Skip to content

Advisory

When the question is bigger than a mandate.

Some clients arrive knowing the outcome. Others arrive with a harder question: where should AI change how we operate, what should we build, what should we stop doing. We do that work too, with one rule that keeps us honest: strategy here ends in shipped software, not in slides.

The operating diagnostic

First, we put your whole operation on one map.

Fifty-plus steps, three colors. Green stays human. Red stays your call. Blue goes to agents.

  • GREEN, the human moments. We never automate these.

  • RED, the decisions. Consolidated, and kept yours.

  • BLUE, hands on keyboard. Zero added value. It goes to the agents.

Today the blue steps belong to your most junior people, with the biggest margin for error. Your sales team spends half its week in there too.

Then we collapse it.

One company we mapped ran 12,450 manual events a year through steps like these. Fifty steps became a dozen. The dozen that need a human kept one.

A map of your operation, every step labeled and color-coded three ways: green discs for the human moments, red diamonds for the decisions, blue squares for hands-on-keyboard work. The blue squares fold away, the red diamonds consolidate into one decision gate, and the surviving steps land on a single run line through scope, build, prove, and ship, ending in a shipped tick.

EMBEDDED

Forward-deployed engineer

A senior engineer inside your team, weekly. They sit in your planning, translate strategy into mandate pipelines, and answer for delivery dates in your meetings, not ours. Runs alongside a standing mandate so what ships stays watched.

Cadence
2–5 DAYS / WEEK
Term
QUARTERLY, RENEWABLE
Output
A MANDATE PIPELINE

RESEARCH

AI research mandates

Custom models, evaluation harnesses, and the hard problems off the beaten path. Same discipline as every mandate: the acceptance criterion is a benchmark number, and the warranty is reproducibility. If a benchmark cannot be written, it starts as a paid probe.

Criterion
A BENCHMARK, SIGNED
Warranty
REPRODUCIBILITY
Yours
MODELS, DATA, IP

DIAGNOSTIC

Operating diagnostic

The probe, aimed at your operation instead of your codebase. Two weeks mapping where hours actually go, which workflows agents can own, and what that is worth. You get the map, the math, and a sequenced mandate program with a price on each step.

Length
2 WEEKS, FIXED FEE
Fee
CREDITS TO MANDATE ONE
Output
PRICED PROGRAM + ROI

All three end in the same place: mandates in the portal, with criteria, evidence, and a warranty. Advice that never becomes a mandate is advice we failed at, and we track it.

How mandates work

ADVISORY FITS WHEN

  • You own a P&L and suspect AI changes your cost base, but nobody can say where first.
  • A transformation program is stuck at the pilot stage and the board is asking why.
  • You need a model or capability that does not exist off the shelf, with proof it works.

IT DOES NOT FIT WHEN

  • You want a strategy deck without the obligation to ship anything. Plenty of firms sell that; we do not.
  • You want staff augmentation by another name. The FDE produces mandates, not billable hours.
  • The honest first step is a 12,000 prebuilt mandate. We will say so and point you there.

Start with the diagnostic.

Two weeks, fixed fee, credited to your first mandate. Write to advisory@salfati.group, or run the code probe first.

Run the free probe