What your client is actually grading you on

Between your company and your client’s there is a strip of no-man’s-land that no tool owns. Your ticketing system doesn’t cover it. Theirs doesn’t either. Email actively makes it worse. Every forward deployed engagement lives or dies in that strip — and your client is grading how you manage it from the first kickoff call, against a rubric they will never write down. Here is the rubric, and the standard ways vendors flunk it.

The rubric

Expectations got reset at the top of the market. Palantir claims its AIP bootcamps take customers from zero to a working use case in one to five days — a self-reported number, but one your buyer has heard pitched, or heard imitated. Meanwhile MIT’s NANDA research (methodological caveats and all) put a number on the buyer’s core fear: most pilots fail because tools never adapt to the organization’s workflows, not because the technology is weak. Speed to value, and depth of fit. Every question below is one of those two in disguise.

What they say What they’re actually asking What convinces them
“What does week one look like?” Are you bootcamp-fast or SI-slow? Something running in their environment inside days
“Who’s our point of contact?” Will our asks vanish into your org? Every request visible somewhere both sides can see, with an owner and a date
“Can you work with our security team?” Will procurement kill this in month four? Evidence assembled before it’s requested
“How custom is this?” Are we buying a product or funding your R&D? A line between configuration and roadmap, drawn in writing
“How do we know it’s working?” Give me ammunition for my board The numbers from the sales deck, tracked openly against reality
“Who else should be in these meetings?” Does this survive if my champion changes jobs? You, insisting on multi-threading early

The fourth row is where AI-era vendors most often stumble. Andrusko notes that most enterprises don’t want to be a vendor’s permanent laboratory — they want predictable implementations. “Forward deployed” reads to a buyer as white-glove, not open-ended. Scope and date the bespoke work, or the client will re-scope it for you at renewal.

The standard failures

Pitfall What the client experiences Countermeasure
Overfitting without a paper trail V2 breaks their customizations; nobody remembers why they exist Document every fork as it happens: what changed, for whom, why
Status theater A weekly email that’s stale on arrival One live view both sides read; kill the email
Blocker latency — on their side Weeks lost to an approval sitting in an inbox Chase client-side blockers daily, politely, visibly
Single-threaded knowledge Your senior engineer is the system of record; they take PTO, the engagement stalls Decisions written down where both sides can find them
No memory between engagements Engagement four relearns engagement three; the client can tell A playbook that compounds across clients

Note what’s not on the list, because no coordination fix touches it: a product that can’t actually do the thing, and a deal sold to the wrong champion. If either of those is your problem, solve it first; process is not anesthesia.

Two rows deserve unpacking because vendors consistently misassign blame. Blocker latency is usually a client-side failure — the access grant, the security sign-off, the SME who won’t book time — but the client will still experience it as your slowness. MIT’s finding that deployments succeed when line managers are empowered, not just central AI labs, suggests who you should be making the plan visible to. And overfitting is not itself the sin — Palantir’s field engineers were explicitly told to overfit. The sin is overfitting amnesiacally, so nobody can later tell load-bearing customization from scar tissue.

Read both tables again: none of these are engineering failures. They are failures of the boundary — asks, status, decisions, and evidence falling into the strip between two companies. That boundary is the surface Gravel was built to own, with agents doing the daily tending that humans reliably drop. But the rubric exists whether or not you buy anything. Print it, and grade yourself before your client does.

Sources

  1. Palantir, "How AIP Bootcamps Work"
  2. Fortune on the MIT NANDA report
  3. a16z, "The Palantirization of everything"
  4. Nabeel Qureshi, "Reflections on Palantir"

Filed under · Client expectations · Deployment · Enterprise sales

The lever in this note is the one we built.

Gravel puts a crew of agents on the coordination half of the forward deployed motion, so one person supervises engagements instead of carrying them.