people know where problems occur. Inexperienced ones just say “no problem.”
Scripts for Reporting to Your Boss
If You Want to Say “Let’s Do It”
“According to McKinsey’s 2025 report, 88% of companies are already using AI, but only 6% are actually generating significant business value. The question isn’t whether to do it, but how to do it so we can be in that 6%.
I recommend we start with a POC on one specific scenario, budget capped at XX, and evaluate results in 3 months. If successful, we expand. If not, we’ve limited our losses.”
Source: McKinsey State of AI 2025 [Archive]
If You Want to Say “Let’s Wait”
“According to S&P Global’s 2025 survey, companies abandoning most AI initiatives jumped from 17% last year to 42%. Gartner also predicts 30% of GenAI projects will be abandoned after POC.
The main reasons are insufficient data quality and unclear business objectives. I recommend we spend 2-3 months organizing data and clearly defining the problem before evaluating whether we need external consultants. This will improve our success rate.”
Sources: S&P Global [Archive], Gartner [Archive]
Decision Recommendations
| Your Situation | Recommendation |
|---|---|
| Short-term needs (1-2 projects) | Consultants are more cost-effective than building a team, but control the scope |
| Long-term strategy (building AI capabilities) | Consultants + internal team in parallel; consultants help start, internal team maintains long-term |
| Completely unfamiliar | Take courses, study cases, build basic understanding first, then decide whether to engage consultants |
Conclusion
AI consultants aren’t a silver bullet, nor are they scammers.
They’re tools. How well tools work depends on whether you know what you want.
Before spending money, ask yourself:
- What problem am I trying to solve?
- Does this problem really need AI?
- Do I have sufficient foundation to support this project?
- Am I willing to bear the risk of failure?
If you have answers to all four questions, you’re ready.
If not, finding those answers is more important than finding consultants.
References
- RAND: The Root Causes of Failure for AI Projects [Archive]
- Gartner: 30% of GenAI Projects Will Be Abandoned After POC [Archive]
- S&P Global: AI Adoption with Mixed Outcomes [Archive]
- Capgemini: The AI-powered Enterprise [Archive]
- McKinsey: The State of AI 2025 [Archive (PDF)]
- Accenture FY2024 Earnings Report [Archive]
- Holori: Cloud Market Size 2024 [Archive]