Three ways to run a programme

Microsoft 365 alone, with Copilot, or with atlanticma.ai

Four things matter to a programme: the hours it takes, the quality of the work, how fast benefits arrive, and whether the decisions hold.

	Microsoft 365 only	Microsoft 365 with Copilot	atlanticma.ai, our estimate
Hours time saved	The baseline. In one transformation, 65 days¹ a month went on recurring reporting and steering work.	2 to 7%^6,7,8 of each person's time, mostly email, search and summarising. No measured change in meetings or shared work.⁶	33% less recurring reporting and steering work, against 39%¹ in McKinsey's own case. 22% of all programme-office time.
Quality of the work, and of the judgement calls	Slowest, but people working unaided reached the right answer 84.5%² of the time on a hard judgement task.	Mixed. A general assistant raised rated quality 38 to 42.5%² on tasks within AI's ability, and cut right answers by 14 to 24 points² on a hard judgement task, where the wrong answers also looked better. Copilot itself did better on summaries and emails and worse on spreadsheets and slides.⁹	About 40%² better drafted work (plans, charters, reports), and built for 0 points lost on judgement calls, which stay with your people. They were right 84.5%² of the time unaided, against 60 to 71%² when they relied on AI. Every meeting proposal carries the quote it came from, and by default a named person approves it.
Speed of the whole programme	The plan is written by hand after close. With no synergy timeline, three-year share price growth was 21%⁵ or less, against 71%⁵ with a one-year target.	0% measured change in meetings or in how long documents take to finish.⁶ An individual drafts a single task 25%² faster, which shortens the task, not the programme.	Programmes 10 to 25%³ shorter. The first plan and charters are drafted in about 4 minutes (our own test). Each month gained is worth one twelfth of your annual benefits.
Decisions	About 1 in 3⁴ acquirers fully achieve their deal objectives.	No evidence of better team decisions. Each person asks their own assistant, from their own documents and their own prompts, so a team can bring several different answers to reconcile in the meeting. 40%¹⁰ of workers report being handed AI-made work that looked finished and was not, at about two hours a time to sort out. We believe that makes group decisions harder, not better.	From day one you work the way 59%⁴ of successful acquirers do, and only 28%⁴ of the rest: owners, dates and a decision record, on one shared record. We believe stakeholder profiles and the modelling of decisions improve group dynamics and decision-making further.

Figures in colour are the change against the Microsoft 365 baseline, and each row compares the same thing across the three columns. The quality trial used a general AI assistant, not Copilot itself. Figures in the right-hand column are our estimates, drawn from the studies marked.

The four lines

What is published, what the application does, and our estimate

1. Hours

Published: One transformation in which agents cut recurring reporting and steering work from 65 to 39.5 days a month, a saving of 39%.¹ Copilot users saved 2 to 7%^6,7,8 of their own time, mostly on email, search and summarising, with no measured change in meetings or shared work.⁶
The application: Does or helps with 8 of the 10 activities in that case.
Our estimate: 33% of recurring reporting and steering work, and 22% of all programme-office time. The workings are in the next section.

2. Quality

Published: A randomised trial of 758 consultants using a general AI assistant, not Copilot itself. With AI, work it suited was rated 38 to 42.5% better. On a hard judgement task it did not suit, right answers fell from 84.5% to 60 to 71%, a loss of 14 to 24 points, and the wrong answers looked better.²
The application: Drafts plans, charters and reports. Shows the source quote for each proposed change, checks it against the transcript, by default asks a named person to accept or reject, and logs every change.
Our estimate: About 40%² better drafted work (plans, charters, reports), and built for 0 points lost on judgement calls, which stay with your people.

3. Speed

Published: One bank prepared its integration about 25% faster. One company modelled synergies in about 2 months where the usual route takes over 12.³ Acquirers that targeted synergies within a year saw 71% share price growth over three years, against 21% or less with no timeline.⁵ A general assistant helped an individual finish a set task 25%² faster, which shortens the task, not the programme. With Copilot there was no measured change in meetings or in how long documents take to finish.⁶
The application: Drafts the strategic direction, a charter for each workstream and a dated work plan from your documents. On one test project with eight workstreams this took about 4 minutes.
Our estimate: Programmes 10 to 25% shorter. The 25% is Bain's³ figure for preparation, which we apply to the whole programme. Each month gained is worth one twelfth of your annual benefits.

4. Decisions

Published: Of 530 executives surveyed, about 1 in 3 said they fully achieved their deal objectives. 59% of those had turned value drivers into initiatives with owners and timelines before integration began, against 28% of the rest.⁴ In a separate survey, 40%¹⁰ of desk workers said they had been handed AI-made work that looked finished and was not, at about two hours a time to sort out.
The application: Gives tasks an accountable owner and a date in the first plan, keeps a decision register, and tracks benefits against a locked baseline. It builds a profile of each key stakeholder and models how a decision will land before it is taken (stakeholder simulation, in beta).
Our view: From day one you work the way 59%⁴ of successful acquirers do, and only 28%⁴ of the rest: owners, dates and a decision record, on one shared record. With a general assistant, each person works from their own documents and their own prompts, so a team can bring several different answers to reconcile in the meeting. We believe that makes group decisions harder, not better. We believe stakeholder profiles and the modelling of decisions improve group dynamics and decision-making further.

Workings

How we reach the hours estimate

McKinsey's case¹ lists ten recurring activities. Against each we show whether the application does it, helps with it, or does not cover it. Where it does or helps, we count the saving McKinsey reported. Where it does not, we count nothing.

Activity	Days a month before	Days saved in McKinsey's case	atlanticma.ai	Days we count
Biweekly and steering committee reporting	12	9	Does it	9
Monthly steering committee preparation	4	1	Helps	1
Other reporting, for example a supervisory board	4	3	Helps	3
Monthly financial reporting	5	2	Helps	2
Performance against plan	5	2	Does it	2
Monthly headcount reporting	10	4	Not covered	0
Status updates and follow-ups	8	2	Helps	2
Checking the quality of initiative data	5	0	Not covered	0
Recalibrating targets	10	2	Helps	2
Valuing initiatives to a common guide	2	0.5	Helps	0.5
Total	65	25.5 (39%)		21.5 (33%)

Across a whole programme office, including planning, meetings and analysis, our estimate is 22% of all time. That rests on our own view of how an office divides its time, and counts nothing for meetings or analysis.

Your numbers

Put the case into your own figures

Choose your kind of programme, enter the numbers you already know, and choose which evidence to apply. The result is an illustration from your inputs and our best estimates above. Use your own currency.

Programme office or integration support, a yearExample figure. One 3,000-person turnaround needed more than 15 full-time staff in its transformation office.¹

Annual benefits in your business caseExample figure. Stronger performers target 75% or more of annual earnings; weaker ones 25% or less.¹⁶

Programme length, in monthsExample figure. No published benchmark.

Hours saved16%. Strict reading: activities the application only helps with count at half.22%. Our estimate for a whole programme office.33%. Our reporting and steering figure¹, applied to the whole office.Your own figure: %

Programme shortened by10%. The low end of our estimate.25%. Bain's measured figure for integration preparation³, applied to the whole programme.Your own figure: %

	Microsoft 365 only	With Copilot	atlanticma.ai
Hours, value a year	0	48,000 at 4%	264,000 at 22%
Benefits brought forward	0	0 0% measured change	1,800,000 1.8 months earlier. Each month is worth one twelfth of annual benefits.

Quality and decisions are not turned into money here.

Return on the licence, across your projects

Projects you run at the same time, on averageA whole number. Count the programmes live at the same time, not the ones you start in a year. Each is taken to be like the one above.

	One programme	Your 3 projects
Hours saved over the programme	356,400 16.2 months at 264,000 a year	1,069,200
Benefits brought forward	1,800,000	5,400,000
Total benefit	2,156,400	6,469,200
atlanticma.ai Enterprise, over the programme 3 projects × $12,000 $36,000 Enterprise pack (25%, minimum $20,000) $20,000	25,200 18,667 a project a year, for 16.2 months	75,600 56,000 a year, for 16.2 months
Return		86 times the licence Net benefit 6,393,600

Priced on Enterprise, our highest price. Independent and Firm cost less, so their return is higher. The price follows the Enterprise formula: $12,000 a year for each project live at the same time, three projects minimum, plus the Enterprise pack of 25% of project fees, minimum $20,000 a year. US dollars. The programme length is the shortened one: your months, less the shortening you chose above. Hours and the licence are both counted over that length.

What we leave out: add-on modules, such as the Earnout Ledger (in beta) on M&A projects, and usage beyond the limits set for each project. Set-up, support and custom development are separate. Quality and decisions add to the return and are not counted.

The studies

What each one measured

McKinsey, "The agentic transformation office", September 2026. One transformation, using McKinsey's own tools. McKinsey estimates agents can cut time on many core tasks by 35 to 40%. Read it
Harvard Business School and BCG, "Navigating the Jagged Technological Frontier", 2023. A randomised trial of 758 consultants working alone with a general AI assistant on set tasks. Read it
Bain, M&A Report 2026. Client cases: a bank whose integration preparation was about 25% faster, and a company that modelled synergies in about two months. Read it
PwC, 2026 M&A Integration Survey. 530 senior executives at companies with revenue of $1 billion or more. Self-reported. Read it
PwC, "The M&A synergies credibility gap". 180 US public deals of $1 billion or more, announced 2016 to 2024. A correlation, as PwC states. Read it
Microsoft Research, "Shifting Work Patterns with Generative AI", 2025. A randomised trial of 7,137 workers at 66 firms. Regular Copilot users spent about two hours a week less on email; meeting time did not change. Read it
UK government Copilot evaluations, 2025 and 2026. Self-reported savings of 19 minutes a day at one department and 26 across twelve. Department for Work and Pensions, cross-government
Humlum and Vestergaard, "Large Language Models, Small Labor Market Effects", 2025. About 25,000 Danish workers. Users of AI assistants saved 2.8% of their working hours on average. Read it
UK Department for Business and Trade, Microsoft 365 Copilot evaluation, 2025. 1,000 licences. In a small observed test, Copilot users did better on summaries and emails and worse on spreadsheet analysis and slides. The department found no robust evidence that time saved improved productivity. Read it
BetterUp Labs and Stanford Social Media Lab, survey on AI-made work, 2025. About 1,150 US desk workers, self-reported. 40% had received AI-made work in the past month that looked polished but lacked substance, and spent nearly two hours dealing with each instance. Read it
L.E.K. Consulting, "Keys to unlocking post-merger integration value: cost synergies", 2024. Public deals with announced synergies, 2012 to 2022. Median acquirers aim for cost synergies of 6 to 10% of the target's revenue. Read it
BCG, "The Real Deal on M&A, Synergies, and Value", 2016. 286 North American deals over $500 million, 2010 to 2015. Synergies are expected within 12 to 18 months. Read it
PrivateEquityInfo, "What a decade of PE holding periods reveals", 2026. US platform exits. Median hold of 5.9 years for 2025 exits. Read it
BD Emerson, transition services agreement benchmark, 2026. 39 separations by SEC-registered companies, 2020 to 2026. Median term of 24 months. Read it
EY, "Four current trends in estimating M&A integration costs", 2019. More than 70 US deals over $1 billion, 2010 to 2016. Integration cost averaged 3% of deal value, with a range of 1 to 7%. Read it
McKinsey, "The numbers behind successful transformations", 2019. 82 listed companies. Stronger performers set targets at 75% or more of trailing earnings; weaker ones at 25% or less. Read it
National Center for the Middle Market. Defines the US middle market as companies with annual revenue of $10 million to $1 billion. Read it
Deloitte, Global Divestiture Survey, 2026. 1,500 executives, surveyed in October 2025. Self-reported. Sellers rank stranded cost as their leading challenge after the sale, and only 11% deal with it in full. Read it
McKinsey, "Profitably parting ways: Getting more value from divestitures", 2013. McKinsey's experience with sellers; no sample is given. Sellers often take up to three years to recover from a divestiture. Read it

How to read this page

Published figures and our estimates

A figure with a small number beside it comes from the study of that number. It is that study's result.
A figure marked as our estimate is our view of what atlanticma.ai is built to deliver, drawn from those studies and from what the application does.
Our estimates are best estimates. They have not yet been measured on a customer's programme, and your results will depend on your programme.