Last updated 2026-09-14 — First publication.

14. What to Expect — and Your Starter CoS Library

You have read thirteen chapters of what to do. This last one answers the question you had before you opened the book: what does it add up to, in numbers, over the next three years — and what do I do on Monday? It is the most practical chapter and the least theoretical, which is how a manual should end.

Four things follow. What to expect from the three numbers on the cover — costs, revenue, growth without hiring — year by year, with the honest ranges I use with clients. The assets you accumulate that competitors cannot buy. Your first ninety days, week by week. The master table of what the CEO does, what the CEO delegates and what the CEO verifies. And, to close, thirty Conditions of Satisfaction ready to be adapted: the to-do list of the whole book, written as commitments you can sign.

What to expect: three numbers, three years

Consider these ranges snapshots, not certainties. They are what I have seen in SMEs that followed the steps of chapter 5 with a qualified consultant, and they assume the first agent was placed on a routine, data-rich process. Companies that skip the audit or start with the most ambitious use case do not get these numbers: they join the 56% of CEOs who report no financial benefit from AI yet.

	Year 1: first agent › first AI Team	Year 2: mixed organization	Year 3: AI-first company
Lower costs	Cost per work unit −20/40% on the processes covered; one function	Operating cost of covered functions −15/25%; agencies and executional suppliers renegotiated on results	Cost structure with 20-30% variable, governed by caps; per-seat software largely replaced
Higher revenue	Conversion +10/30% where commercial agents are placed; response times from days to hours	New products and services tested at 3-5x the previous rate; 1-2 new markets served with the same structure	Revenue from products, markets and services that did not exist before: 10-20% of total
Growth without hiring	Capacity of the covered function ×1.5-2 with the same people	Capacity ×2-3 in covered functions; hiring concentrated on judgment and relationship roles	Revenue per employee up 30-50%; headcount stable, roles changed
What you have built	Baseline, first supervisors, first CoS register	Mixed org chart, dashboards, decision playbooks, evidence file	Proprietary data read by agents, redesigned processes, people fluent with agents

Two notes. The ranges in year one are the most reliable, because they are measured on one process against a baseline. The ranges in year three depend on you: on whether the second and third steps were taken, and on how many functions were covered. And the cost line is not a headcount line: Gartner found no correlation between workforce cuts and returns from autonomous technologies. The companies that get the numbers above reassign people to judgment, relationship and supervision; the ones that only cut get budget room and nothing else.

Data point. Only 6% of companies attribute a meaningful share of profit to AI, and 73% of them redesigned their workflows fundamentally, against 25% of the others. Companies with strong foundations — data, processes, governance — are three times more likely to report meaningful returns. Sources: McKinsey, The State of AI 2026; PwC, 29th Global CEO Survey 2026.

The assets nobody can buy

In three years every competitor will have the same models at the same price. The advantage will be in three things that accumulate only with time and use, and that this book has been about from the first page.

Redesigned processes. Every agent placed with CoS forces a process to be described, measured and simplified. After two years you have the only complete map of how your company works — and a company that knows how it works changes faster than one that does not.

Proprietary data read by agents. Customer behavior, quotes, complaints, machine signals, supplier performance: data you always had and never read, now read every day and feeding decisions. When models learn from the work they do inside companies — the missing capability from chapter 1 — this is what they will learn from.

People fluent with agents. Supervisors who know how to write a CoS, read a dashboard, declare a breakdown and stop an agent. This skill does not exist on the market in the numbers you will need; you grow it, one step at a time, or you do not have it.

Your first ninety days

The plan in steps (chapter 5) says what to do; this says when. It is the calendar I use for the first quarter with a new client.

Weeks	What happens	Who	Output
1-2	Inventory of AI already in use (chapter 12); baseline of hours, costs and errors on the five processes that absorb most time; the three numbers as of today	CEO, finance, function heads, consultant	Inventory; baseline sheet; the written lines (what we never do)
3-4	Choice of the first agent: routine, data-rich, low-risk, result visible in weeks; CoS negotiated and written; supervisor named; classification gate passed	CEO, one function head, consultant, legal	Structured task with CoS; named supervisor; monthly review date
5-8	Agent designed, tested on past data, launched with thresholds; supervisor trained on the job; weekly check of the CoS	Function head, supervisor, consultant	First live agent; logs; first weekly readings
9-12	First monthly review against the baseline; breakdowns analyzed; decision on step two (first AI Team) with CoS for the function; second agent chosen	CEO, function head, finance, consultant	Result versus baseline; decision to climb, slow down or stop; plan for the next quarter

Ninety days is enough for one proof. It is not enough for a transformation, and anyone who promises one in a quarter is selling something else.

What the CEO does, delegates and verifies

Chapter 12 drew this table for the rules. This is the same table for the whole program: the one page to keep on your desk.

Step	The CEO does	The CEO delegates to	The CEO verifies with
0. Audit	Decides to start; names the accountable person; writes the lines	Finance (baseline), consultant (audit), function heads (process maps)	Baseline sheet; inventory; the three numbers dated and signed
1. First agent	Chooses the function; signs the CoS; sits in the first review	Function head (CoS and supervisor), consultant (design), legal (gate 1)	Monthly CoS review; result versus baseline
2. First AI Team	Sets the function's CoS at 3-6-12 months; approves budget and cap	Function head (chief's CoS), HR (roles and training), finance (cost per work unit)	Function dashboard; return per agent; supervisor feedback
3. Mixed organization	Presents the plan to the company; decides which functions and in what order; renegotiates suppliers	Executives (their functions), HR (reskilling), consultant (governance), lawyer (contracts and countries)	Margin; capacity; revenue per employee; evidence file; quarterly review
Always	Uses AI visibly; keeps the three decisions of chapter 12; reads the three numbers monthly	Everything that is execution	CoS — never activity reports

Your Starter CoS Library

Thirty Conditions of Satisfaction, ten domains, each in the full form: what must be true, how much, by when, verified how, what if not. The thresholds are starting points from real projects; yours come from your baseline. Use them as templates for the negotiation with your function heads and your consultant, not as targets to impose. An updated and downloadable version lives in the online edition.

#	Domain	What must be true	How much	By when	Verified how	If not
1	Customer service	First response to routine requests	Under 2 min on 80% of requests	4 consecutive weeks	Support dashboard	Escalation rate reviewed; thresholds renegotiated
2	Customer service	Escalation to a person	Under 20% of requests, with reason recorded	Monthly	Support dashboard	Agent scope narrowed
3	Customer service	Customer satisfaction on agent-handled requests	Above 4.5/5	Monthly	Post-contact survey	Agent paused on the failing category
4	Sales	Lead qualification and handover	95% of leads answered within 2 min, 24/7	Monthly	CRM	Routing rules reviewed
5	Sales	Qualified leads that close	At least 30% within 60 days	Quarterly	CRM	Qualification criteria renegotiated with sales
6	Sales	Technical quote turnaround	Within 24 working hours on 90% of requests	Monthly	Quote log	Bottleneck identified; agent or person reassigned
7	Marketing	Recovery of abandoned carts and open quotes	At least 12% recovered within 72 hours, incentive under 5%	Monthly	Sales platform	Message sequence and incentive reviewed
8	Marketing	Revenue from existing customers	+20% in 6 months, unsubscribes under 0.3% per send	6 months	CRM and email platform	Frequency rules tightened
9	Marketing	Product information for AI assistants	100% of catalog complete and consistent across channels	Monthly	Catalog audit	Data owner named; agent re-scoped
10	Operations	Unplanned downtime	−30% in 12 months, false alarms under 10%	Quarterly	Maintenance log	Sensor data and thresholds reviewed
11	Operations	New products or services tested	At least 6 market tests a year, cycle under 8 weeks	Quarterly	Product committee	Process from idea to test redesigned
12	Operations	Outgoing defects	−25%, every complaint with root cause within 5 days	Monthly	Quality system	Quality agent scope reviewed
13	Supply	Capital tied up in stock	−15% with stock-outs under 2%	Monthly	ERP	Reorder rules and lead times reviewed
14	Supply	Supply-chain exceptions resolved	Within 2 days on 90% of cases	Monthly	Exception log	Escalation path to buyer reviewed
15	Supply	Supplier delivery performance visible	100% of orders tracked, delays flagged within 24 hours	Monthly	Purchasing dashboard	Data feed from suppliers renegotiated
16	Finance & admin	Documents handled without human intervention	80%, error rate under 0.5%	Monthly	Finance	Exceptions categorized; rules added
17	Finance & admin	Monthly closing	Within 5 working days	Monthly	Finance	Bottleneck process assigned to an agent
18	Finance & admin	Cost per agent and per work unit	Reported monthly, within cap, with trend	Monthly	Finance dashboard	Cap enforced; agent paused above cap
19	Product	Customer feedback read and classified	100% of reviews, returns and complaints, weekly summary	Weekly	Product dashboard	Sources connected; owner named
20	Product	Personalized proposal or configuration	Delivered in hours, not days, on 90% of requests	Monthly	Quote log	Configuration rules completed
21	Product	Multi-language support	Requests handled in 5 languages with under 24-hour turnaround	Quarterly	Support and sales logs	Language scope reduced or agent retrained
22	Organization & people	Supervisors trained and active	One named supervisor per agent, trained before launch	Per launch	HR training record	Launch postponed
23	Organization & people	Function head's time	70% on strategy and governance, 30% on routine, within 12 months	Quarterly	Self-assessment and calendar review	Routine tasks reassigned to agents
24	Organization & people	Reskilling paths	Every operator whose task moves to an agent has a written path within 30 days	Per agent	HR	No further agent in that function until done
25	Compliance	Transparency notices	100% of customer-facing agents disclose AI; AI content labeled where required	Before launch and quarterly	Legal review	Agent paused
26	Compliance	Human on the loop	Named person, thresholds written, stop tested, for every agent	Before launch	Evidence file	Launch postponed
27	Compliance	Evidence file	Complete and findable by anyone in 10 minutes	Quarterly	Internal audit	Accountable person closes the gaps in 30 days
28	Cost governance	Monthly AI budget with automatic block	Cap set per agent and per function, 100% of agents	Before launch and monthly	Finance dashboard	Agent without cap not launched
29	Cost governance	Return per agent	Positive within 6 months of launch on cost, revenue or capacity	6 months	Finance review against baseline	Agent redesigned or retired
30	Cost governance	Supplier and model changes	Advance notice, re-test before change on 100% of agents	Per change	Change log	Contract clause enforced; supplier reviewed

What to take away

Expect real numbers in year one on one process, and the cost, revenue and capacity of the whole company to move in years two and three — if you take the steps, redesign the work, and keep people on judgment. The assets that last are the ones you build: processes, data, people. Ninety days is enough for the first proof; the master table tells you what is yours and what is not; the thirty CoS are the book turned into commitments. Riding does not mean running. It means having a plan you can accept, slow down or stop at every step — and now you have one.

Your to-do list.

Write today's three numbers — cost per work unit on your heaviest process, conversion, revenue per employee — with the date.

Pick five CoS from the library, adapt the thresholds, and put them in front of your function heads and your consultant.

Book the first monthly review now, ninety days out, with the baseline on the table.

Continue reading with your free reader account.

Frequently asked questions

How long will it take before I see real financial results from AI?

On the one process where you place the first agent, expect cost per work unit down 20-40% and conversion up 10-30% within year one — these are the most reliable numbers of the whole projection because they are measured against a real baseline. Structural, company-wide results take longer: two to three years, and only if you take the second and third steps rather than stopping after the first agent.

What should realistically happen in my company's first 90 days with AI?

Weeks 1-2: an inventory of AI already in use and a baseline of hours, costs and errors on the heaviest processes. Weeks 3-4: the first agent chosen, with Conditions of Satisfaction negotiated and a supervisor named. Weeks 5-8: the agent designed, tested and launched with thresholds. Weeks 9-12: the first monthly review against the baseline and a decision on step two.

If AI cuts costs, do I need to lay off staff to see the return?

No — and the data argues against it. Gartner found no correlation between workforce cuts and returns from autonomous technologies: companies that get the real numbers reassign people to judgment, relationship and supervision roles, while companies that only cut headcount get budget room and nothing else.

Is there a ready-made list of AI project goals I can start from?

Yes — a Starter CoS Library of thirty Conditions of Satisfaction across ten business domains (customer service, sales, marketing, operations, supply, finance, product, people, compliance, cost governance), each with what must be true, how much, by when, and how it is verified. Treat them as negotiation templates to adapt with your own function heads and consultant, not as targets to impose as-is.

Sources

Comments & reviews

No comments yet.