Chapter 16 · Part 5: Growth

Scaling the Product and the Company

“Organizations which design systems are constrained to produce designs which are copies of the communication structures of these organizations.”

— Melvin Conway

Twelve customers, then sixty

When Maya joined Brindlepath, it had twelve customers. Three years later, it had sixty. Almost nothing that worked at twelve worked at sixty without change.

At twelve customers, each implementation was a small craft project: an engineer spent weeks on each customer’s record integration, and the product had grown dozens of customer-specific settings, some of which nobody remembered the purpose of. At sixty, that approach would have needed a hundred engineers just for implementations. At twelve, Maya knew every customer’s chief nursing officer by name. At sixty, she could not. At twelve, two product teams coordinated over lunch. At sixty, seven teams needed a planning rhythm, shared standards and clear ownership, or they would trip over each other every week.

Scaling is not just doing more of the same. It is changing how the product and the company work, so that growth makes things better instead of breaking them.

Configure, don’t customize

The single most important product decision in scaling business software is how you handle the differences between customers. Every customer believes its workflow is unique. Some differences are real; many are habits. How you respond shapes your margins, your speed and your product’s quality.

At one end, every customer uses the same product in the same way. At the other, every customer has its own code. Healthy products sit near the left: one product, shared by all, with configuration, settings that customers or your team can change without engineering work, for the differences that genuinely matter, and extension points, such as open interfaces, where customers or partners can build what is unique to them.

Customization, building special code for one customer, feels like good service. It wins the deal in front of you. It also creates code that must be maintained, tested, secured and upgraded for that one customer forever, which is why Chapter 6’s three smallest customers were losing money. Brindlepath adopted a simple rule: a request for something customer-specific is first examined as a potential configuration available to everyone; if it cannot be, it goes to a partner or a services engagement, not into the core product.

Scaling the technology

Product leaders do not need to design architectures, but they do need to understand the trade-offs, because scaling problems show up as product problems: slow screens, outages, failed integrations, delayed features.

Multi-tenancy. Most SaaS products serve many customers from shared infrastructure, with each customer’s data kept strictly separate. It is far more efficient than running separate copies, and it requires careful design to make sure no customer can ever see another’s data.

Standard integrations. Replace one-off integrations with a standard connector, configured per customer. Brindlepath’s record connector, from Chapter 7, cut the time to integrate a new hospital from months to weeks.

Reliability targets. As customers depend on the product, reliability becomes a feature. Many teams set service level objectives (SLOs), explicit targets for availability and speed, and track how much of their allowed unreliability, their error budget, they have used. When the budget runs low, the team shifts effort from features to reliability. This turns an argument into a rule.

Observability. At scale, you cannot rely on customers to tell you when something is wrong. Monitoring, logging and alerting should show problems before customers notice them, especially for integrations, where a silent failure can mean patients missing from the queue.

Performance and cost. As usage grows, so do infrastructure and AI costs. Track cost per customer, as Chapter 6 recommended, and treat sudden increases as bugs.

Balancing quality with growth

Growth puts pressure on quality from every direction: more customers, more features, more engineers, more code. Without deliberate effort, quality erodes slowly and then all at once.

A few practices protect it:

  • Protect the capacity allocation from Chapter 5. Reliability, security and technical debt get their share every quarter, especially when growth is fastest.
  • Automate testing and release, so quality checks scale with the team rather than depending on heroics.
  • Use feature flags and staged rollouts to limit the impact of any single change.
  • Watch quality measures alongside growth measures: incident counts, error budgets, support tickets per customer, time to resolve.
  • Do blameless reviews after every significant incident, and fix the system, not just the symptom.

Compliance at scale

Compliance obligations grow with the business. New customers bring new security reviews. New states and countries bring new privacy laws. Larger customers ask for independent certifications. AI features bring new regulations.

Scale compliance the way you scale everything else: make it systematic.

  • Invest in certifications before you need them. A SOC 2 report or HITRUST certification takes months. Start before the big deal that requires it.
  • Build compliance into the platform. Audit logging, access controls, data retention and regional data storage built once, used everywhere.
  • Maintain the trust package from Chapter 11 as a living product.
  • Track regulatory change. Assign someone to watch the regulations that affect you and translate changes into product requirements early.

Expanding market reach

Growth eventually requires reaching beyond the first market. There are a few classic paths, each with its own risks.

Deeper in the same customers. More users, sites and modules within existing customers. Usually the cheapest growth, built on the customer success work in Chapter 14.

More customers in the same segment. Repeating what works with similar customers. Depends on a repeatable sales and implementation process.

Adjacent segments. New types of customers with similar problems. For Brindlepath, that meant large physician groups and health plans, which also needed to coordinate care after hospital discharges. Adjacent segments often need different buyers, messages and features, so treat each as a new product-market fit question, using the discovery methods in Chapters 2 and 3.

New geographies. New countries bring new regulations, languages, payment systems and buying habits. In healthcare, they also bring entirely different health systems. International expansion is rarely a matter of translation.

Expand one direction at a time, and only when the core is strong. A company that expands while its core product is still struggling usually ends up with two struggling products.

Scaling the organization

Melvin Conway’s observation at the top of this chapter has become known as Conway’s law: the structure of a system tends to mirror the communication structure of the organization that builds it. If three teams must coordinate to change one workflow, the product will show the seams.

The practical lesson is to design teams around the product you want. Brindlepath reorganized at about forty engineers from teams organized by technology layer to teams organized around customer journeys: the navigator’s day, discharge and integration, outcomes and reporting, and the patient’s experience, with a platform team providing shared services. Each journey team could deliver most changes on its own.

As the organization grows, the product leader’s job changes too:

  • From doing to enabling. Less time on individual features, more on strategy, hiring, coaching and the system that teams work within.
  • From conversations to written communication. Strategy pages, decision logs and product briefs replace hallway agreements.
  • From one roadmap to many. Each team owns its roadmap within a shared strategy and portfolio view.
  • From knowing every customer to knowing the patterns. Customer advisory boards, research programs and data replace personal relationships with every account, though product leaders should still talk to customers every week.

Brindlepath in practice: the configuration rule

When Brindlepath adopted its configuration rule, the product had accumulated 47 customer-specific code paths, settings and integrations over its first years. Dev Raman’s team estimated that maintaining them consumed close to a quarter of engineering capacity, mostly in testing and upgrades.

The team sorted them into three groups. Nineteen could become configuration available to every customer: for example, a customer-specific risk threshold became a setting any hospital could adjust, with sensible defaults. Fourteen belonged with partners or services: custom reports, for instance, moved to the outcomes module’s export and a reporting partner. And fourteen could simply be retired, because the customers who had asked for them no longer used them, or had never used them at all.

The migration took most of a year, done customer by customer, with notice and support. Some conversations were difficult; one customer had built a workflow around a special integration and needed months of help to move. But by the end, the share of engineering capacity spent on customer-specific maintenance had fallen to a small fraction of its former level, and the time to onboard a new customer had dropped sharply.

The rule also changed sales conversations. Instead of promising custom work, sellers could show prospects the configuration options available to every customer, the partner ecosystem for specialized needs, and the product’s open interfaces. Prospects who needed something truly unique were pointed to services or partners. A few deals were lost. Many more were won faster, and every customer benefited from a product that improved for all of them at once.

Common mistakes

Saying yes to custom work to win deals. Each custom request looks small. Together they become the maintenance burden that slows every release and erodes your margins.

Scaling a product that has not found fit. More sales and marketing on top of weak retention accelerates churn, not growth.

Letting reliability become an afterthought. As customers depend on you, outages cost more. Service level objectives and error budgets turn reliability into a managed commitment.

Expanding in every direction at once. New segments, new geographies and new products together stretch a growing company past breaking point.

Ignoring Conway’s law. Teams organized by technology layer produce products whose seams show to customers.

Keeping early-stage habits too long. Practices that worked with two teams, such as decisions in the hallway and a roadmap in one person’s head, break quietly at seven.

Try this

Make a list of every customer-specific setting, integration and code path in your product. For each, ask whether it could become a configuration available to every customer, whether it belongs with a partner, or whether it should be retired. Estimate the engineering time each one costs per year. Then pick the three most expensive and put a plan for each on the roadmap. Most teams find that a handful of special cases consume a surprising share of their capacity.

How AI changes this

AI changes the economics of scale. AI tools can automate parts of implementation, support, testing and documentation, reducing the people needed per customer. Companies can serve more customers with smaller teams than ever before. That is an opportunity to grow efficiently, but only if the core product is standardized enough for automation to work. AI does not rescue a product built on customization.

AI costs scale with usage. Unlike most software costs, AI inference costs grow with every request. At scale, small inefficiencies become large bills. Engineering for AI cost, by using smaller models where they are good enough, caching results and limiting unnecessary calls, becomes a real discipline.

AI governance must scale too. When many teams build AI features, they need shared standards: approved models and providers, evaluation requirements, safety reviews and monitoring. Chapter 17 describes the practices; scaling means making them part of the platform rather than leaving each team to invent its own.

AI can help leaders see the whole system. Summaries of incidents, customer feedback, delivery measures and roadmap progress across many teams help a product leader spot patterns that would otherwise take weeks to assemble. The judgment about what to do remains human.

Chapter checklist