You Probably Don't Need AI

That is an odd thing for an AI company to lead with, so let's be precise about what we mean. AI works. We use it every day and it has changed how our business runs. What we do not believe is that every business should be buying it right now, and we think the industry's refusal to say so out loud is why so much of the money being spent on it is disappearing.

The prevailing sales pitch is fear. Move now or be left behind. Your competitors are already doing it. The window is closing. It is effective, it is exhausting, and it produces exactly the outcome you would expect: a great many businesses buying a powerful tool before anyone has decided what job it is for.

So this piece does two things. First, it makes an honest case for why a fair number of businesses should not be starting with AI yet, with the specific signals that tell you which one you are. Then, if you are not in that group, it walks through the system we actually run internally, the one that took us from enthusiasm to something that sticks. It is less complicated than you would think.

Everyone has started. Almost nobody has finished.

McKinsey surveyed 1,993 respondents across 105 countries for its November 2025 State of AI report. Adoption is close to universal: 88% now use AI somewhere in the business. But only 39% can point to any effect on enterprise-level EBIT, and just 7% say AI is fully scaled across the organisation.

Use AI in at least one business function 88%

Report any EBIT impact at enterprise level 39%

Say AI is fully scaled across the organisation 7%

Source: McKinsey & Company, The State of AI in 2025: Agents, Innovation, and Transformation, November 2025. The 7% figure is reported in McKinsey's accompanying analysis, December 2025.

Bain reaches the same place from a different direction: fewer than 20% of organisations have scaled their generative AI efforts in any meaningful way. Note what these numbers are not telling you. They are not saying the technology underdelivers. They are saying that the overwhelming majority of attempts stall somewhere between the first exciting demo and anything that changes how work gets done.

Almost nothing here is a technology failure. It is a long series of tools bought before anyone named the job.

Which brings us to the reframe this whole piece rests on. AI is not an answer. It is a tool that makes a team better at solving problems and measurably more productive at work they already understand. It does not generate a strategy you do not have. It will not fix a process that is broken, and it certainly will not tell you which parts of your business are worth improving. Point it at a well-understood problem and it is remarkable. Point it at a vague ambition and you get the 61%.

Four signs it is not your turn yet

These are the situations where we would tell you to spend the money elsewhere. Not forever, but not this quarter.

Your constraint is demand, not capacity

If your team has spare hours and your problem is not enough customers, giving everyone eight hours a week back solves nothing. Fix the pipeline first. AI will still be here.

Nothing you do happens twice the same way

Genuinely bespoke, low-volume work gives the tool nothing to learn from and no pattern to apply. Some businesses really are like this, and there is nothing wrong with that.

Nothing is written down, and nobody intends to start

This is the big one, and the rest of this article explains why. If your standards, pricing and methods live only in people's heads, there is nothing for the tool to stand on.

Nobody will own it

Not a working group. A named person with real time for it. If that person does not exist and you are not willing to create them, the initiative has a predictable ending.

If two or more of those describe you, our honest advice is to wait, and to spend the budget on whatever is actually limiting you. That is not us being coy. Half of what we get asked to build should not be built, and saying so early is cheaper for everyone than discovering it in month four.

If none of them describe you, the rest of this is for you.

What it looks like when it does stick

We are an AI-native company, so we are an easy case. But the interesting part of our own rollout is not that our developers got faster. It is that the gains showed up in every function, including the ones nobody markets AI to.

Fig. 2 · Our own team

Hours back per week, by role

Illustrative. Our own team, and our measurement is still being refined.

Developer ~10h

Jira and Xray tests · scope pricer · sprint readiness · code and PR helper

Project manager ~8h

Meeting follow-up · sprint dashboards · retro logger · status packs

Sales ~7h

Proposal builder · branded decks · weekly leads plan

Operations ~6h

Meeting follow-up · retro logger · document generation

HR ~4h

One-to-one prep · job specs · candidate assessment

We publish those with the caveat attached because the caveat is the point. They are our own numbers, they are self-reported, and we are still tightening how we measure them. Anyone showing you a precise productivity percentage for a technology this new is telling you a story. What we are confident about is the shape: the biggest gains landed on the roles that spend their days turning context into documents, and none of them are engineering roles.

The part everyone skips: context

Here is the thing that separated our second year of using this from our first. The model is not the differentiator. Your competitor can license the identical one this afternoon, for the same price, and be running it before dinner. Model quality is a commodity and it is getting more commoditised every quarter.

The model is rented. Your context is owned.

Your context is your clients and the whole history of every deal. It is the reasoning behind decisions you made years ago. It is the specific, hard-won way your business actually wins work, which is written nowhere and is not the way the website describes it. Nobody else has that, and no model arrives with it. A generic assistant with no context gives you generic output, which is precisely the disappointment most people describe after their first month.

So the work is not choosing a model. It is getting your context somewhere the model can reach it, and keeping it true.

Your knowledge comes in two kinds

This is the single most useful distinction we have found, and treating both kinds the same way is the most common reason these projects rot quietly six months in.

Static

Changes rarely

Your standards, how you operate, your strategy, your pricing.

Curate it. Keep it small, keep it in one guarded place, and make a human approve every change before it counts as true.

Dynamic

Changes constantly

Emails, calls, meeting transcripts, your CRM.

Never copy it. Leave it in the system that owns it and read it live. A copy is stale the moment you make it.

Connect, do not copy. The source is always fresher than your snapshot, and the moment your team catches the assistant confidently quoting something six weeks out of date, you have lost them. Which is why we date every fact in the static core. A confident, stale answer is far more damaging than no answer at all.

The static core is also a team sport, not a documentation project handed to one unlucky person. It stays small and stays trusted only because the people using it are the people maintaining it. Find a champion to own it and make it a real part of their job rather than a side project, or it will drift within a quarter.

Two loops, and you can start the first one this week

You do not need a finished knowledge base. You need a first one and a way for it to grow. That is loop one. Loop two is what keeps it honest once real work starts flowing through it.

Fig. 3 · Loop one

Setting it up

Set up the store. Use AI itself to distil your existing documents into clean core knowledge.

Create a Project and wire it straight to that core knowledge.

Do the actual work inside that Project, so every output draws on your context.

Feed the delivery back, so real work keeps the core growing.

Fig. 4 · Loop two

Keeping it true

Work captures context automatically, in the flow of delivery.

AI drafts it clean, straight from the raw mess.

A human approves it before it counts as truth, and every fact gets dated.

It flows into the core, so context compounds instead of decaying.

The honest bit

The capture in loop two is automatic and it works well. The approval gate is the part we are still tightening. If you build this, plan for the discipline, not just the plumbing. That is where it gets hard, and it is not a technical problem.

Build skills on skills

Once the context exists, the leverage comes from stacking. We taught the tool our house style once, covering how a riivo document is laid out and how a riivo deck looks, then made those foundational skills. Everything built on top inherits the brand and formatting for free, so each workflow skill only has to do one real job.

So the proposal builder assembles a client proposal from our knowledge and comes out on brand without being told. The quoting skill turns a scope into a consistently priced quote. Meeting follow-up turns a recording into notes and actions. The monthly one-to-one skill pulls the prior month together into a balanced view before a manager walks into the room. Four different jobs, one shared foundation, and nobody re-explaining what a riivo document looks like ever again.

Why our team adopted it so fast

This is the part that surprises people, and it is the opposite of what you would guess.

Our whole team is on Claude Team licences. Around three quarters are on the standard seat at $25 per person per month on monthly billing, which handles everyday work perfectly well. The rest are on the premium seat at $125, granted by role and known usage. People motivate for it, and annual billing brings both figures down further. A small overage is allowed and it is capped, so there are no surprise bills.

The constraint is the point

Because not everyone had unlimited usage, the team learned to manage their context windows, their tokens and their model selection.

If we had handed everybody an unlimited licence on day one, that discipline would never have formed. The scarcity did the teaching. It is the cheapest training programme we have ever run, and we did not design it on purpose.

The other half of the adoption story is that we let people play first. Experiment freely, share what worked in the open, then take the genuinely good ones and harden them into the core so everyone gets them. Codify the wins, embed them where people already work, and measure what they gave back. Enthusiasm arrives on its own when someone sees a colleague's trick save them an afternoon.

Set against the time it gives back, the monthly licence cost is not the number that matters. Which is worth sitting with, because cost is almost never what actually stops these projects. Discipline is.

Do this next week

If you cleared the four signs above, here is the whole first step. It costs a licence and an afternoon, and you do not need us for any of it.

Pick a data store

Notion or Confluence are both quick to start with. The choice matters far less than starting.

Populate it with AI

Point it at the business and project documents you already have and let it distil them into clean core knowledge.

Create a Project on top

Point it at that store, then run your real conversations and workflows inside it. Every output improves from there.

Start small and let it compound. You are not trying to build the finished thing. You are trying to build the first version of something that gets better every week, which is a much easier target and the only one that has ever worked.

riivo AI Blueprint

Still not sure whether you are in the group that should wait?

That is the conversation our discovery call is for. We map how your processes actually run, find where the time genuinely goes, and work out whether AI earns its place before anyone writes a line of code. The four-week AI Blueprint that follows only makes sense if it does.

The call is free, and if the honest answer is that you should spend the money elsewhere this year, we will say so.

Book a free discovery call →

References

McKinsey & Company. The State of AI in 2025: Agents, Innovation, and Transformation. November 2025. mckinsey.com
McKinsey & Company. AI at Work but Not at Scale. Week in Charts, December 2025. mckinsey.com
Bain & Company. How Do Companies Create Value with AI? June 2026. bain.com
Per-role time savings and licence mix: riivo internal, self-reported and illustrative. Measurement is still being refined.