Quick answer
Most "AI strategy" pitches to Canadian mid-market oil and gas operators are framed wrong - looking for transformation when the real work is workflow-level deployment. The Augmentation Edge eBook walks the framework: three workflows where AI consistently delivers measurable hours-saved value (JIB reconciliation, vendor contract review, CVE triage), three where it doesn't yet (field operations, production forecasting, autonomous incident response), why CFO ownership is the strongest pilot success predictor, and the 90-day pilot pattern that scales only what proves out.
Inside this guide.
Credentials, affiliations & memberships.
James's perspective is shaped not only by 25 years of operator experience but by active participation in the global communities setting the agenda for technology leadership, AI policy, and digital sovereignty - from Calgary to the United Nations in Bangkok.
The layoff that cost more than the salaries.
In 2023, Klarna's CEO told the world that AI could already do all of the jobs that humans do. The company stopped hiring. It cut roughly 700 customer service positions. It reported "$10 million" in savings and was hailed as the most aggressive AI-first transformation in fintech.
By early 2026, Klarna was quietly rehiring. Customer satisfaction had cratered. Edge cases overwhelmed an AI trained on routine queries. The CEO admitted, in print: "We focused too much on efficiency and cost. The result was lower quality, and that's not sustainable."
Klarna is not the cautionary tale because it tried AI. It is the cautionary tale because of how it tried AI.
Gartner published a follow-up in February 2026 predicting that half of all companies that cut headcount specifically for AI will rehire by 2027. IBM, after replacing most of its HR division with an AI chatbot, ended up reinvesting in engineering, strategy, and client engagement to compensate for what the bot couldn't do.
Meanwhile - and this is the part that should make every executive pause - MIT's NANDA Initiative studied 300 enterprise AI deployments and found that 95% delivered zero measurable P&L impact. Not 50%. Not even 70%. Ninety-five percent.
So what does the 5% know that the 95% don't? It isn't a better model. It isn't a bigger budget. They use the same tools everyone else does.
This eBook is about that misapplication - and the unglamorous, deeply un-trendy work it takes to do it right. It is written for the owner or executive of a 10 to 100-person company who has been told, repeatedly, that AI will let them "do more with less." That promise is usually a euphemism for layoffs. It is almost always wrong. And it is the single fastest way to land yourself in the 95% that wasted the money.
There is a better way. It involves the same staff you already have. It is just harder to fit on a slide.
You're probably approaching this wrong.
Let's start with the boardroom conversation that has been happening, in some form, in roughly every mid-market company on the continent for the past 24 months.
It is asked seriously. By smart people. In good faith. And it is the wrong question - not because Karen is irreplaceable in some sentimental sense, but because the question reveals a misunderstanding of what AI is actually good at, what Karen actually does, and what the math of replacement looks like once you carry it past the announcement.
The replacement fantasy vs. the real world.
The replacement fantasy goes like this: AI tools cost less than a salary. AI tools work 24/7. AI tools don't take vacation, don't have kids, and don't quit. Therefore, AI tools = salaries minus problems = profit.
Every part of that equation is wrong in a way that only becomes obvious about nine months after you act on it.
- Headcount cuts treated as the primary ROI metric.
- Institutional knowledge walks out with the severance package.
- AI confidently fails on edge cases. Customers notice immediately.
- Quiet rehiring at month 9 - often more expensive than the original "savings."
- Engagement tanks. Hiring becomes harder. Boardroom narrative collapses.
- Time saved per role, captured as growth capacity.
- Karen still knows where the bodies are buried - and is now twice as fast.
- Humans catch the edge cases AI flubs. Quality goes up, not down.
- Compounding productivity year over year. No expensive re-do.
- Engagement and retention improve. You become a place people want to work.
What actually happens in the nine months after
- Month 1–2: Announcement, layoffs, celebratory press release. Stock or valuation ticks up. Executives feel decisive.
- Month 3–4: AI tool deployed. Routine queries handled. Metrics look great. Someone makes a deck.
- Month 5–6: First edge cases hit. AI handles them poorly. Customers notice. Internal staff start working around the AI.
- Month 7–8: Senior people start covering for the AI. Their actual work suffers. Quality drops. Morale drops harder.
- Month 9+: Quiet rehiring begins. Job titles change. The savings line is revised. Someone who knows where the bodies are buried writes a confidential memo.
The 95% that fail (and why).
In July 2025, MIT's NANDA Initiative published The GenAI Divide: State of AI in Business 2025. The report was based on 52 executive interviews, surveys of 153 leaders, and analysis of 300 public AI deployments. It examined an estimated $30–40 billion in enterprise AI spending.
The headline finding became one of the most-quoted statistics of the year. The chart below tells the whole story.
The natural reaction is to assume that the 5% have something the others don't. A better model. A bigger budget. A unicorn data scientist. They don't. They use the same Claude, the same Copilot, the same ChatGPT Enterprise, the same Gemini as everyone else.
What separates them is a set of decisions that are almost embarrassingly mundane. MIT's researchers, alongside follow-up work by Gartner, S&P Global, and Forrester, converged on four structural causes of failure. Each one is a leadership decision, not a technology problem.
The four causes of failure.
If you read those four causes carefully, you'll notice none of them is about the AI. They are all about how organizations introduce, govern, and manage new technology. Which brings us to the next observation, which most AI vendors will not enjoy.
AI is just another technology.
A confession from someone who has been in IT leadership since the late 1990s: every wave of technology has been introduced by people who insisted, with absolute conviction, that this one was different.
The internet was going to eliminate retail. (It did not.) Email was going to eliminate the office. (Briefly true in 2020, then it wasn't.) The cloud was going to eliminate the IT department. (Ask any CIO how that turned out.) Mobile was going to eliminate the desktop. (Have a look at your finance team.) Blockchain was going to eliminate the middleman in every industry. (Let's not talk about that.)
Each of those technologies was genuinely transformative. None of them did what their loudest evangelists predicted. They all did something more interesting, more uneven, and more dependent on the boring work of integration than anyone wanted to admit at the time.
Reality: e-commerce + physical retail.
Reality: bigger IT departments.
Reality: both, forever.
Reality: in progress. See this eBook.
AI is in the same lineage.
Generative AI is not magic. It is statistical pattern matching at industrial scale, attached to a remarkable interface. It is genuinely transformative for some tasks, surprisingly unreliable for others, and the line between the two is not always intuitive.
Harvard Business School and Boston Consulting Group, in a study of 758 consultants, coined the term "the jagged technological frontier" to describe this - AI improves performance on tasks within its capability boundary, and degrades performance on tasks just outside it. The frontier is jagged, not smooth. It moves. And the only way to map it for your business is to actually use the tools on your actual work.
This is, again, not new. Every previous wave of technology had a jagged frontier. Spreadsheets were transformative for budgets and disastrous for narrative reasoning. Email was transformative for asynchronous coordination and disastrous for nuanced negotiation. The wise organizations learned the frontier and built workflows around it. The unwise ones tried to use the new tool for everything, declared it a failure, and went back to the old way.
Why this perspective matters
If AI is just another technology - extraordinary, but in the same lineage as everything that came before - then the playbook for adopting it is also not new. It is the same playbook used for ERP rollouts, for cloud migrations, for the introduction of CRM in the 1990s. Identify the real problems. Pilot small. Train the users. Measure honestly. Adjust. Scale what works. Decommission what doesn't.
What is new is the speed at which the tools improve. Models that were state of the art eighteen months ago are now charmingly outdated. This argues against expensive, multi-year custom builds locked to a specific vendor's 2024 capability. It argues for staying flexible, building thin layers on top of the strongest available tools, and assuming you will swap underlying models several times in the next five years.
The augmentation math.
The replacement narrative is loud because it is simple. The augmentation narrative is quieter because it requires arithmetic. The arithmetic is worth doing.
Over the past three years, a series of rigorous field experiments - not vendor whitepapers, peer-reviewed academic work from MIT, Stanford, Harvard, NBER, and the OECD - has produced a remarkably consistent picture of what AI does to the productivity of existing workers when introduced well.
Three patterns inside those numbers are worth pausing on, because they reverse the assumptions that drive most replacement decisions.
Pattern 1 - The gains are largest for less experienced workers
The Stanford study of 5,179 customer support agents - still the largest controlled deployment of AI tooling ever published - found that agents with two months of experience and AI assistance performed at the level of agents with six months of experience without it. AI compressed the learning curve. The implication is uncomfortable for the replacement narrative: AI doesn't make the junior people redundant. It makes them more valuable, faster.
Pattern 2 - AI amplifies skill rather than substituting for it
In the BCG study, consultants using GPT-4 on tasks within the AI's capability boundary outperformed their unassisted peers by a wide margin. On tasks just outside the AI's capability - the jagged frontier - assisted consultants actually performed worse, because the AI confidently produced plausible-looking nonsense and the consultants didn't catch it. The skill that mattered most was knowing when to trust the AI and when to override it. That skill comes from experience. It is the senior people's edge.
Pattern 3 - The compounding effect is bigger than any single number
None of those productivity gains is, by itself, world-changing. A 14% improvement in customer support is a nice quarterly win. A 25% improvement in consulting throughput is a real margin improvement. But the compounding effect across functions is where the actual transformation lives. A finance team 20% more efficient. A sales team 15% faster at proposals. A marketing team 30% faster at content. An IT team 40% faster at routine tickets. Layer those gains across an organization of 10 to 100 people, and you have absorbed the workload growth that would have required two or three additional hires - and your existing team is doing more interesting work.
That is the math. It is unglamorous. It cannot be announced in a press release. It does not generate a stock bump. But it compounds - year after year - into a genuine competitive edge. And critically, it builds on the institutional knowledge already in the building. The Karen problem, solved.
So if augmentation is the answer, why does it feel harder to execute than replacement? Because the hard part - the part everyone skips - is the next chapter.
Change management: the part everyone skips.
Most discussions of AI implementation are dominated by technologists discussing technology. This is roughly as useful as a discussion of marriage dominated by jewelers discussing rings.
The technology is the easy part. Anyone can buy a Copilot license. The hard part is the human system that has to absorb the change, develop new habits, build trust in the outputs, redesign workflows around the tool, and continue performing while doing all of it. This is not new work. It is the same change management discipline that has existed since organizations started introducing new technology to humans, which is to say, since organizations existed.
Notice what is not in those numbers. "Picked the wrong vendor." "Bought the wrong model." "Didn't have enough data scientists." The technology variables matter, but they are not where the failures happen. The failures happen in the soft, unsexy work that gets the least executive attention and the smallest budget line.
Why AI makes this harder, not easier.
Most prior technology rollouts had a clear before and after. The old system was switched off, the new system was switched on, and people adapted to a new fixed reality. AI is not like that. AI is a moving target. The model improves every few months. The tasks it can handle today are different from the tasks it could handle six months ago. The right prompts evolve. The right workflows evolve. The right balance of automation and human review evolves.
This means AI adoption is not a project with a defined end. It is a continuous capability that the organization has to build. The companies that treat it as a one-time rollout - train everyone once, declare victory, move on - find themselves eighteen months later with stale habits, outdated prompts, and a workforce that is using the AI worse than they used it at launch.
The three roles you have to actually staff.
In every successful AI adoption I have led or advised, three roles end up doing the heavy lifting. They are rarely on the announcement slide. They are always on the success.
A practical framework: the Delta Playbook.
The core idea of this eBook in one paragraph: take the people you already have, make them measurably more productive with AI, capture the delta as real business outcomes rather than headcount cuts, and reinvest some of that delta into training those same people to take on higher-value work. Repeat. Compound. Win quietly.
That is the Delta Playbook. It works for a 10-person professional services firm. It works for a 75-person energy services company. It works for any organization where the bottleneck on growth is the capacity of existing staff to handle more interesting work.
The five-step delta workflow.
Find the actual time sinks
Two weeks of 1:1s. Not surveys. The answers will surprise you.
Match sinks to AI capability
Honest mapping. Know the jagged frontier before you commit.
Pilot small, with the willing
Volunteers, never conscripts. Real work. Four to six weeks.
Honest pilot review
Not a celebration. Not a kill meeting. A real review.
Scale carefully, train continuously
Build capability, not one-off rollouts. Prompt libraries matter.
Step 1 - Find the actual time sinks (not the imagined ones)
Before you buy a single license, spend two weeks asking your team - in one-on-ones, not surveys - where their time goes that they wish it didn't. The answers will surprise you. The CFO who you assumed needed help with analysis actually spends most of her week chasing down expense reports. The senior engineer you assumed needed help with code is actually losing four hours a week to status updates. The salesperson you assumed needed help with prospecting is actually drowning in CRM data entry.
This list is gold. It is also almost never what the executives think it is. The work AI is best at, in 2026, is exactly this layer - the time tax that sits between your people and the work you actually hired them to do.
Step 2 - Match real time sinks to real AI capability
AI is genuinely good at: drafting first versions of documents, summarizing long inputs, extracting structured data from unstructured sources, answering questions about a known corpus of documents, writing routine code, transcribing and summarizing meetings, generating variations of marketing copy, and triaging incoming requests by category and priority.
AI is genuinely bad at: anything requiring real-time information it wasn't trained on (without retrieval), tasks with strict accuracy requirements where confident-sounding wrong answers are worse than no answer, multi-step reasoning across many systems without specialized agent infrastructure, and anything that depends on tacit knowledge of your specific organization that isn't written down anywhere.
Step 3 - Pilot small, with the willing
Pick two or three use cases. Pick them from the list of real time sinks, not from a vendor's case study. Recruit volunteers - never conscripts - from the affected team. Give them training, a sandbox, a clear measurement plan, and four to six weeks to actually use the tool on real work.
Step 4 - Honest pilot review
After the pilot, hold a structured review. Not a celebration. Not a kill meeting. A genuine review. What worked? What didn't? Where did the AI surprise us positively? Where did it disappoint us? Where did the workflow need redesigning, not just augmenting? What governance and data-handling rules did we discover we needed?
Step 5 - Scale carefully, train continuously
If the pilot worked, scale it - but slowly enough that you can actually train the next group of users properly. If it didn't, kill it cleanly and try the next use case. Either way, build training, prompt libraries, and governance materials as you go. The companies that win at AI are the ones building internal capability, not the ones doing one-time rollouts.
Quick wins by function.
Theory is useful. Specific examples are more useful. Below are use cases that have produced measurable, repeatable productivity gains across the kind of companies this eBook is written for - 10 to 100 people, real businesses, real budgets, real customers. None of these require building anything custom. All of them can be running inside 30 days with off-the-shelf tools and a sensible governance policy.
Governance & data safety: the boring part that saves the company.
Every executive who has been in a boardroom during a data breach knows the look. It is the look of a person who is realizing, in real time, that all of the cost savings they generated by skipping the boring governance work are about to be repaid, with interest, in the next four to six weeks.
AI is a data exposure surface most companies have never had before. The free consumer versions of the major models, by default, may use your inputs to improve future versions of the model. The terms vary. The defaults change. And every employee who pastes a contract, a customer list, a financial model, or a confidential strategy memo into a chat window is making a data-handling decision on behalf of your company - usually without knowing they are doing it.
The four governance decisions.
Measuring what matters.
The vendors will tell you to measure adoption rates, license utilization, and the number of prompts run per user. These are not the metrics that matter. These are the metrics that justify the next renewal.
The metrics that matter are the business outcomes - and there are surprisingly few of them, but they need to be tracked honestly from day one. If you can't measure them, you don't know whether the AI is working, and you definitely can't have the conversation with the board about expanding the program.
The five metrics that actually matter.
- Time saved per role, per week - measured by asking the actual users, not by inferring from license usage. Calibrated quarterly. This is the foundation of your delta calculation.
- Quality of output for AI-assisted work - error rates, rework rates, customer satisfaction scores on AI-assisted vs. non-AI-assisted work. If quality drops, you are not winning, regardless of what the time data says.
- Capacity absorbed without new hires - when business volume grows, how much of that growth was handled by existing staff with AI rather than by adding headcount? Most credible CFO-friendly metric of the bunch.
- Employee engagement and retention in AI-impacted roles - are people more engaged or less? Are they staying or leaving? The replacement playbook reliably tanks engagement; the augmentation playbook reliably improves it.
- Shadow AI prevalence - what percentage of AI use is on approved tools with approved data, versus personal accounts with company data? You want this trending toward 100% on the approved side.
What not to measure (or measure carefully).
Be cautious about measuring headcount reduction as a success metric. It is the metric that drives organizations into the 95% that fail. It also creates incentives that are diametrically opposed to the change management work that actually produces the gains. The moment your team believes that AI success means their colleague gets laid off, adoption stops being voluntary and becomes adversarial.
Be equally cautious about measuring "hours of work eliminated." The honest answer is usually "hours of low-value work eliminated, replaced with hours of higher-value work that wasn't getting done before." That is the win. Don't accidentally tell your team the goal is fewer hours; tell them the goal is better hours.
The 90-Day Plan.
A specific, executable 90-day plan for a 10 to 100-person company that wants to start using AI well, without firing anyone, and with the explicit goal of capturing the delta as growth and quality rather than as cost cuts. Adapt to your context. Don't skip the unsexy parts.
Days 1–30 · Foundation
- Week 1: Executive sponsor commits in writing. Tone is set publicly. The CEO begins using the tools personally and talking about it.
- Week 1–2: Identify the internal champion (the operator everyone goes to). Recruit them. Give them air cover.
- Week 2: Procure business or enterprise tier licenses on one chosen primary platform. Stop the free-tier shadow AI before you start anything else.
- Week 2–3: Write the AI usage policy. Two pages, plain English. Approved tools, approved tiers, data rules, who to ask if unsure. Communicate it.
- Week 3–4: Run the time-sink interviews across the company. Identify the top 10–15 candidates for AI assistance. Pick two or three for pilot.
Days 31–60 · Pilot
- Week 5: Train the pilot users. Real training - half a day minimum, hands-on practice on actual work, not a recorded webinar.
- Week 5–9: Run the pilots. Track time savings, quality outcomes, qualitative team experience. Adjust weekly. Capture prompts that work.
- Week 8: Mid-pilot honest check-in. Kill what isn't working. Double down on what is.
Days 61–90 · Scale & Train
- Week 10: Structured pilot review. What worked, what didn't, what surprised us, what governance gaps did we find.
- Week 11: Build internal prompt libraries from what worked in pilot. Document the redesigned workflows.
- Week 11–12: Roll out training to the next wave of users. Same depth as the pilot training - half day minimum, hands-on, on real work.
- Week 12: First measurement report. Where are we on the five metrics? What is the delta looking like? What is the plan for the next 90 days?
One more wave.
Twenty-five years in this field, across six continents and most of the energy, defense, mining, and manufacturing sectors, the pattern is now familiar enough that it borders on tedious. A new technology arrives. The loudest voices predict it will eliminate jobs, eliminate industries, eliminate the need for thinking itself. The companies that panic and act on those predictions tend to regret it. The companies that engage seriously, augment carefully, manage the change like the human work it is, and capture the delta as growth rather than as cuts - those are the companies still standing two decades later, doing things their competitors literally cannot do.
AI is the latest version of this. It is not the last. There will be other waves - agents that take multi-step actions, models that reason across modalities, AI integrated so deeply into the tools we already use that we stop noticing it. The companies that build flexible, augmentation-based adoption habits now will absorb those next waves without breaking stride. The companies that bet everything on replacement projects will have to start over, again, with smaller teams, less institutional knowledge, and even louder vendor noise.
If you take one thing from this eBook, take this: your existing staff are not the bottleneck on your AI transformation. They are the asset. The institutional knowledge in their heads is the moat. The judgment they exercise on hard problems is the differentiator. AI does not replace that. AI multiplies it - when introduced with care.
The Klarna CEO eventually said something honest, which is rare enough at that level that it is worth quoting in full. "From a brand perspective, a company perspective, I just think it's so critical that you are clear to your customer that there will always be a human if you want."
Yes. Also: from an operations perspective, a quality perspective, a culture perspective, a retention perspective, a risk perspective, and a long-term competitive perspective.
There will always be a human. The companies that win the next decade will be the ones who made those humans better, not the ones who tried to make them disappear.
Vencer Group
Vencer Group is a Calgary-based managed IT, M&A technology, and digital transformation partner, founded in 2006. Nineteen years in business. Thirty-plus M&A transactions delivered. Zero data breaches across eleven years of managed security operations. Delivery across four continents - with live infrastructure under management right now in Bangkok and Jakarta, and sister NOC/SOC operations running 24/7 out of Canada and Singapore.
Most regional MSPs sell hours. Vencer delivers outcomes - the kind operators, CFOs, and boards all need. Three engagement models (Bundled, Co-Managed, and Fractional), three core tiers plus Field add-on, six delivery areas, and a single team that actually knows your industry - energy, M&A advisory, healthcare and pharmacy, professional services, and international operations.
Where AI fits into all this.
Vencer is one of the few Calgary IT partners that has designed, built, and deployed an AI Workforce Transformation Platform in production with a long-term client - not a deck, not a pilot, a live deployment. The AI Readiness Workshop that supports this work runs in four modules over a half day: the AI landscape and what is real, data safety and platform-by-platform risk, use case mapping and quick wins for your specific industry, and a 90-day roadmap with named owners and a budget. Every decision-maker in the room. No hype.
For companies ready to go further, Vencer offers a JV / co-investment model where Vencer builds AI capability alongside you, you own the outcome, and we are aligned on whether it actually works.
Calgary, AB T2P 3J4
Operator opinion built from field work. Not legal, regulatory, or certified security advice. Every organization carries different variables. Use this as a thinking framework, not a compliance checklist.
→ Book the 30-min review