Resources
When not to build
The useful answer is sometimes: not yet. Scope is a decision, not a courtesy.
Most consultancies are paid to say yes. The incentive is a statement of work, then a team, then a timeline that assumes the constraint is already a build. Sometimes it is not. The useful answer — the one that saves the next two quarters — is: not yet.
AKAF is a lean consultancy. We do not keep a bench warm. Saying “don’t build this” costs us a project. It also keeps the ones we take honest. Production, or we say don’t build it, is not a tagline. It is how we stay small.
Signals the work is not a build yet
Stop before a fixed-scope build when any of these are true:
- Two numbers, no owner. Finance and Operations cannot say the grain in one sentence without a fight. A dashboard will not settle that. Neither will an AI demo.
- The constraint is still a feeling. “We need a lakehouse” or “we need AI” is a category, not a job. If you cannot name the decision that is blocked, you are shopping for a platform.
- Nobody will run it on Monday. The person who would own alerts, access, and the first failure is unnamed — or already at capacity with no mandate to take it.
- The source of truth is a workbook with a hidden column. You can ship a pipeline on top of that. You will be reconstructing the undocumented rule in production for months.
- Success is a demo date. If the only milestone is a room that nods, the work will optimize for the nod.
None of these mean “never.” They mean the next honest delivery is smaller than the one on the slide.
What to do instead
The substitute for a premature build is usually one of four things:
- Write the grain. One metric, one sentence, including what is excluded. If the sentence cannot be said out loud, that disagreement is the deliverable. See one number everyone trusts.
- Name an owner. Not a steering committee. One person who can be wrong in public. No owner, no build.
- Mark unofficial extracts unofficial. Kill the second gold, or admit you do not have a certified table and need a pipeline before a report — or before production AI.
- Run a short discovery. Map the systems, the blocked decision, and a buildable scope. You leave with a written shape of the work — not a slide deck, and not a team already billing against a guess.
Discovery is still work. It is typically one to two weeks, senior-led. It is cheaper than a three-month build pointed at the wrong constraint.
When a build is the right next step
We will quote a fixed-scope build when:
- The constraint is named, and a decision is blocked on it.
- The systems in play are known well enough to estimate without theatre.
- There is an owner who will take the system after go-live.
- Production is the bar — a pipeline that lands, a number a room can defend, a site that launches, AI with an exam and a cap — not a prototype to “explore options.”
Duration follows the constraint. A personal site is weeks. A lakehouse or an app is longer. Continuity after handover is optional, month to month, so the system stays yours: alerts, small changes, a named person. Not a bench of juniors sitting on the clock.
How we say no without wasting the week
Send the site, the data problem, or the deadline. If the shape is unclear, that is the first conversation — not a proposal padded to look like certainty. We would rather tell you the work is a definition, a pipeline, a report, or not yet, than start a build we would not put our name on.
That same honesty shows up in production AI vs demo-ware: if the warehouse is not trusted, wrapping a chatbot around it is not a product. Typical engagement shapes are discovery, a fixed-scope build, then optional continuity. We quote after we understand the work.
If this is the constraint: tell us what you were about to buy, and what decision it was supposed to change. We will tell you whether to build it.