Reflections

The right question

From “What are our competitors doing?” to “Where can we create more value?”: five lenses that connect AI to measurable value in the P&L.

Gökhan Mataracı7 min read

What question are we starting with?

Since 2023, conversations about AI in executive meetings have often begun with the same question: “What are our competitors doing?” A more meaningful question is: “Where can we create more value?”

The first question takes us to someone else’s solution. The second takes us to our own business, customers and P&L.

The current picture makes this distinction important. According to PwC’s 2026 CEO Survey, 66% of CEOs in Türkiye say AI has made no difference to their revenue.[1] Globally, only 12% report benefits in both cost and revenue.[2]

The question being asked is not, of course, the only reason for this outcome. But in a period when capital is expensive and demand is weak, we no longer have the luxury of “trying it and seeing what happens.” We need to know from the outset which area of value every AI budget is intended to address.

Someone else’s medicine

We do not go to the doctor and ask which medicine our neighbor is taking. We ask for tests first. In business, however, we often take a good example from the industry and try to adapt it to ourselves.

Industry examples are not useless; they show that something is possible. But symptoms may be shared while root causes are not. Two companies may both be losing customers. At one, the cause may be service quality; at the other, price. The medicine that helped your neighbor may not treat your condition.

Even the best industry example is only a hypothesis.

Where can we create more value?

The entire article rests on this question. At the executive table, I make it more concrete:

Where in our P&L can we create more value by building on our own best performance?

This question does three things. It begins not with a process or the big picture, but with the statement management reviews every month. It does not stop at the symptom; it moves to the cause. And it looks for the benchmark inside the organization, not outside it.

I use five lenses to ask this question. None is specific to a particular industry. But first, a confession: none of them is new either.

Levers we have known for thirty years—and still keep in a drawer

We have known how customer profitability is distributed since the early 1990s.[5] The leakage between list price and pocket price was described in 1992.[7] Economists have measured productivity differences between units of the same company for decades.

So the real question is this: if we have known these things for so long, why are companies still not acting on them?

One important reason was the cost of performing and maintaining these analyses. They were done manually and often remained one-off exercises. A project would calculate customer profitability once, the report would go into a drawer, and the numbers would be outdated within months. Robert Kaplan, one of the architects of activity-based costing, wrote in 2004 that many companies had abandoned the method because of the cost of building and maintaining it.[6] The required information was also scattered across different systems and departments.

AI is not inventing new levers. It is making the levers we already know usable.

A one-off analysis becomes continuous. An annual report moves to the moment of decision. Some of this work does not require AI; sound calculation, integration and automation are enough. AI makes a difference where records are fragmented, unstructured or open to interpretation: emails, contracts, requests, complaints and meeting notes.

01

You may not control the unit price, but you can control the quantity

In Türkiye, many cost items appear untouchable: energy tariffs, the minimum wage, rent and foreign-currency software licenses. Yet every cost item consists of two factors:

Cost = Unit price × Quantity consumed

You often cannot change the price. Quantity, however, can often be managed: energy per unit produced, overtime and waiting time, unused software licenses and empty return journeys. According to Zylo’s 2026 index, only 54% of the SaaS licenses purchased by the organizations it studied were being used.[4]

Why was it missed? Consumption is distributed across dozens of categories and hundreds of locations. Monitoring each one requires human effort. In many organizations, a detailed review is therefore postponed until the budgeting cycle.

Which decision changes? The unit manager intervenes on the day a deviation occurs rather than waiting until month-end. Before approving overtime, the manager reviews the demand forecast; before renewal, the manager cancels an unused license.

AI enters here by forecasting demand and distinguishing the reason for a deviation: is this month’s increase in energy use caused by the production mix, or by equipment beginning to fail?

02

The P&L tells you what you spent, not why you spent it

The chart of accounts shows cost by type or function. But no line says, “the cost of serving this customer.” Robert Kaplan writes that almost every customer-profitability study reveals a similar picture: 15–20% of customers generate all the profit, and sometimes more.[5] Some of the rest break even, while others consume that surplus by operating at a loss.

In other words, many companies may be serving some customers at a loss without seeing it at customer level. The same lens can be applied when an exporter says, “the exchange rate hurt us.” The aggregate figure may say so. At the level of customer, product and market, however, some business may still be highly profitable while some has become loss-making.

Why was it missed? Calculating cost by customer and product required knowing how much time each activity spent on each one. That information was gathered through surveys and interviews and quickly became outdated.

Which decision changes? For a loss-making customer, the sales manager changes the service model: minimum order size, delivery frequency, a fee for special requests or the price.

The cost to serve can be calculated from order, shipment and return records with a well-designed costing model. AI makes a difference by reading fragmented records such as emails, requests and complaints to identify where and why the cost to serve is rising: which customer constantly requests urgent shipments, and whose demands consume the team’s time?

03

Leakage occurs between line items, not within them

According to McKinsey’s analysis of S&P 1500 companies, a 1% improvement in price, with volume held constant, increases operating profit by roughly 8%. In the same analysis, a 1% reduction in variable cost raises profit by about 5.5%, while a 1% increase in sales volume raises it by about 2.5%.[7] Price is therefore the lever that moves profit fastest.

Price is lost on the journey from list price to pocket price: discounts, volume rebates, absorbed freight, returns and payment terms. In Türkiye, payment terms are among the least visible examples. The sales team is rewarded for revenue and offers 90-day terms. The cost of those terms appears not under sales, but under financing expense. Assume an annual funding cost of 40%. Ninety-day terms create a financing burden of roughly 10% of the invoice (simple interest: 40 × 90 / 365 ≈ 9.9%). Yet that cost is not visible when the discount is approved.

Why was it missed? Each element of leakage sits in a different system owned by a different department. Seeing the total required manually combining each transaction across several systems. Even when this was done, it was usually retrospective and based on averages.

Which decision changes? When preparing an offer, the sales representative sees the pocket-price impact of both the discount and the payment term. They can negotiate price instead of terms, or terms instead of price. Leakage is prevented before it occurs rather than reported afterwards.

AI helps by finding additional discounts, free services and exceptions scattered across contracts, emails and special agreements, and bringing them into the calculation.

04

What is easiest to cut is not necessarily what should be cut

When cost pressure arrives, headcount is often the first thing that comes to mind. Personnel expense is visible, and the amount to be addressed is easy to calculate. Waste, by contrast, is distributed across dozens of categories and remains invisible. Managers cut the item they can find.

Research suggests that downsizing based only on headcount may not deliver the expected result. In a study of S&P 500 companies by Cascio and colleagues, companies that reduced employee numbers alone did not significantly outperform their industry averages financially. Those that combined downsizing with asset restructuring achieved better results.[8] In a Bain survey of more than 700 senior executives, 80% said they did not expect the savings to last.[9]

Why was it missed? Finding fragmented waste was far more laborious than counting people. Under pressure, the easiest item to measure came to the fore.

Which decision changes? The cost discussion moves from “Whom should we let go?” to “Which work should we stop doing?”

Process records and reporting are sufficient to measure waiting times and repeated steps. AI can read free-text requests, approval notes and correspondence to reveal where non-value-adding work and approvals sit. It can also take on some routine tasks directly. Here, AI is a tool for redesigning work, not simply reducing headcount.

05

Your best benchmark is your own best day

According to data compiled by Chad Syverson, among plants in the same narrowly defined U.S. manufacturing industry, a plant in the top 10% of productivity produces nearly twice as much output from the same inputs as one in the bottom 10%.[10] A study by Nicholas Bloom and colleagues covering 35,000 U.S. manufacturing plants shows that 40% of the variation in management practices exists between plants within the same company.[11]

There is therefore likely to be a substantial difference between your best branch, team or sales representative and your average. But not all of that difference may come from management style or behavior. Your best salesperson may have a different customer portfolio; your best plant may have different equipment. We therefore look at the difference that remains after accounting for customer, product and operating conditions. The aim is performance that can be repeated under comparable conditions. Investigating this gap is especially valuable because an example of better performance already exists inside your own organization.

Why was it missed? It was possible to see the difference; identifying its cause and spreading it was harder. What the best representative did differently often remained embedded in that person’s experience. Training and manuals could transfer only part of the knowledge.

Which decision changes? When calling the next customer, an average-performing representative can see what the best representative did in a similar situation: which offer, in which sequence and at what time.

Measuring the difference between units is a conventional analytics task. AI contributes by controlling for differences in conditions, searching across dozens of variables for factors that may explain the remaining gap, incorporating meeting notes and correspondence, and turning the best behavior into a recommendation at the moment of decision.

This lens also offers a diagnostic clue:

If you are losing money even on your best day, it may be time to put the strategy on the table as well.

In that case, the market, product mix and business model should be discussed alongside operational improvements.

Why should AI projects start with the P&L?

The five lenses share one point: the value is already in the P&L. What is missing is a mechanism that can capture it every day and in every decision. That mechanism is usually not a report, but a repeated decision: a price, a discount, a payment term, a shift plan or a customer call.

This is precisely where AI creates measurable value. An improvement of a few percentage points in a single decision may appear insignificant. In a decision repeated tens of thousands of times a year, it changes the P&L.

This approach has another advantage: it does not require waiting for perfect infrastructure. Invoice, order, discount and collection records are data almost every organization already maintains. The initial diagnosis can often be completed within weeks, and the first intervention can follow soon after.

Closing

At a time when confidence in AI is eroding, we do not need another promise. We need to begin with the right question: not “What are our competitors doing?” but “Where can we create more value?”

How large is the gap between your best team and your average? And who has measured it so far?

Sources

  1. PwC Türkiye, 29th Annual Global CEO Survey — Türkiye results, executive summary (2026)↩
  2. PwC, 2026 Global CEO Survey press release (January 2026)↩
  3. Bloomberg HT, Şirketler için susuz yaz [A Dry Summer for Companies] (June 2026)
  4. Zylo, How Much Is Wasted on SaaS Spend? (using 2026 SaaS Management Index data)↩
  5. Robert S. Kaplan, A Balanced Scorecard Approach to Measure Customer Profitability, HBS Working Knowledge (August 2005)↩↩
  6. Robert S. Kaplan and Steven R. Anderson, Time-Driven Activity-Based Costing, Harvard Business Review (November 2004)↩
  7. Marn, Roegner and Zawada, The power of pricing, McKinsey Quarterly (2003); Marn and Rosiello, Managing Price, Gaining Profit, Harvard Business Review (1992)↩↩
  8. Wayne F. Cascio, Clifford E. Young and James R. Morris, Financial Consequences of Employment-Change Decisions in Major U.S. Corporations, Academy of Management Journal 40(5) (1997)↩
  9. Bain & Company, Getting a competitive edge through cost transformation (October 2015)↩
  10. Chad Syverson, What Determines Productivity?, Journal of Economic Literature (2011)↩
  11. Bloom, Brynjolfsson, Foster, Jarmin, Patnaik, Saporta-Eksten and Van Reenen, What Drives Differences in Management Practices?, American Economic Review (2019)↩