Journal/Strategy/A field guide to picking your first wedge.
A field guide to picking your first wedge.
The first AI workflow you ship is rarely the most valuable one. It is the one that earns you the credibility to ship the second.
Published
Dec 30, 2025
Reading time
7 minutes
Category
Strategy
Fig. 11 · Object OBJ-1230Trajectory: pilot 1 → pilot 2
The most consequential decision in any AI rollout is the first one: which workflow to pick. It is also the decision teams spend the least time on, because it does not feel like the work. The work feels like building. Picking is upstream of all of it.
This is a short field guide to choosing the first wedge — informed by the projects we have run, and by McKinsey's State of AI 2025, which puts most of the realised EBIT impact in back-office redesign rather than front-office novelty.
01. Pick a wedge with visible friction
The first wedge should be a workflow where the people doing it are already complaining. Not in surveys. In hallways, in standups, in the small irritated emails that get sent at 5pm.
This matters for one reason: when the wedge ships, the people who feel the relief are the people who can vouch for the project. They will tell other teams. They will defend the budget at next year's planning. A wedge that improves a workflow nobody complained about is a wedge that has no champion when the org chart shifts.
The first wedge does not need to save the most money. It needs to save someone you trust the most aggravation.
— working principle
02. Pick a wedge that ships in one quarter
If the wedge cannot be in production within thirteen weeks, it is the wrong wedge. Not because longer projects are bad — they are sometimes correct — but because they are wrong as the first one.
The first wedge has to compound. It has to ship and produce a number, and the number has to fund the second wedge. A wedge that ships in week thirty has missed the budget cycle, the morale cycle, and the credibility cycle. It will not produce a second.
This rules out, for the first project, anything that requires a custom data lake from scratch, a multi-system integration with three vendors, or a custom-trained model. Those things are doable, sometimes correct, but never first.
03. Pick a wedge with a known answer
Counter-intuitively, the right wedge for a first project is one where the answer is already known — a person on the team already does the work, and does it correctly. The job is to scale them, not to invent something new.
Wedges where the answer is unknown — generative ideas, novel research, anything that begins with "what if we" — are the second or third pilot, not the first. The first one is about proving you can ship. Proving you can ship is easier when you can also evaluate the output against a person who already knows the right answer.
04. Pick a wedge whose owner wants it
The single best predictor of whether a wedge ships is whether the manager whose team is affected actually wants it. Not "thinks it is interesting." Wants it. Will champion it. Will defend the eval. Will sit through the change-management conversations.
If the wedge is technically the right one but the owning manager is lukewarm, pick a different wedge. There are always two or three good candidates in any organisation. The right one is the one with a real owner.
Rule of thumb: if the manager closest to the workflow cannot describe the win in one sentence — without using the word "transformation" — the wedge is not yet defined.
A small list of wedges that fit
In our experience, the following wedges fit all four criteria more often than not:
A draft-and-review queue for one category of inbound message.
A renewal alert system across a CRM and a billing tool.
A reconciliation between two systems that should agree but don't.
A daily digest for a leader who currently does it manually.
A first-pass classifier for incoming requests, routing to the right team.
Any of them can be the first wedge. The cheapest, smallest one with a real champion is almost always the right answer.
A short closing
The teams that go on to ship pilots two and three are the teams that picked pilot one with this much care. The teams that don't, almost always picked pilot one for the wrong reasons — because it was visible, because a vendor pitched it, because someone read an article. The first wedge is a strategic choice. Treat it that way.
Najvažnija odluka u svakom AI puštanju je prva: koji tok rada izabrati. To je ujedno i odluka na koju timovi troše najmanje vremena, jer se ne oseća kao posao. Posao se oseća kao izgradnja. Biranje je uzvodno od svega toga.
Ovo je kratak vodič kroz teren za biranje prvog zahvata — zasnovan na projektima koje smo vodili i na McKinsey-jevoj The State of AI 2025, koja većinu realizovanog EBIT efekta smešta u redizajn back-officea, ne u front-office novost.
01. Birajte zahvat sa vidljivim trenjem
Prvi zahvat treba da bude tok rada gde se ljudi koji ga obavljaju već žale. Ne u anketama. U hodnicima, na standupima, u malim ljutitim mejlovima koji se šalju u 17h.
Ovo je važno iz jednog razloga: kada zahvat krene, ljudi koji osećaju olakšanje su ljudi koji mogu jamčiti za projekat. Reći će drugim timovima. Branit će budžet na sledećem godišnjem planiranju. Zahvat koji popravlja tok rada na koji se niko nije žalio jeste zahvat koji nema šampiona kada se organizaciona struktura pomeri.
Prvi zahvat ne mora da uštedi najviše novca. Treba da nekome kome verujete uštedi najviše živaca.
— radni princip
02. Birajte zahvat koji se isporučuje u jednom kvartalu
Ako zahvat ne može biti u produkciji u trinaest nedelja, pogrešan je zahvat. Ne zato što su duži projekti loši — ponekad su tačan izbor — nego zato što su pogrešni kao prvi.
Prvi zahvat mora da se akumulira. Mora da se isporuči i da proizvede broj, a broj mora da finansira drugi zahvat. Zahvat koji se isporuči u tridesetoj nedelji propustio je budžetski ciklus, ciklus morala i ciklus kredibiliteta. Neće proizvesti drugi.
Ovo isključuje, za prvi projekat, sve što zahteva prilagođeno jezero podataka od nule, integraciju više sistema sa tri dobavljača ili posebno obučen model. Te stvari su izvodljive, ponekad tačne, ali nikad prve.
03. Birajte zahvat sa poznatim odgovorom
Suprotno intuiciji, pravi zahvat za prvi projekat je onaj gde je odgovor već poznat — neko u timu već radi taj posao, i radi ga ispravno. Posao je da ih skalirate, ne da izmislite nešto novo.
Zahvati gde je odgovor nepoznat — generativne ideje, novo istraživanje, sve što počinje sa „šta ako bismo" — drugi su ili treći pilot, ne prvi. Prvi je o dokazivanju da možete da isporučite. Dokazivanje je lakše kada izlaz možete uporediti sa osobom koja već zna pravi odgovor.
04. Birajte zahvat čiji vlasnik to želi
Najbolji pojedinačni prediktor da li će zahvat biti isporučen je da li menadžer čiji je tim pogođen zaista to želi. Ne „misli da je zanimljivo". Želi. Brani. Brani evaluaciju. Sedi kroz razgovore o upravljanju promenom.
Ako je zahvat tehnički pravi ali je menadžer-vlasnik mlak, izaberite drugi zahvat. U svakoj organizaciji uvek postoje dva ili tri dobra kandidata. Pravi je onaj sa stvarnim vlasnikom.
Pravilo: ako menadžer najbliži toku rada ne može da opiše pobedu u jednoj rečenici — bez upotrebe reči „transformacija" — zahvat još nije definisan.
Mala lista zahvata koji odgovaraju
Po našem iskustvu, sledeći zahvati ispunjavaju sva četiri kriterijuma češće nego ne:
Red „nacrt-i-pregled" za jednu kategoriju dolazne pošte.
Sistem obaveštenja o obnovi preko CRM-a i alata za naplatu.
Usaglašavanje dva sistema koji bi trebalo da se slažu ali se ne slažu.
Dnevni rezime za lidera koji to trenutno radi ručno.
Prvopropusni klasifikator dolaznih zahteva, koji ih usmerava ka pravom timu.
Bilo koji od njih može biti prvi zahvat. Najjeftiniji, najmanji sa stvarnim šampionom skoro je uvek pravi odgovor.
Kratko zatvaranje
Timovi koji nastavljaju da puste pilote dva i tri timovi su koji su izabrali pilot jedan sa ovom dozom pažnje. Timovi koji ne, skoro uvek su izabrali pilot jedan iz pogrešnih razloga — jer je bio vidljiv, jer ga je dobavljač predložio, jer je neko pročitao članak. Prvi zahvat je strateški izbor. Tretirajte ga tako.
Filed under: STRATEGY · METHOD First published: Dec 30, 2025
