Shared AI capability & business performance
Keep what your people learn, connect it to the work and give leaders evidence of what is improving.
CSM · Video · 10 min read · Updated 19 November 2026
The short answer
Watch: How do you turn individual AI use into company-wide capability?
Watch how CSM connects individual AI use to shared knowledge, workflows and business results.
What is the difference between individual AI use and company capability?
Individual AI use helps a person complete a task. Company capability means the business has a dependable way to repeat that work: approved information, clear responsibilities, reusable guidance, controlled actions and a record of the result. Another authorised colleague should be able to use the process without reconstructing everything from the beginning.
A useful answer in a chat is a starting point. To become company knowledge, it needs context: which question it answered, what evidence supports it, who reviewed it and where it applies. To become a workflow, it also needs an owner, a trigger, a handover and an acceptance criterion.
At CSM, our starting point is the work the business needs to complete. We connect the information, judgement and systems around that work so useful learning can improve the next job.
Does the business really forget when someone closes a chat?
The video uses that phrase to describe a practical organisational gap. A saved conversation can still be difficult for a colleague to find, interpret or safely reuse. Retaining the text does not establish that its answer is correct, current, approved or relevant to a different customer.
The question is whether useful knowledge reaches the next appropriate person at the right point in their work. If a team repeatedly uploads the same background, makes the same corrections or asks the same expert to explain an exception, there may be no dependable process for retaining that learning.
Start by examining one recurring correction. Where is it recorded? Who decides whether it should become shared guidance? How does a later user discover it? Those questions are more useful than assuming every conversation should be copied into a central store.
Why might personal productivity improve without company earnings improving?
An employee can draft faster while the work still waits for missing information, review or approval. The time released may be absorbed by additional activity, or the output may need substantial checking. Faster tasks do not necessarily improve the complete process.
McKinsey’s 2026 State of AI survey reports that 80% of respondents said AI improved their individual productivity, while 37% reported a positive contribution to their organisation’s EBIT. These are reported survey findings, not a measured return for every business or proof that a particular operating model causes better results. Read the McKinsey research.
For your business, establish a baseline for completed work, elapsed time, human effort, errors and cost. Then decide how any released capacity will be used. Additional profitable work, an avoided future hire and a reduction in an existing expense are different benefits and should be measured separately.
What knowledge should the company keep?
Keep approved information that helps people complete repeatable work: customer requirements, current commercial rules, product or service definitions, relevant evidence, decision criteria and examples of acceptable outputs. Capture why a correction mattered, rather than saving only the corrected wording.
Suppose an estimator identifies an exclusion relevant to a particular type of project. Record the condition, the supporting evidence and the person authorised to approve it. Specify where it applies and when it needs review. Do not transfer confidential project information into another customer’s workflow.
Separate source records, company rules and temporary working notes. A draft answer is not automatically policy. A useful knowledge system makes these differences visible and gives each important item an owner, a review date and an appropriate audience.
How does one person’s learning become reusable knowledge?
Use a simple review loop: capture the proposed improvement, assess its evidence and scope, approve the change, update the relevant playbook and test representative work. Record the version so you can understand which guidance produced a particular result.
For example, a sales colleague may correct an AI draft that promises an unsupported delivery time. The reusable lesson could be to retrieve the approved service terms and flag missing dates for review. The lesson is not to copy that customer’s negotiated deadline into every future proposal.
The process owner should decide whether a correction is general, customer-specific or temporary. Improving instructions and reference material does not itself retrain the underlying model. Test whether the change reduces the original error without introducing new ones.
Do you need to replace all the systems you already use?
Start by identifying the systems that already hold authoritative information and the gaps between them. A company-wide approach can build around existing tools. It does not require every record to move into one database or every employee to use an identical interface.
Agree which system owns each important fact, how records are matched and what happens when two sources disagree. A customer name alone may not be enough to link records reliably. A connection also needs permission checks, monitoring and a clear response when updates stop arriving.
CSM’s integration experience provides a starting point. Each client implementation still needs its own access, field mappings, rules and checks. An available connector does not establish that a particular workflow is complete or that the information is correct.
What does a connected workflow look like in practice?
Consider an enquiry that needs a proposal. The workflow retrieves the authorised customer record, relevant requirements, approved service information and current commercial terms. It checks for missing information before preparing a draft.
AI can organise the evidence and draft the response. Controlled rules can enforce required fields and route the item to an appropriate reviewer. The reviewer sees the sources, unresolved questions and proposed commitment, then approves the exact version or sends it back with a reason.
After approval, the system records the outcome and creates the agreed follow-up. A missing record, failed connection or absent approver follows a defined exception route. A useful improvement is retained through the review loop so the next proposal benefits from it.
This is an illustrative workflow, not a claim that a particular client has deployed those exact steps. The design and controls must fit the work and its consequences.
Where do agents fit, and how much should they do autonomously?
An agent needs a bounded job description: required inputs, expected output, permitted tools, acceptance criteria and circumstances that require a person. Begin with preparation and assistance where that lets the team inspect results before consequential actions occur.
The video’s progression from 50% to 80% of a task describes an ambition to improve coverage over time. It is not a guaranteed starting point or improvement curve. Define what the percentage measures and include human review, exceptions and maintenance when assessing the result.
Expand autonomy only when testing supports it and the permissions are appropriate. Some decisions should continue to require human approval. More independent action is useful only if it improves the outcome without introducing unacceptable risk or supervision costs.
How do you avoid turning shared knowledge into a shared mistake?
Keep evidence attached to important claims and make uncertainty visible. Limit access to relevant information and distinguish instructions approved by the business from material retrieved from documents or external sources. A source may contain useful facts without being authorised to tell the system what to do.
Test missing evidence, conflicting instructions, outdated records, denied permissions and failed connections. Require the workflow to stop or escalate where its conditions are not met. A prompt asking an agent to seek approval is not a substitute for the action itself checking an approval record.
Give someone responsibility for monitoring the system after launch. They need a route to investigate problems, revoke access, restore service and update the playbooks. Shared capability needs maintenance as the business, information and tools change.
What does commercial intelligence add?
Commercial intelligence brings the operational evidence together around leadership decisions. It should show where work is blocked, what exposure exists, which outcomes are changing and who can act. A red, amber or green status needs a clear definition and a visible source.
For example, proposal volume may rise while approval turnaround gets worse. Looking at both can reveal a review bottleneck that an activity dashboard would miss. The appropriate response might be clearer briefing, improved evidence or additional review capacity.
Attach an owner and next action to important exceptions. Leaders should be able to inspect the underlying records and understand how current they are. The value comes from better intervention and follow-through, not simply from producing a more attractive report.
How would CSM take this from discovery to live use?
We begin with a leadership priority and follow a real piece of work with the people doing it. We identify where they search, retype, wait and check, then assess opportunities by commercial impact, time and capacity, key-person dependence, risk and readiness.
The selected priorities form a ninety-day roadmap with owners, scope and success measures. A prototype tests important assumptions using appropriately authorised examples. The full build needs reliability and permission checks, followed by a pilot with the users who will complete the work.
Go-live includes training, monitoring, support and acceptance criteria agreed for that implementation. At the next review, compare the result with the baseline and reprioritise. Ninety days is a review cadence, not a promise that every possible system can be delivered within that period.
Our fractional team works alongside the people who understand your business. The aim is to combine their knowledge with implementation capability and leave clear ownership of the resulting process.
How should you measure whether the business is improving?
Agree the definitions before the pilot. Track completed work, turnaround, human review time, rework, exceptions and recurring costs. Record how estimates are calculated and distinguish a proposed benefit from one observed in live use.
Do not add overlapping savings twice. If drafting and review improvements affect the same job, measure the total change across that job. If saved hours are used for additional work, assess the value of that extra capacity rather than reporting the same hours as a cash cost reduction.
Keep revenue tracked, revenue attributed and profit improvement separate. A reporting system observing revenue does not mean it caused that revenue. A rise in activity does not establish an improvement in EBITDA. Use a client-specific results record to explain the baseline, evidence, assumptions and next decision.
What should leaders agree before committing?
Agree what is included, what success means, who contributes information and decisions, and what remains outside the first scope. Clarify data access, security responsibilities, testing, training and ongoing support.
Document ownership and handover for custom code, workflows, data and configuration. Third-party subscriptions, hosting and maintenance may continue even when you own the custom work. Any performance guarantee needs a written baseline, measurement period, responsibilities and remedy.
Calculate payback from the benefits the business can actually realise and the full implementation and running costs. The aim is a more dependable and valuable business; no system can promise to eliminate every person-related dependency or automatically increase a sale valuation.
Where should you start?
Bring one recurring job that slows the team down, one recent example and the person who knows where it gets stuck. Ask what the next colleague would need to complete the job without repeating the same investigation.
That is enough to begin a useful discovery conversation. For more detail, explore process mapping for AI agents, building an AI Commercial Brain and reducing key-person dependency.
Questions for your leadership team
Which AI-assisted task keeps being reinvented by different people?
What useful corrections are repeatedly made but never retained?
Who approves shared guidance and keeps it current?
Which source owns each important fact in this workflow?
What can an agent prepare, change or send, and who approves it?
Where might faster output create a larger queue for another team?
How will we measure net benefit after review, exceptions and costs?
Who maintains the process when the original expert or builder is unavailable?
Full video transcript
The spoken script is reproduced below; production notes and on-screen directions have been omitted.
Every day, your people ask AI for help. And every day, your business forgets the answer.
ChatGPT, Claude, Copilot, Gemini. One person, in one chat, on one task. Then the chat closes, and it’s gone.
That helps them. It doesn’t help the business. You can’t see how they’re using it, or whether they’re using it the right way. They re-upload the same context every time. Nothing is built into a business process. So the business isn’t compounding. And when that person leaves, everything they’ve learned leaves with them.
McKinsey surveyed more than seventeen hundred business leaders. Eight in ten said AI had made them more productive. Only thirty-seven percent said it had improved the company’s earnings.
There’s no brain built into the business.
That’s where we come in, and we do it very differently. We’re CSM. We believe systems run the business, and people run the systems. So we build AI into the business itself. We connect your data, capture what your people know, and build it into your processes and your systems. What one person works out, the whole company keeps.
That’s how you build a business worth buying: more profitable, more efficient, with stronger EBITDA and no dependency on any one person.
We’ve spent five years and hundreds of thousands building that brain, with a full team led by Kevin O’Brien, in our own business and in more than fifty others. That’s our unfair advantage, and every client gets the benefit of it.
Every system we build must deliver measurable value. Across those businesses, we’ve tracked five hundred and eighty million euro in revenue. More than thirty-two thousand hours of employee time saved. A hundred and fifty business processes improved. Seventy-five systems and automations put in place.
You can see it in the platforms we’ve built: a construction operating system, a mission control, a finance operating system, and Site-to-Quote, which turns a site visit into a quote.
Every client gets their own portal, where we track their exact results. And every ninety days, we report on what’s been delivered, what it achieved and what comes next.
But don’t take our word for it.
It starts with one real piece of work. We follow it from the first request to the finished result, and we find where it gets stuck. Then we score every opportunity: the commercial impact, the time it gives back, and what stops if one key person leaves. The top priorities become your ninety-day roadmap. We prototype on your real examples, build it, pilot it with your people and go live. Then we score again, so progress compounds.
Underneath, it’s built from the data up. Your data, connected through more than three thousand integrations we’ve already built. Your knowledge and playbooks, so AI follows your way of working. Then agents: an AI workforce built around your people. Each one gets a job description, and starts by taking on half of a task. Then sixty percent. Seventy. Eighty.
And on top, commercial intelligence. It shows you what’s in the green, what’s in the amber and what’s in the red, where the constraint is, and where your leaders need to step in. So if your business could talk to you, what would it say?
Every system is completely custom. It’s yours for life. We put performance guarantees in place, and the time it saves can pay back the investment within a few months.
So bring us one piece of work that slows your team down. Book a thirty-minute call, and we’ll show you what’s possible.
Build shared AI capability with CSM
Bring one process that repeatedly slows your team down. Start with a digital discovery workshop to map the work, agree the priorities and define a sensible first implementation.