resources

An AI transformation roadmap for a 300–2,000-person company (12 months)

Four quarters, one workflow at a time. The deliverables, the owners and the failure mode of each quarter, written for mid-market companies that cannot afford an enterprise timeline.

September 12, 2026

12 min read

BEFORE THE ROADMAP

THE FOUR QUARTERS

WHAT TO MEASURE

An AI transformation roadmap for a 300–2,000-person company (12 months) - cover image

TL;DR

Mid-market companies adopt AI as fast as anyone. RSM’s 2025 Middle Market AI Survey found 91% using generative AI, but only 25% had it fully integrated into core operations, and of the 79% with a roadmap, only 37% called that roadmap well-formulated.

Deloitte’s ROI research puts a typical AI use case at two to four years to satisfactory return, against the seven to twelve months executives expect. A twelve-month roadmap closes that gap through sequence, not speed.

The four quarters: foundation (owners, definitions, one system of record per object), one production workflow with a baseline, two or three more on the same data layer, then an operating model the company runs without outside help.

Each quarter has a way of going wrong that is visible in advance: the readiness score nobody acts on, the workflow that ships without a baseline, the third use case that rebuilds its own pipeline, and the governance cadence that lapses when the sponsor moves on.

Most companies of this size already have a roadmap. That is the problem.

RSM’s 2025 Middle Market AI Survey asked 966 executives at mid-sized companies about generative AI. Ninety-one percent said they use it. Seventy-nine percent said they have a roadmap for it. Then the survey asked whether the roadmap was well-formulated, and the number fell to 37%. Only a quarter of respondents described AI as fully integrated into core operations. The typical mid-market company, in other words, has a document called a roadmap, a set of tools in use, and a gap between the two that nobody has sequenced.

Two more numbers from the same survey explain why. Seventy percent of respondents said they need external support to move forward, and 39% said they lack the in-house expertise to do it themselves. That matches what we see: a 600-person software company with one Salesforce admin, a RevOps lead who also runs the HubSpot instance, and a CFO who signed the AI license because the board asked about it. There is no platform team to absorb a bad sequence. Whatever order the work happens in is the order the company lives with.

So this is a roadmap for that company. A 300–2,000-person technology business, Salesforce or HubSpot as the revenue system of record, a small operations team, twelve months. The quarters below are specific about deliverables and owners because vague roadmaps are the kind RSM’s respondents already have. Adjust the scope of each quarter to your size. Do not adjust the order.

Why twelve months, and why the enterprise playbook does not transfer

Deloitte’s research on AI returns, drawn from 1,854 executives across EMEA in 2025, found that a typical AI use case takes two to four years to reach satisfactory ROI. The same executives expected seven to twelve months. Only 6% reported payback inside a year. KPMG’s Q4 2025 AI Pulse survey of large enterprises tells the same story from the top of the market: 59% of companies with more than $1 billion in revenue expect ROI from agents within twelve months, while only 26% have actually deployed any.

An enterprise can carry a three-year payback on a portfolio of initiatives. Gartner’s September 2026 C-suite survey found that organizations tracking AI ROI as a portfolio see positive returns on 81% of initiatives, which is a sound strategy when you have forty initiatives and a chief data and analytics officer to run them. A mid-market company has three or four, one budget cycle, and a sponsor whose patience is measured in quarters. The enterprise playbook, with its center of excellence, its multi-year data platform and its parallel pilots, assumes slack that does not exist at 600 people.

The mid-market advantage runs the other way. MuleSoft’s 2026 Connectivity Benchmark counts 957 applications in the average enterprise, 27% of them integrated. A mid-market company might run sixty, and the ones that matter for revenue fit on one whiteboard. The foundation that takes an enterprise two years can take a focused team one quarter, provided the decisions that gate it are made first and made once. That is the bet this roadmap makes, and it is why it starts with decisions rather than technology. The four quarters are laid out in the timeline that follows.

Diagram: Four quarters, one workflow at a time

The four quarters

Quarter one: decide, then clean

The first quarter buys nothing. It opens with three decisions the executive team makes in a room, in writing. One accountable owner for the transformation, by name, with the target metric in their goals. One flagship workflow, chosen because the board already tracks its number: lead-to-opportunity conversion, first-response time on support cases, forecast accuracy. And one system of record per business object, declared. The customer lives in Salesforce. Invoices live in billing. The spreadsheet in customer success is retired or becomes a report.

With those made, the quarter becomes hygiene with a purpose. Run the readiness assessment and score it honestly. Write the definition set: lead, opportunity, customer, churn, revenue, one version each, used by every dashboard. Stand up one governed report on the flagship metric so a number can be traced from a board slide back to the records that produced it; this is the baseline the whole year will be measured against. Inventory the tool estate and retire what nobody owns. Salesforce Ben’s 2026 admin survey found only 2% of admins describe their org as clean and well-maintained, so budget real hours for the unused custom fields, the picklist with three spellings of the same value and the integration user nobody can name.

Deliverables: readiness scorecard, ownership map, written definition set, one defensible dashboard with the baseline recorded, pruned tool inventory.
Owners: the executive sponsor for the three decisions; the head of business systems or RevOps for the assessment and definitions; the CFO for the baseline number.
What goes wrong here: the assessment is run, scored and filed. Nobody acts on the failures, because acting on them means a conversation between marketing and finance about what a lead is. A quarter-one scorecard with no name attached to each red item is the first sign the year will end in pilots.

Quarter two: one workflow in production, with a number in front of it

Quarter two ships exactly one thing, and ships it to production rather than to a pilot. The distinction is operational. A pilot runs on hand-prepared data and reports to a sponsor. A production workflow runs on the system of record, crosses every system it touches through a maintained integration, runs under permissions designed for the agent, and has a baseline recorded before go-live and a named person who can switch it off.

For a Salesforce shop the flagship is usually a service or sales agent on Agentforce; for HubSpot, one of the Breeze agents pointed at the same object the quarter-one dashboard measures. Either way the work is the same. Document the decision rules a human applies today (when to escalate, what never gets promised, which fields get updated), write them as test cases, map the agent user’s permissions to a human in the same role and no further, and connect the workflow end to end. Deloitte’s 2026 State of AI in the Enterprise survey of 3,235 leaders found only 21% of organizations have mature governance for autonomous agents. That is the quarter-two deliverable most companies skip, and the one that decides whether the workflow is still running in quarter four.

Deliverables: one AI workflow in production; documented decision rules and test cases; agent permission design; integrations across every system the workflow crosses; a before-and-after number on the flagship metric; a rollback owner and a review cadence.
Owners: the workflow’s business owner (head of support, head of sales development, whoever holds the metric) for scope and sign-off; the Salesforce or HubSpot lead for the build; the executive sponsor for the go/no-go.
What goes wrong here: the workflow goes live without the baseline, because the dashboard from quarter one was never finished, and the team spends quarter three arguing about whether it worked. The second failure is quieter. The agent ships on a spreadsheet extract because the billing integration was hard, and the workflow now has a human bridge that breaks the first weekend the analyst is away.

Quarter three: scale on the same foundation, or stop

Quarter three adds two or three workflows. The number matters less than the rule that governs it: each new workflow reuses quarter one’s data layer and quarter two’s governance pattern. Same definitions, same system of record, same permission model, same review cadence, same test-case format. The measure of the quarter is cost per additional use case, and it must fall. If the third workflow needs its own data pipeline, its own definition of a customer or its own integration user, the roadmap has drifted back into piloting, and the right move is to stop and repair the foundation before continuing.

This is the quarter where the mid-market advantage shows. Accenture’s 2025 study of 2,000 companies found that firms scaling one strategic bet were roughly three times as likely to beat their ROI expectations as firms spreading effort across many. BCG’s October 2024 survey of 1,000 executives reached a similar conclusion about AI leaders: they pursue half as many opportunities and scale twice as many. A company of this size cannot run eight pilots. It can run three workflows that share everything, and that turns out to be the stronger position.

Deliverables: two or three additional workflows in production on the shared foundation; a cost-per-use-case figure with the trend visible; a light platform rhythm, such as a weekly review across workflow owners, that replaces the project-by-project approach.
Owners: each workflow’s business owner; the head of business systems as keeper of the shared layer, with the authority to say no to a workflow that wants its own pipeline.
What goes wrong here: enthusiasm. The first workflow worked, three departments want their own, and the second one is built by a different vendor on a different data model because it was faster. By quarter four there are two foundations, and the company is back among the mid-market firms RSM counted as integrated only in parts.

Quarter four: turn it into how the company runs

The last quarter converts a program into an operating model. The governance rhythm gets written down: who reviews what the agents did, how often, who can switch any of them off, and what happens to a workflow whose metric drifts. Role definitions are updated to assume AI in the loop, which is a real change for a support lead whose team now supervises resolutions rather than performing them. A quarterly readiness review goes on the calendar. And a ranked pipeline of next use cases is built, scored on metric and readiness rather than on which executive asked loudest.

The IBM Institute for Business Value’s 2026 CEO study found 83% of CEOs believe AI success depends more on adoption than on the technology, and Gartner’s May 2026 research found only 20% of executives believe their workforce is AI-ready. Quarter four is where that gap is closed or ignored. It is also where consumption pricing turns from a line item into a management topic. With Agentforce metering a standard action at 20 Flex Credits and HubSpot charging per resolved conversation, the operating model needs someone watching the bill against the outcome every month.

Deliverables: a governance charter; updated role definitions for the teams that work alongside agents; a quarterly readiness review on the calendar; a ranked use-case pipeline; a monthly cost-versus-outcome report per workflow.
Owners: the executive sponsor for the charter; HR and the functional leads for roles; the CFO for the cost report. By the end of the quarter the external help, if there was any, should have handed over.
What goes wrong here: the sponsor is promoted, leaves or moves on, and the cadence lapses within two review cycles. Gartner’s June 2025 AI maturity research found 45% of high-maturity organizations keep AI projects in production for three or more years, against 20% of low-maturity ones. The difference is almost entirely whether the rhythm survived a change of hands.

Underneath all four quarters sit the three decisions from quarter one. They are restated in the card below because in practice they are the thing that gets renegotiated in month seven, when a new executive arrives and asks why marketing does not own the customer record.

Diagram: Three decisions that cost nothing and decide everything

What we actually see

A pattern from engagements, offered as observation rather than a controlled study. The company has usually done a version of quarter two already, out of order: a service agent or a lead-scoring pilot shipped in a few weeks, demoed to the board, and now running on a data extract one person refreshes. It works. It has no baseline, because nobody recorded the metric before go-live, so the question of whether it worked is answered with anecdotes. When the team tries to add a second workflow, it discovers the first one’s definitions were local to the support team, and the sales team uses different ones.

What happens next is the useful part. The company goes back and does quarter one in about six weeks, because the decisions were always available and only needed a room. The existing pilot is re-pointed at the system of record and given a baseline reconstructed from historical data, with an honest caveat about its quality. From there the roadmap runs roughly as written. Total elapsed time is closer to fifteen months than twelve, and the extra three are the cost of having started with the tool. We have not seen a company of this size skip quarter one and reach quarter three intact. We also have not seen every company.

What to measure, and what to ignore

Five numbers, reviewed monthly by the sponsor and the CFO. Everything else is reporting for its own sake.

The flagship metric, before and after, from the governed dashboard. This is the only number the board needs.
Readiness score by quarter, from the same assessment run in quarter one, so improvement is visible and regression is caught early.
Cost per additional use case, from quarter three onward. It should fall. If it rises, the foundation is being rebuilt somewhere.
Share of AI workflows in production versus in pilot. Deloitte found only 25% of organizations have moved 40% or more of their pilots into production; a company on this roadmap should be well past that by quarter three.
Cost versus outcome per workflow, monthly, once consumption pricing is live.

Ignore prompts per employee, tools licensed, pilots launched and hours of training delivered. They measure activity. Gartner’s May 2026 survey found 88% of employees with enterprise AI access also use personal AI tools, which is a reminder that usage will happen with or without a roadmap. What will not happen by itself is a workflow, in production, with a number attached, that survives the person who built it.

PwC’s 29th Global CEO Survey, published in January 2026, found that only 12% of 4,454 CEOs say AI has delivered both cost and revenue benefits, and that companies with strong foundations were three times as likely to report returns. Twelve months of decisions, one workflow at a time, is how a mid-market company gets into that 12%. It is unheroic by design. It is also the only version we have seen survive the second year.

Frequently asked questions

What is an AI transformation roadmap?

A sequenced plan that takes a company from scattered AI usage to AI embedded in governed core workflows, with named owners and measured outcomes, over a defined period. For a 300–2,000-person company that period is typically twelve months. The sequence is what makes it a roadmap rather than a list of pilots: decisions and foundation first, one production workflow second, scale third, operating model last.

How long does AI transformation take for a mid-sized company?

Twelve months to a working operating model, if the sequence is respected. Deloitte finds a typical AI use case takes two to four years to reach satisfactory ROI, largely because companies start from the tool and rebuild the foundation afterward. A mid-market company that makes the ownership, definition and system-of-record decisions in the first quarter usually has a measurable result by the second.

What should be in the first quarter of an AI roadmap?

Three decisions and a cleanup. Name one accountable owner with the metric in their goals, choose one flagship workflow the board already tracks, and declare one system of record per business object. Then run a readiness assessment, write the definition set, build one governed dashboard with the baseline recorded, and prune the tools nobody owns. No AI is purchased in quarter one.

Who should own the AI transformation in a mid-market company?

One executive whose goals contain the business metric the AI is meant to move, usually the COO, CRO or head of business systems. IBM reports 76% of organizations now have a Chief AI Officer, but the title matters less than the fact that a name is attached to an outcome and that person can refuse a workflow that wants its own data pipeline.

How is a mid-market AI roadmap different from an enterprise one?

Less slack and a shorter foundation. An enterprise can run AI as a portfolio, absorb a multi-year payback and staff a center of excellence. A mid-market company has one budget cycle and a handful of systems, so it must sequence strictly and scale one workflow at a time. The compensation is speed: the foundation that takes an enterprise two years can take a focused mid-market team one quarter.

How do you measure whether an AI roadmap is working?

Five numbers: the flagship business metric before and after, readiness score by quarter, cost per additional use case, the share of AI workflows in production versus in pilot, and monthly cost against outcome per workflow. Ignore activity measures such as prompts per employee or pilots launched. If cost per use case is rising, the foundation is being rebuilt somewhere and the roadmap has slipped back into piloting.

NEXT STEP

See where your organization stands before the next AI purchase.

A short conversation is enough to tell whether the AI Readiness Health Check, 36 checks on your Salesforce or HubSpot org, is the right first step. No decks, no obligations.

book a strategy session

Read the 36 checks →

WRITTEN BY

Tal Oryon

MORE NOTES

Perspective

The world is changing. Is your organization part of it?

In one year the platforms your company runs on rebuilt their pricing, their products and their partnerships around agents. A note to executives on the one decision that is still theirs.

AI transformation