What Piers Linney Says About Implementing AI in an SMB

By Lazar Milicevic · Published October 15, 2026 · 10 min read

You have read the AI think-pieces, tried ChatGPT a few times, and still have no deployed workflow saving anyone hours. Piers Linney, co-founder of Implement AI, argues that this gap, not the technology, is where most small businesses stall. His advice is scattered across interviews and podcasts rather than packaged as one named framework, so this post pulls it together, labels what is opinion versus evidence, and turns it into a checklist you can run this quarter.

Who Piers Linney is, and what to treat as evidence

Piers Linney MBE is co-founder and Executive Chairman of Implement AI and a former Dragons' Den investor (Tech Show London speaker page). Implement AI was founded in 2023 with Dr Aalok Shukla, and says it has guided more than 120 UK businesses (About Implement AI). That count is self-reported by the vendor.

One caveat before the takeaways. I could not find a single named, published "Linney framework" with defined steps. What exists is a consistent set of positions repeated across interviews and the Implement AI Podcast: start small, pick a measurable ROI, write an AI policy, use the tools yourself, and think about growth rather than only cost-cutting. The rest of this post treats those as advice from an operator who sells implementation services. That makes him informed and also interested. Weigh his claims accordingly, and note that vendor ROI figures on Implement AI's own marketing pages are not independently verified, so I do not repeat them here.

Takeaway 1: Start with one workflow that pays back fast

The core advice: pick workflows or blockers with an immediate ROI, and focus on areas where the technology works today (Brandpie interview). In practice, that rules out "transform the business with AI" and rules in "stop hand-sorting the support inbox."

A workable selection filter for a team of 1-10 people:

Score your top five candidate tasks against those five rows. Pick the one with the most "good" answers. If two tie, pick the one where a mistake is cheaper.

Write the baseline down before you build anything. A spreadsheet row is enough:

workflow: inbound lead follow-up
owner: sales lead (you)
baseline:
  avg_time_to_first_reply: "measure for 2 weeks"
  leads_per_week: "count from CRM"
  leads_never_contacted: "count from CRM"
target_after_30_days:
  avg_time_to_first_reply: "set a number you can verify"
review_date: "30 days after go-live"

Without a baseline, you cannot tell a good pilot from a busy one.

Takeaway 2: Know which ROI lever you are pulling

On the Implement AI Podcast, Linney and Shukla describe two levers: reducing the cost of serving customers, and capturing revenue that is currently being missed (Apple Podcasts listing). Linney also says AI is about growth and new revenue, not only cost-cutting (Brandpie).

For a small business, that maps to two categories of automation:

Cost-to-serve levers

• Triage and draft replies to support email
• Extract invoice data and route it for approval
• Summarize calls and update the CRM

Missed-revenue levers

• Follow up with every inbound lead within minutes, not days
• Re-contact old quotes that went quiet
• Flag customers who haven't ordered in a while

Most solopreneurs default to the first category because the pain is visible. The second is often worth more, because a lead that never got a reply has a cost you never see on a timesheet. Here is an illustrative example with round numbers (not a benchmark): if you get 40 inbound leads a month, never reply to 10 of them, and your average job is worth $1,000 with a 20% close rate, those 10 missed leads represent roughly $2,000 of expected revenue per month. Your own numbers will differ, but the exercise is worth doing before you decide what to automate first.

Takeaway 3: Use the tools yourself before you buy anything

Linney says leaders should personally use tools like ChatGPT and Google Gemini (Brandpie). I agree, for a builder's reason: you cannot write a good spec for a task you have never tried to delegate.

A one-week owner exercise that costs nothing:

1. Take five real tasks from your inbox or CRM.
2. Paste the input into a general-purpose assistant and ask for the output you would normally produce.
3. Mark each result: usable as-is, usable with edits, or wrong.
4. Note why the wrong ones failed (missing context, ambiguous rules, bad input data).

That list of failure reasons is your requirements document. It tells you what context the automation needs, which rules you have never written down, and where a human must stay in the loop. If you skip this step and hand a vague brief to a consultant or a tool, you pay for them to discover the same things.

Takeaway 4: Write the AI policy before the pilot, not after the incident

Linney argues every company needs an AI policy, and that it must run through HR, training, compliance and data security (BM Magazine). He also cites research that almost a third of employees use AI at work without their employer knowing, so-called shadow AI. The article doesn't name the underlying study, so treat that number as an anecdote rather than a statistic you can quote.

You don't need a 30-page document. For a small team, one page that answers these questions is enough:

ai_policy_v1:
  approved_tools:
    - "list the specific tools and plan tiers"
  data_never_to_paste:
    - customer personal data
    - payment details
    - credentials and API keys
    - unreleased financials
  human_review_required_for:
    - anything sent to a customer
    - anything touching money
    - anything legal or contractual
  logging: "where AI-generated outputs and approvals are recorded"
  owner: "one named person"
  review_cadence: "quarterly"

Two practical notes. First, a policy that bans everything just pushes usage underground. Name approved tools so people have a legitimate path. Second, if your automation touches an inbox, a CRM, or accounting data, decide now which systems the agent may read and which it may write to. Read access to a whole mailbox is a bigger exposure than most people assume.

Takeaway 5: Strategy first, technology second

A podcast episode with Paul Farrelly of Freeman Clarke covers why successful AI adoption starts with business strategy rather than technology (Apple Podcasts listing). For an SMB, "strategy" can be answered in three sentences:

• Which number do we want to move this year (revenue, margin, response time, capacity)?
• Which two or three processes most directly drive that number?
• Which of those are repetitive and digital enough to automate?

If you can't answer the first question, no tool will fix it. The OECD's December 2025 discussion paper on SME AI adoption sorts small firms into four categories: AI Novices, AI Explorers, AI Optimisers and AI Champions, and names connectivity, AI-enabling inputs, skills and finance as the key enablers (OECD paper). Use it as a self-check: if your "inputs" are scattered across personal inboxes and paper, your first project is cleaning up data, not deploying an agent. That is not glamorous, but it is where most small-business AI projects quietly fail.

What a staged rollout looks like (and why vendor timelines vary)

Implement AI's public materials show how a vendor structures this. Its AIOS Pilot is a 30- to 90-day programme that deploys one AI agent to prove impact before scaling (AIOS Pilot). Its AIOS Workforce page describes a four-week rollout: connect systems in week 1, train the worker in week 2, run in parallel with humans for review in week 3, and client sign-off in week 4 (AIOS Workforce).

Note that the vendor's own pages give different timelines for different products, and I wasn't able to reconcile them, so don't quote any single number as "how long AI deployment takes." The pattern is what matters, and it applies whether you use a vendor, a freelancer, or build it yourself:

1. Connect: wire up the systems the workflow touches, read-only first.
2. Train/configure: encode the rules and examples from your owner exercise.
3. Shadow: run alongside the human process and compare outputs. Nothing goes to customers unreviewed.
4. Sign off: promote to live only against the success metric you wrote down at the start.

Implement AI describes three agent types: interactive workers (voice, SMS, email conversations), analyst workers (analyze every conversation at scale), and action workers (operate the software the team uses) (AIOS Workforce). That taxonomy is a useful way to scope your own project. Most SMB first projects are an analyst (summarize and classify) or a light interactive worker (draft replies for approval). Action workers that operate software autonomously deserve the tightest permissions and the longest shadow period.

The KPIs the AIOS Pilot lists are a decent template for your own scorecard: cost-per-interaction, incremental revenue, SLA compliance, customer sentiment, and utilization versus a human benchmark (AIOS Pilot FAQ). Pick two or three, not all five.

The checklist: from theory to deployment

Here is the practical checklist, assembled from the advice above. Treat the sequence as the point.

Week 0: Decide

• Name the one business number you want to move
• List five repetitive digital tasks; score them with the filter table
• Choose one workflow; decide whether it's a cost-to-serve or missed-revenue lever
• Record the baseline for 1-2 weeks

Week 1: Learn and protect

• Run the owner exercise on five real tasks; log why failures failed
• Write the one-page AI policy; name approved tools and one owner
• Decide read vs. write access for every system the workflow touches

Weeks 2-3: Build and shadow

• Build the narrowest version that does one job
• Keep a human approving every outbound or money-touching action
• Run in parallel with your existing process; compare outputs side by side
• Log every correction; that log is your improvement backlog

Week 4+: Decide with data

• Compare against the baseline you wrote down
• Promote, adjust, or kill. Killing a pilot that missed its number is a success of the process
• Only then pick workflow number two

Build, buy, or hire help

Linney's firm sells implementation, so his advice naturally includes working with specialists. That isn't a criticism; it's a trade-off you should evaluate honestly.

On pricing: Implement AI lists AIOS Command from £250 per seat per month with an optional £750 one-off assisted setup (AIOS Command). Its data-handling claims are UK-hosted client data, no training of public models on that data, and Cyber Essentials certification (AIOS Workforce). These are vendor statements; check current pricing and request the security documentation yourself before signing anything. Whoever you choose, ask the same questions: what systems does it need access to, where is data stored, what happens when it's wrong, and how do I turn it off.

How BizFlowAI approaches this

BizFlowAI builds AI automations for solopreneurs and small teams: practical systems that work, without the buzzwords. The approach follows the sequence described above. Pick one workflow with a measurable baseline, scope permissions tightly, run it in shadow mode against the human process, and only promote it when the numbers hold up. If your checklist from this post leaves you with a candidate workflow and a baseline, that is the right starting point for a conversation.

Work with BizFlowAI

If you'd rather have this built for you, that's what we do: production AI automation for solo founders and small teams — agents, integrations, and document pipelines that actually ship.

Request a free discovery call — 30 minutes, we map the highest-ROI automation in your workflow. No pitch deck, just engineering.

More guides like this on the BizFlowAI blog.