Answering 2025’s Top Five AI Questions
At twisted loop, we’ve spent the past year helping scale-ups, investors, and ambitious business leaders navigate the AI shift in their companies, moving from pilots and prototypes to AI-enabled operating models. Below, we’ve distilled the five questions we hear most often, and the practical ways leaders are turning AI from an experiment into an advantage.
1. How do I measure the effectiveness of my AI usage?
Measuring ROI from AI isn’t about tracking how much software you’ve bought - it’s about understanding how your organisation is transforming. The real return comes from how work gets done differently: freeing people from repetitive tasks, enabling faster, better-informed decisions, and unlocking time for value-creating work.
That’s why we often frame it as Return on Intelligence, not just Return on Investment. It’s about how much smarter your business becomes because of AI. You can measure that through:
Decision speed: Are decisions being made faster or with higher confidence?
Forecast accuracy: Are your predictions improving due to better insights?
Customer experience: Are customers happier, responding faster, or converting at higher rates?
Employee satisfaction: Are your teams spending more time on meaningful work?
These signals often precede the financial returns that boards tend to focus on. Over the short term, the monetary impact might not look dramatic - efficiency gains and time savings take time to compound - but those who use AI to fundamentally re-design business models or open new revenue streams see ROI faster.
To make measurement practical, build AI evaluation directly into your processes. Define the “before” metrics (decision time, conversion rate, hours spent) and measure them again after deployment. AI can even help here: models can self-audit, tracking usage, efficiency, and anomalies automatically.
Ultimately, ROI from AI is both a financial and cultural signal. If your teams are adopting it willingly, applying it to new problems, and talking about new ways to create value, that’s already a powerful indicator of success.
2. Confidentiality – How do I keep my data secure?
Data confidentiality isn’t about avoidance; it’s about design. You don’t say “no” to AI - you say “yes, safely.”
Our AI in Deals framework lays out two critical layers: understanding your tools and putting up the guardrails.
Know your tools
Never use consumer AI models (the free versions of ChatGPT, Claude, Gemini, etc.) for sensitive work. Their terms often allow for prompts and data to be reused in model training, which means any confidential client information, deal data, or internal materials could leak.
Instead, use enterprise-grade AI solutions built for business use. These bring:
Data isolation: Your inputs aren’t used for model training
Encryption: Both in transit and at rest
Compliance: Most are certified under GDPR, ISO 27001, SOC 2, and can run on private or even on-premise clouds
Contractual protection: Clear policies on data retention and deletion
Think of it like the difference between public Wi-Fi and your company VPN: one’s convenient, the other’s secure.
Build the guardrails
Good governance will enable innovation. Establish:
Clear AI policies defining acceptable use, ownership, and responsibility
Mandatory enterprise tooling for any client or confidential data
Role-based access controls so people only see what they need
Controlled sandboxes for testing new AI tools safely
Data minimisation and anonymisation for exploratory use
Supplier due diligence - especially for vendors touching your data rooms or CRM
At a technical level, you can also add protective layers between your prompts and the model. For example, stripping out identifiable details before the request leaves your environment, or using “policy-as-code” automation to enforce permissions and monitor for breaches.
Secure AI is about giving your teams access to tools they trust. Build that trust early, and adoption will follow.
3. Should I use third-party AI/LLM providers or build bespoke AI in-house?
For SMEs, third-party AI platforms (like OpenAI, Anthropic, or Microsoft Copilot) are usually the right choice. Building from scratch requires infrastructure, ML engineering, and expertise that most businesses don’t need or have.
The benefits of third-party AI are clear:
Low upfront cost and predictable seat-based pricing
Faster time to market
Automatic upgrades - you benefit from each new model release
Reduced technical risk and baked-in security
However, off-the-shelf tools have limits. They’re general-purpose, seat-priced, and don’t always integrate cleanly with your data. That’s where bespoke solutions add value - when you have proprietary data, workflows, or intellectual property that offer real differentiation.
Bespoke builds give:
Full data control - you can log usage, filter confidential inputs, and design your own audit trail
Custom integration with CRMs, data warehouses, and existing systems
Usage-based pricing (API calls, not per seat)
The trade-off is that you take responsibility for uptime, compliance, and maintenance.
So, the simple rule of thumb:
If you want a tool, go third-party.
If you want transformation, build bespoke (or hybrid)
The smart middle ground is abstraction - building flexible layers between your apps and the underlying models. That way, if you ever switch providers or fine-tune an open-source model later, your core logic and data remain intact.
We recently did this for a UK private equity firm, building an intelligence layer over their existing Copilot infrastructure so they could interrogate all of their deal notes, CRM data, meetings, and shared documents within one view.
4. How do I know what I’m building won’t go to waste as new developments emerge?
This is a valid concern in a field moving as fast as AI. The answer lies in design thinking and modular architecture.
If your ROI horizon is short - say, three to six months - go ahead and build now. You’ll have captured value before the landscape shifts. But if your ROI depends on a two-year timeline, you’re likely over-scoping. Start smaller. Prototype. Learn fast.
To future-proof what you build:
Use standard connectors and open APIs
Adopt modular (or “componentised”) architecture. Each element - data ingestion, model layer, interface - can be swapped as better tools emerge
Consider Model Context Protocol (MCP) or similar frameworks that allow your internal models to plug into new external ones seamlessly
Design for replacement, not permanence. Like a race car with interchangeable components: improvements shouldn’t require a full rebuild
The point isn’t to build something that lasts forever - it’s to build something that evolves gracefully.
5. How do I stay on top of AI developments?
You can’t track everything. But you can build smart habits that keep you informed and inspired.
Follow two streams of information
Vendors - for the latest features and capabilities
Independent analysts and communities - for comparisons, opinions, and unbiased use-case insights
Seek real-world use cases. Newsletters and bulletins are great, while conversations and events bring AI to life even more. Ask people: “How are you using it?”. Real examples beat press releases every time
Experiment personally. Try new tools on something you care about - whether that’s your workflow, a hobby, or helping you negotiate a price for a new car. Learning through relevance is the fastest route to understanding
Evolve your expertise, not just your tech. Keep deepening your subject-matter knowledge. The more expert you become in your field, the better you’ll spot where AI can add value
Curate, don’t consume. Pick two or three trusted sources (newsletters, analysts, or podcasts) that align with your learning style. For some, bullet points work better than long-form; for others, conversations or conferences spark more ideas
The goal isn’t to chase every trend. It’s to build an informed perspective so you can distinguish signal from noise - and act when it matters.
Final Thoughts
Every decade has its inflection point, the moment when hesitation quietly becomes risk. 2025 is that moment for AI.
The organisations thriving two years from now won’t be the ones that guessed the right model; they’ll be the ones that built cultures, systems, and decision-making processes agile enough to adapt to whatever comes next.
How twisted loop can help your business
Our team has been delivering data and AI solutions for over 30 years.
We know how to start fast, align strategy to business goals, and build momentum with confidence.
We offer a range of AI Strategy and Rapid Prototyping services, with delivery timescales as short as two weeks - so you can choose the balance of speed, rigour, and scope that best fits your business priorities.
Want to rapidly accelerate your AI strategy or test a high-impact use case?
