Vertical Integration

A company only does part of its own work itself. The rest goes to someone else. Pagedeck rents servers rather than running them, hands card payments to a payment processor, and sends help-center translation to an outside agency. Outsourcing is cheaper and faster: the money and months that a server room would consume go into the product instead.

But every outsourced step ties you to someone else's terms. When the processor raises its fee from 2.5 percent of revenue to 3.2 percent, your cost rises by exactly that much. Negotiating leverage runs in proportion to volume, and a company with $150,000 in annual revenue has none against a large processor. Quality behaves the same way. When an outsourced support agent answers a customer, the answer goes out, but the product problem the question revealed never reaches the engineering team. What was handed over was a response; what was lost was the information the conversation carried.

Vertical integration is the decision to take that step back. Process payments yourself, put employees on support, run your own servers. It isn't done because it's cheaper — early on it's usually more expensive. It's done to be able to see what happens in that step and fix it directly, which makes the choice of which step to take back the whole of the decision.

There are two directions, and which one you're moving in gets settled first.

1. Backward integration — making something you used to buy. Moving from rented servers to your own data center, or from purchased components to in-house production. The aim is control over cost and supply.
2. Forward integration — taking back a sales channel or customer touchpoint you had delegated. Moving from distributors to your own sales team, or bringing outsourced support in-house. The aim is seeing what customers actually experience.

How much to take back is a separate question. You can acquire the company that does the work, build it from scratch internally, or take part of it while continuing to use the existing provider. That third option is the common starting point: process 20 percent of payments yourself, confirm the real cost and failure rate, then move the rest. Running both in parallel costs more for a while, and it buys the only figures worth deciding on — your own, rather than a vendor's pricing page read backward.

The verdict comes from three questions. First, does this step determine value the customer can feel? Translation quality does; a few milliseconds on the payment screen doesn't. Second, is there enough volume? Doing it yourself creates fixed cost that needs transaction volume to spread across, and that calculation runs off the cost structure — the breakdown of outgoing money into fixed and variable. A processing fee is variable; in-house processing is fixed, so at low volume integration loses money. Third, can you actually do the work well? Take the step back only when all three answers are yes.

In a software company the decision surfaces in roughly four places: payment processing, customer support, infrastructure, and the sales channel. All four typically start outsourced and come up for review as the company grows, usually in that order, because that is roughly the order in which each one's volume gets large enough to matter.

What separates the yes from the no is whether the step touches competitive advantage — the thing you do better that competitors can't easily copy. How documents get searched is something Pagedeck has to build itself. Card processing isn't: handling payments in-house never becomes a reason a customer picks you. Steps that don't touch the advantage get decided on the cost calculation alone, and when the volume isn't there, they stay outside.

The reverse decision comes from the same place. If a 45-person company keeps two engineers maintaining a payment system, those two aren't building product. Returning them to the product is a focus strategy decision — pouring all resources into one narrow area — and in a small company, integration and focus often sit on opposite sides of the table.

It also separates from the strategy types next to it. Diversification widens the business into new markets; vertical integration takes the adjacent steps of the same business, so no market gets added. It widens the boundary of work you already do, which is why it shows up in a strategy document as a change in what the company operates rather than a change in what it sells.

"Processing fees run 3 percent of revenue. Has anyone worked out what doing it ourselves costs?"

Checking whether integration actually pays on cost.

"Leave translation outsourced and bring support in-house. The information we're losing there is worth too much."

Judging the decision step by step rather than as one policy.

"Two engineers are tied up maintaining the payment system. I want to revisit whether that's right."

Reviewing whether something taken in-house should go back out.

Pagedeck keeps three of those steps outside and moved one in. Servers, payments and help-center translation are external; customer support came in-house three months ago. The reason wasn't cost. Over six months of outsourced support, more than 400 questions accumulated about the same feature, and that fact never reached Mia.

Since the move, the support lead counts questions by category each week and hands the tally to Mia. Response times got slightly slower and payroll went up. Three of the five product improvements shipped last quarter came out of that tally. Grace treats that as a trade worth making and has left the decision in place, with a note to revisit it if the tally stops producing anything.

The most common mistake is expecting the cost to drop. Remove the party taking a cut and the saving looks automatic. But the fee a processor charges contains the savings that company built through scale. The per-transaction cost of a company handling ten million transactions is not the per-transaction cost of you handling ten thousand. On top of that sit the fixed costs integration creates: six months of development, two people on maintenance, and someone awake at night when it breaks. Compare only the fee against those three left out and integration always looks cheaper — and about eight months after the move, the arithmetic says cost went up.

The other mistake is treating integration as irreversible. Sending something back out feels like an admission of failure, especially when the original move was announced internally as a strategic one. But whether integration is right changes with scale. Volume can grow until it becomes right, and strategy can shift until the step no longer matters. Decide in advance never to reverse it and two engineers stay parked outside the product indefinitely. Which is why an integrated step deserves the same three questions once a year.