You're Behind on AI. Here's How to Start.

A leader I'm working with came back from an industry event a few weeks ago genuinely rattled. Every peer organization in the room was talking about what they'd done with AI — the pilots, the productivity numbers, the roadmaps — and their company had nothing to show. No strategy, no pilot, not even a sanctioned tool. They felt embarrassingly behind, and they wanted to fix it immediately.

I hear a version of this constantly, and the panic it produces is dangerous, because it pushes leaders toward the worst possible first move: buy everything, announce a big initiative, and launch a dozen pilots at once. That's how you join the large group of companies that adopted AI and got almost nothing for it. Here's the more reassuring truth — starting late, done deliberately, is closer to an advantage than a liability.

Being behind is not the problem you think it is

The headline numbers make it sound like you've missed the boat: 78% of organizations now use AI in at least one business function (WRITER). But look at the second number and the panic should ease — only around 29% of organizations report significant ROI from generative AI (WRITER). Most companies that "have AI" are in roughly the same place you are, just with more sunk cost and more shelfware to clean up.

That gap is the opportunity. The companies ahead of you mostly got there by scattering effort — everyone got a chatbot, nobody measured anything, and the productivity stayed theoretical. You get to skip that phase entirely and go straight to the disciplined version. Late and deliberate beats early and aimless.

Start with one workflow, not a strategy

The single most important decision is to resist doing everything. Organizations that try to transform five workflows at once usually fail to show measurable results in any of them; the ones that pick a single workflow, prove value, and expand from there are the ones that compound their gains. So your first move isn't a strategy document — it's choosing one workflow.

A good first workflow has four traits. Pick where they overlap:

• High friction — a task people find slow, tedious, or error-prone today. The pain makes the win obvious.
• High frequency — it happens often, so you accumulate evidence fast and the improvement actually adds up.
• A clear owner — one person accountable for the outcome and able to change how the work gets done.
• A measurable outcome — there's already a number attached: time, cost, error rate, or turnaround.

For most organizations the safe, high-value starting points look similar: drafting and summarizing routine documents, triaging and routing incoming tickets or requests, searching internal knowledge that's currently trapped in people's heads, reviewing work for errors before it ships, or reconciling records. Notice these are all internal and low-stakes enough to fail safely, but frequent enough to matter. Don't start with the customer-facing, regulated, or irreversible work — you earn that later.

Measure the before, or you'll never prove the after

This is the step the embarrassed-and-in-a-hurry crowd skips, and it's the one that determines whether you can defend the work in three months. Before you introduce any tool, write down the current numbers for your chosen workflow. How long does it take? What does it cost per unit? How often does it have to be redone? What's the turnaround people actually experience?

I've argued before that your AI pilot usually isn't failing — your scoreboard is. The fix is to baseline first. If you can't state where you started, every later claim about improvement is just a feeling, and feelings lose budget arguments. A workflow-level number measured against a defined baseline for a quarter is the most defensible signal you can bring to leadership (CIO).

Run a small, owned, time-boxed pilot

Now introduce AI into that one workflow, with one team, under a named owner, for a fixed window — a quarter is plenty. Keep the guardrails simple and real: a sanctioned tool (not a pile of personal subscriptions), a clear rule about what data can and can't go in, and a human reviewing the output. You don't need a governance bureaucracy to start, but you do need those three things from day one.

Resist scaling early. The small blast radius is a feature — it's where you learn what breaks before it breaks in front of customers.

Measure outputs, speed, and quality — together, never one alone

The whole point is to improve outputs, increase speed, and raise quality, so measure all three at once. This matters because they trade against each other, and watching only one hides the cost on the others. A small scorecard — five to eight metrics — is plenty (CIO). For a first workflow:

• Speed — cycle time from start to finished, AI-assisted versus the baseline. The headline win.
• Output — volume completed per person per week. Are you actually getting more through the door?
• Quality — error or rework rate: the share of work that has to be corrected or redone. This is the metric teams skip, and the one that catches the failure mode where speed just ships mistakes faster.
• Cost — what the AI usage actually costs, tracked against the time or money saved.
• Adoption depth — are people using it daily, or did it get tried once and quietly abandoned?

Track quality with the same seriousness as speed. AI can make a team look dramatically faster while quietly increasing defects; if you only watch the speedometer, you'll celebrate right up until the rework lands. Speed that costs you quality isn't a win — it's a deferred bill.

Done right, the gains are real and large. Roles redesigned around AI tend to see roughly a 37% productivity improvement, versus about 12% from traditional automation alone (Azumo), and organizations that actually reach production report multiples on their investment (Azumo). But those are the numbers for teams that measured and redesigned the work — not the ones that just handed out logins.

Prove it, then template it

When the quarter's numbers are good — faster, more, and no worse on quality — you do two things. You declare the win with evidence, not vibes. And you write down how you did it: the workflow you picked, the guardrails, the tool, the baseline, the result. That playbook is the real asset. Your second workflow should be faster to AI-enable than your first, and your third faster still, because you're no longer starting from zero — you're running a process.

A healthy place to be a year from now isn't "AI everywhere." It's two to four workflows genuinely redesigned around AI, each with a named owner, a success metric, and a quarter of real data behind it. That's a defensible, compounding foundation — and it's well ahead of the peers who scattered.

Your first 90 days

• Days 1–30: Pick the workflow and baseline it. Choose the one high-friction, high-frequency workflow with a real owner, and measure its current speed, cost, and quality before touching anything.
• Days 31–60: Stand up a sanctioned tool and run the pilot. One team, one owner, simple guardrails — approved tool, data rules, human review. Start logging the same metrics you baselined.
• Days 61–90: Compare, decide, and document. Put baseline versus result in front of leadership, decide whether to scale, and write the playbook for the next workflow.

Ninety days from a standing start gets you something many of those conference peers may not actually have: one workflow measurably better, with the receipts to prove it.

The takeaway

If you're starting from nothing, the worst response to feeling behind is to sprint in every direction at once — that's exactly how the majority ended up with adoption and no return. Pick one workflow that hurts. Measure it honestly before you change anything. Run a small, owned pilot with light guardrails. Track speed, output, and quality together so you improve the work without quietly degrading it. Then prove it and template it. You don't catch up by doing the most AI; you catch up by being the rare organization that can actually show what its AI did. Late isn't the problem. Aimless is.

And if you'd rather not navigate that first quarter alone, that's the kind of work my firm, Prosigliere, does — we embed AI-forward engineers and fractional leaders to help teams pick the first workflow, ship it, and measure it. (Disclosure: Prosigliere is my employer.) You can start with a quick call at prosigliere.com.

Sources: WRITER: enterprise AI adoption in 2026 · Azumo: enterprise AI adoption statistics · CIO: measuring AI-enabled success