Manifesto
The PointProven Manifesto
The deck is where thinking gets judged
A few people around a table, a deck on the screen, an ask on the last page. The client has to decide whether the recommendation holds up. Your team has to defend it.
For consultants, the deck is the product: judged against MBB output, built without the MBB machine. It's what the client keeps when the engagement ends, and what sells the next one. Forwarded to someone who missed the meeting, it has to make the case without you.
A good deck is hard to build because writing is thinking. Forcing an analysis into a structured argument is how you discover what you actually believe: finding the governing thought, testing what supports it, deciding what to leave out. Those hours wrestling a storyline into shape were never formatting overhead. They were the job.
With AI, thinking arrives untested
Type a sentence, get a polished deck. You read it back and it sounds right. The thinking feels like it already happened. But it was skipped, and nothing on the page says so.
AI decides without you. The AI chose whose question the deck answers, what the answer is, and what counts as evidence. None of those decisions were shown to you.
AI imitates rigor. A model can explain MECE and still produce overlapping reasons for a recommendation. The supporting facts can be invented. An invented number looks exactly like a sourced one.
AI doubles down. Ask for help with a weak argument and AI strengthens the wording while accepting the premise. A misread assumption becomes a finding, then the anchor of the recommendation. The longer the brief, the more of it the model forgets. The evidence that would have caught the error was in the notes the whole time.
The result is slop: work that looks finished, advances nothing, and hands the cleanup cost to whoever reads it.
AI can't make the call. It can break the client's trust.
Generating isn't deciding. Someone has to decide whether the evidence warrants the recommendation and whether the client can act on it, and sometimes the answer is no. A generated deck arrives with that decision unmade. The client will ask you why, and "the AI did it" is a confession that nobody was in charge.
Pasting is disclosing. The client's material is in your custody, and custody means you answer for every place it goes. Their numbers. The strategy they have not announced. Paste it into a personal account and the only protection left is the vendor's promise. By default the vendor keeps it and trains on it, and the terms have been rewritten more than once. A promise the vendor can rewrite is not custody.
Ignoring AI concedes speed and leverage
The upside is real. Done well, AI gives a small team the reach of a large one: the first structured draft in minutes, the formatting that ate the evenings gone, the partner's time spent on the argument rather than the alignment. That is leverage.
No option is free. Keep AI out of the firm and you give up that upside. Paste and approve and the call goes unexamined. Rebuild the argument after the slides are made and you have given back the speed you came for. None of those is a serious way to work.
AI built for consulting should capture the craft
AI built for consulting should build the argument the way the craft builds it, and then argue against it. The decision stays with the consultant. The client's material stays the client's. The speed follows from the order of work.
1. The deck is an answer. Start with the question.
There is a decision someone has to make, and the deck exists to help them make it. AI should begin with who is deciding, what they need to decide, and what would convince them. Settle that in an outline you can mark up before it has become twenty slides. Rework is where the time goes.
2. A partner, not a flatterer.
The partner reviewing your deck sends you back to the premise. So AI built for consulting should be engineered to argue against its own answer: every recommendation tested against the evidence that would sink it, every number carrying a source, every gap flagged rather than written past. The checks should run on every draft and be tested on arguments they ought to reject. Sometimes the argument loses, and the deck you decide not to send is the best work of the week.
3. AI drafts. You decide.
AI takes the drafting. Whether to make the recommendation at all was always the job, and it stays with you. AI should put that decision in front of you while the argument is still an outline. You should be able to explain why you accepted the argument, what you turned down, and what would change your mind. Those conditions belong in the deck the client keeps. The deck is generated. The judgment isn't.
4. Your client's trust is not training data.
Custody does not end at the upload. We are building to this standard: client material is used for the request, never used to train a model, and kept only as long as the client wants it kept. A model has no instinct for the engagement wall, so the wall is built in: one client's material never informs another client's deck. Nobody on our side can read it, not even with the keys to the building. The client can verify that. Certificates describe how a vendor intends to behave. Architecture decides what a vendor is able to do.
5. Knowledge is in the model. Craft isn't.
AI knows the frameworks, and everyone has the same AI. What differs is what is built around it. Consulting craft is learned in review, one returned page at a time: a standard of rigor, a taste for what a partner will reject, a sense of how a CFO reads a page differently from a CMO. That standard has to be built into a harness, then tested on engagements the system has never seen, with the failures kept so the next model has to earn its place against them. So we do one thing, for one profession, and we are relentless about perfecting it.
We're building that discipline into PointProven
PointProven is the AI presentation platform we're building for consulting teams. The methodology has a McKinsey pedigree. We've spent decades applying it and teaching it. Now we're building it into the work, where you can use it on a deliverable due tomorrow.
Judge what we build against these convictions, starting with the next recommendation you have to defend. If one of them is wrong, tell us. We read every reply.
PointProven. Presentations that think.