Decisions

The calls that didn't have an obvious right answer

The calls that didn't have an obvious answer. What got picked, what got refused, and what came back, with the numbers attached. All eight calls in one place; three are highlighted on the consulting home.

What transfers: the reasoning, more than the answer. Each call below names what got refused as clearly as what got chosen, because that is usually the part that carries to a different company.

Eight technology decisions with the numbers attached: what the situation was, what got decided, what got refused, and what it returned. From my executive work and NLT Labs' own builds.

Prior executive role · post-acquisition platform strategy

What had to merge, and what didn't

Situation: Four acquisitions brought overlapping enterprise platforms under one roof, and roughly seventy managers and developers with them. The assumed answer was one merged product.
The call: Treat “should we consolidate” as three questions, not one. Identity, authentication, security and infrastructure became common. The domain engines stayed put: different teams, genuinely different processes, and a rewrite no customer would see. Between them, published APIs and agreed data flows, not a shared codebase.
Said no to: The all-or-nothing framing, in both directions. A full merge would have rewritten engines that already worked. Fully separate stacks would have meant two of every control plane.
Returned: Year-one scope narrowed to what had to be common, each product line kept its roadmap, and consolidation moved to later years, when a case existed.

What transfers“Consolidate” is three questions, not one: what must be common, what stays local, and what only needs a contract between them.

Prior executive role · platform standardisation

One platform, many companies

Situation: One multi-tenant platform and a long queue of companies arriving on it, each with its own names for the same things.
The call: One page of definitions per core record, one named owner per area, a small group with the final word. Core fields were fixed: a location has a name, a site and an address, and nobody could repoint them. Outside that core, companies extended freely. Tested against real data, then frozen, with a change log.
Said no to: Agreeing everything up front, and letting one company redefine a core field to suit itself. A location that means something different in every tenant breaks reporting, integrations and conversation at once.
Returned: The platform carried 110+ enterprise retail and restaurant brands across 100+ countries and 16 languages on one set of definitions.

What transfersStandardize the core records and the shared names; let companies extend past that freely. The test is whether one tenant's change can alter what a shared field means to everyone else.

Prior executive role · PE-backed healthcare SaaS

Infrastructure cost down, with uptime going up

Situation: A healthcare platform scaling fast on infrastructure that cost more every quarter, serving users who can't absorb a maintenance window.
The call: Move core platform workloads to AWS, treating zero downtime as the constraint rather than the goal, and negotiate the migration itself. Providers fund migrations they are winning, especially when you decommission what you leave. Have that conversation first, not after the invoices.
Said no to: Savings that come out of the reliability budget.
Returned: Annual infrastructure spend came down, and uptime and operational resilience improved rather than being traded away.

What transfersCloud savings that cost you reliability aren't savings, they're a deferred incident with interest. And the biggest line item is often negotiated, not engineered.

Prior executive role · security and compliance

A compliance program built before the deals needed it

Situation: Enterprise and government buyers were asking audit questions the company had no program to answer.
The call: Build the SOC program from the ground up: vendor selection, cross-department controls, the system narrative, and the company-wide information security policy.
Said no to: Answering questionnaires one deal at a time and calling it a program.
Returned: Multiple SOC 1 and SOC 2 Type II audits passed over many years, with national retail, restaurant, and logistics brands on the customer list. Later extended at the same company to ISO 27001 and FedRAMP LI-SaaS.

What transfersCompliance sequenced ahead of the pipeline is a revenue lever. Sequenced behind it, it's a stalled quarter.

Prior executive role · production agent AI

The agent that books for real customers

Situation: An AI feature that doesn't just answer, it takes the action. In the reservation product of an enterprise workplace management platform, the failure mode isn't an awkward sentence, it's a wrong booking on a real customer's calendar.
The call: Build it as a production reservation agent, then gate every change behind a five-level evaluation covering outcome, path, details, quality and safety.
Said no to: Open-ended autonomy. Strict tool schemas, policy-driven prompts, and an explicit confirmation step before anything commits.
Returned: A reservation agent running in production against real customer traffic, with every change validated on sampled live traffic before release.

What transfersAn agent that takes actions needs a different bar than one that writes text. The confirmation step is the product, not a safety afterthought.

Prior executive role · AI platform

One retrieval platform, not five

Situation: Several product teams, each heading toward their own retrieval stack for their own AI feature.
The call: One shared platform, and one team that owned it: ingestion, chunking, embeddings, vector retrieval and orchestration behind a common API, with provider choice in configuration rather than hard-wired per product. That team owned the architecture and the controls, so the API stayed a real contract.
Said no to: Per-product retrieval stacks, and a single-vendor commitment made before anyone knew which model would win.
Returned: One shared API and an internal model catalogue several products drew from, so the second and third AI features cost less to build than the first.

What transfersThe first AI feature is a product decision. The second is a platform decision, cheaper made before you need it. A shared platform with no team owning it becomes five platforms again.

NLT Labs · reference build

Eight days to the second storefront

Situation: I needed to know whether a fleet built for one storefront was a bespoke artifact or a transferable kit.
The call: Point the same kit at a second, unrelated storefront and run it live, instead of writing an architecture doc claiming portability.
Said no to: A generalization layer up front. Building the abstraction before the second real case is how platform teams spend a quarter on the wrong seams.
Returned: Eight days from first commit to a second live storefront. Nine of the 30 agents exist only to check the other 21.

What transfersThe honest test of “platform or one-off” is a second real customer, not a design review.

Prior executive role · margin and org scaling

Ten points of margin while the org tripled

Situation: A services-heavy business moving to pure software, with revenue compounding and headcount climbing to match. Gross margin is usually what quietly gets worse in that transition.
The call: Run the services-to-software move as a margin program with its own targets, rather than treating margin as whatever fell out of the roadmap.
Said no to: Buying growth with headcount. Every implementation that stayed bespoke was a services contract wearing a software label.
Returned: Ten points of gross margin added while the organization grew from 28 people to 80.

What transfersThe revenue mix moves first; the cost base follows only if somebody makes it. Margin is a decision, not an outcome.

The scale those calls were made at: a global enterprise SaaS business grown two and a half times in ARR.

The workplace platform I was CTO of is covered in the 2025 Verdantix Buyer's Guide to Hybrid Workplace Solutions, an independent analysis of 21 hybrid workplace providers. Verdantix found that its machine-learning analytics help firms refine space utilization and hybrid strategies.

Except where explicitly labeled "NLT Labs," results described on this page are from my work as a technology executive at other companies, not from NLT Labs client engagements, and are anonymized where needed. They're shared as experience, not as a promise of comparable outcomes.

Next step

Facing a call like these?
Talk it through first.

Thirty minutes, no obligation. Bring the decision you're sitting on and I'll walk through it the same way: what's running, the options, what to say no to, and what it should return.

Book 30 minutes → Email instead

Prefer to write? hello@nltlabs.ai