What’s a piece of conventional wisdom in your industry that you think is completely wrong? What should companies do instead? And if you could only track three metrics to predict success, which would they be?What’s a piece of conventional wisdom in your industry that you think is completely wrong? What should companies do instead? And if you could only track three metrics to predict success, which would they be?
The belief I disagree with most is that AI is a technology project. Companies hand it to IT, approve a pilot, and wait for a demo. The demo often works. Then nothing changes, because no senior leader owned the question of where AI should change the business and where it shouldn’t. Pilots lose their champions and progress fades. The technology rarely fails. The ownership does.
Companies should treat AI as a leadership agenda instead. Pick one repeatable workflow that slows the team down. Tie it to a number the business already tracks, and make a senior leader accountable for that number. Prove value inside the live process, then expand to the next workflow. Each win informs the next one.
If I could track only three metrics to predict success:
- The share of AI work in production rather than in pilot. It separates companies building real capability from companies collecting demos.
- One business number per use case, agreed before the build starts, such as lead volume, deal cycle length, response time, or margin. If nobody can name it at scoping, the project is unlikely to make it past the pilot.
- Client return rate. In a services business, whether this year’s clients come back is the best predictor of next year. Today, 73% of our clients return for long-term partnerships.
What problem did you set out to solve that others were overlooking? Was there a specific moment that convinced you it needed to be solved? And if you were starting from scratch today with no budget or brand, what would your first three moves be?
The problem was a talent gap the world wasn’t seeing. Pakistan had strong engineers. About 60% of my graduating batch went on to start tech companies. But there was no reliable bridge between that talent and clients in the US and other markets. We started tkxel in 2008 to build that bridge and to make “built in Pakistan” mean world-class quality. Five of us put in $500 between us. We had no international connections and no experience winning overseas clients.
The moment came in our first real pitch, to a US health tech company. Their CTO asked how we planned to run a tech company from a country with power outages. It was a fair question. We didn’t argue with it. We went and built, and we have let our delivery answer that doubt ever since.
If I started from scratch today with no budget or brand:
- Pick one narrow problem I understand deeply and become the best at solving it before going wide.
- Build AI into delivery from day one. At tkxel, AI-assisted delivery now runs on about 70% of active projects. It saved an estimated 18,000 engineering hours over 12 months. A new firm can be small and still deliver like a large one.
- Let clients do the selling. Ask every client for a public review that includes a real number. Some of our longest relationships started with a buyer who found us online and chose us based on independent ratings.
What’s the toughest trade-off you’ve had to make between quality, speed, and cost? What did you sacrifice, and why was it the right decision?
Early on, the hardest trade-off was turning work away. In 2018, we were still taking projects as small as $10,000. As our team and client base grew, we raised our minimum to $40,000. We gave up volume and some quick revenue. In return, we could commit stable teams to fewer clients and own their outcomes. Clients pay a services company for quality, and quality is hard to deliver when your best people are spread across too many small projects.
The harder lesson was about speed. After early success, we spun off a gaming studio and a digital agency. Both passed seven figures quickly. Both failed within a year after I handed them to new CEOs. We had chosen speed of expansion over the basics. It taught me that scaling isn’t about doing more. It’s about doing the basics better and staying close to your people.
From your experience, what’s the biggest reason projects fail? What happens when companies avoid that mistake? Have you uncovered any surprising patterns in your client data over the years?
In AI work, the most common failure point isn’t the build. It comes right after the first proof of concept. The demo works, but nobody has tied it to a business outcome, and no senior leader owns it. Executives approve the project and then watch from a distance, so it never reaches the strategic conversations it needs in order to scale.
When companies avoid that mistake, the project has a number from day one and a leader accountable for it. Teams trust it because they see time saved. Leaders back it because they see the metrics move. The first use case then becomes the template for the next one.
Two patterns stand out. The first comes from client work: the relationships that grow are the ones where we own an outcome, not just a list of tasks. One client started with us taking ownership of backlog items inside their own team. Three years later, that work includes a full product rebuild with AI voice and video calling built in.
The second comes from our own house. In AI Unlimited, our internal build program, teams evaluated more than 60 use cases and built 40 working solutions. We deliberately dropped at least 20 because they lacked value, feasibility, or operational fit. Being willing to stop matters as much as knowing what to build.
What kind of results do clients typically see within the first year? And what separates your highest-performing clients from everyone else?
I won’t quote a typical first-year number. Every engagement measures something different, and an average would hide that. What I can share is what clients have measured:
- An agriculture sector organisation made internal document discovery 50% faster with an AI knowledge assistant.
- A security and mobile patrol services company cut missed sales opportunities by 35% and moved from lead to close 25% faster.
- A customer intelligence platform lowered its cloud costs by 40%, about $120,000 a year, and now delivers insights 6x faster.
- A software delivery firm cut its backlog by 30 to 40% after our engineers took ownership of modules inside their team.
What separates our highest-performing clients is behaviour. A senior leader owns the outcome. The metric is agreed before anything is built. And the first win is treated as the start of a program, not the end of a project.
Can you walk us through one of your most challenging client transformations? What was the problem, what changed, and what outcomes did the client achieve?
One of the most complex was a Saudi professional services group with nine service lines. Those lines range from M&A transaction support and IPO readiness to human capital, finance, legal and governance, risk, and compliance.
The units had grown at different speeds. Some were advanced in technology, and some were not. Some shared systems without sharing a plan. The result was siloed systems, disconnected processes and no company-wide approach to data. Leadership wanted to turn its corporate service units into centres of excellence. That needed one strategy, not nine.
The hardest part was getting nine units with different priorities to agree on one direction. Before designing anything, we set business priorities with each unit’s leaders. We also documented success metrics and digital KPIs for every service line.
The client chose us after a competitive evaluation. A seven-person core team, supported by our Chief AI Officer and subject matter experts, worked in three waves so each group of functions got dedicated attention:
- We assessed processes, applications and data in every unit.
- We designed how each unit should operate once digitised, rather than digitising what already existed. Along the way, we identified where automation and AI change the shape of the work.
- We built a portfolio of initiatives prioritised on impact and feasibility, including initiatives that serve several units at once.
- We designed a company-wide data architecture and an enterprise AI architecture.
- We pulled every unit’s AI ideas into one GenAI use-case catalogue, so AI is evaluated once instead of nine times.
This was a strategy and architecture engagement, so the outcomes are alignment and momentum rather than cost savings. All nine service lines now work to one roadmap covering 2026 to 2029. Wave one was signed off in February 2026. In August 2026, the client moved its first initiative, dispute resolution in its Legal unit, into implementation. Their Managing Director put it this way: “Tech is one of the most complex things to work in, but they maintained a level of simplicity, so our non-technical people clearly understood.”
What does your quality control process look like? Is there anything you would never compromise on, even if a client requested it? And how do you manage scope creep while maintaining a strong relationship?
Quality is designed into how we run the work, not checked at the end. On our AI Pods engagements, KPIs are agreed at scoping. The handoff is measured against those KPIs rather than signed off on a feeling. We work in Agile with weekly scrum meetings so every stakeholder sees the same picture. We are certified to ISO 9001 for quality management and ISO 27001 for information security.
There are two things we won’t compromise on, even if a client asks: data security and human oversight of AI. Our published Responsible AI Principles cover privacy, security, fairness, transparency, and human oversight. We hold ourselves to the same standard. On our own content platform, nothing is published without expert approval. If a client asked us to remove a human checkpoint where judgment matters, or to cut security corners to hit a date, we would decline.
For scope creep, we set expectations in the contract. When the scope is clear, we work on a fixed price. When the scope will evolve, we bill for the hours our team actually puts in, so change is planned for rather than argued over. The rest is communication. One client told us what stood out was our ability to adapt quickly to changing priorities while still delivering consistently on time. That is what keeps scope conversations early and calm.
What does your ideal client look like in terms of mindset, readiness, and goals? On the other hand, are there any clients you would choose not to work with, and why?
Our best fit is a growing mid-market business, a funded startup, or a private equity-backed company. Mindset matters more than size. The ideal client has a senior leader who is personally involved and open to changing how work is defined, not just speeding up what already exists.
On readiness, a client doesn’t need everything in place. The usual blockers are limited technical bandwidth, scattered data, unclear use cases, and internal resistance. We can solve the first two. The last two need the client’s leadership: a clear sense of which workflow hurts and a sponsor who will champion the change.
The goals we work best with are specific and owned. More leads, shorter deal cycles, or better margin, with a named person accountable for the number.
We would choose not to work with a client who wants AI for the announcement rather than the outcome. We’d also pass on a client who wants a vendor to take orders rather than a partner who will challenge them. Much of our value comes from telling clients where AI should be used and where it shouldn’t.
What capability are you investing in today that clients aren’t asking for yet but soon will be? And if AI automates a large portion of your industry, what human element will remain indispensable?
The capability we are investing in most is forward-deployed engineering. These are engineers who work inside a client’s business and own AI systems in production, not just in pilots. Today, most clients ask us for a proof of concept. What they will need next is someone accountable for keeping AI running, measured, and improving once it is live. Our partnerships support this. tkxel is a Select Services Partner in Anthropic’s Claude Partner Network and a Select tier member of the OpenAI Partner Network.
The second is the cost of running AI. Once AI is in production, usage becomes a real line item. We are extending our FinOps and cloud cost optimisation practice to AI workflows so clients are ready for that before it shows up in the budget.
If AI automates a large part of our industry, the human element that remains is judgment. The question has shifted from “can we build it” to “should we build it.” The people whose value grows are not the ones who execute faster. They are the domain experts and problem definers who understand the customer and decide what good looks like. That judgment comes from experience, and it can’t be automated.
How has your presence on Goodfirms influenced the way potential clients discover or evaluate your company? And what has your overall experience with Goodfirms been like?
Mid-market buyers do their research before they ever get on a call, and independent reviews are a big part of that research. Our Goodfirms profile holds a 4.9 rating from 44 verified reviews. Reviewers most often mention our communication, technical expertise, and on-time delivery. That lets a prospect judge tkxel in our clients’ words rather than ours. We have also had long-term clients tell us they selected us based on independent ratings and value for cost.
Our experience with Goodfirms has been positive. We first shared our story with Goodfirms in 2019. What we value most is that the reviews are verified and specific. That is what a careful buyer needs, and it keeps us accountable to every client we work with.