AI implementation succeeds or fails based on what happens to your team after the system goes live. The automation has to target the real bottleneck, freed capacity has to be redeployed to higher-value work, and the team has to adopt the new workflow. Most agencies plan for the first. Almost none plan for the other two.
Based on the working methodology of Pixels and Clicks, Gartner research on GenAI project abandonment (2024), and McKinsey data on transformation program outcomes. October 2026.
You run supply chain or operations at a mid-market consumer brand. You hire an AI agency to automate a workflow that eats 10 hours of your team’s week.
The agency demos something impressive, builds it, and invoices you.
Six weeks later the system works exactly as spec’d. Your team doesn’t use it. The person it was built for still runs the old process on Mondays because the new tool doesn’t match how they actually work.
The failure is organizational. The agency never planned for what changes inside your company after the system goes live. This take from @giyu_codes captures the pattern we see repeatedly:
Gartner estimated that 30 percent of GenAI projects would be abandoned after proof of concept by the end of 2025.
McKinsey puts the failure rate for transformation programs at 70 percent. The dominant cause in both is organizational: people, workflows, and adoption.
This is how we think about AI at Pixels and Clicks. We built the practice around what happens after the system goes live. Here’s the process, step by step.
We map the work before we touch the system
The first thing we do in any engagement is map the real workflow. Who does the work, in what order, in which tools, and what breaks when something unexpected comes in.
The owner’s description of a process and the team’s execution of it rarely match.
The CEO says reconciliation takes an hour. The team says four. The gap is the manual copy-paste between WMS and ERP, and the email chain that fires every time a PO doesn’t match.
An automation built off the wrong map becomes a perfectly working system the team ignores. Companies cycle through three agencies this way. Each one automated what leadership described. Each one missed what operations actually did. @giyu_codes makes the same point here:
That’s why every Pixels and Clicks engagement starts with a paid assessment.
The OTIF Deduction Assessment, the Data-Bridge Friction Assessment, and the AI Forecasting Feasibility Assessment all follow the same sequence: look at the data, map the real process, find the bottleneck. Then decide whether to build anything at all.
What you walk away with: a process map showing every step your team actually takes, the time each step costs, where the gaps are between what leadership described and what operations does, and a recommendation on whether to automate, restructure, or leave it alone. You own the map whether or not you hire us to build anything.
Free hours are wasted without a redeployment plan
The most common pitch in AI consulting is “this will save your team X hours per week.”
That pitch is true. It’s also incomplete.
Saving 10 hours doesn’t save 10 hours of payroll. That person is still on your team. The hours don’t vanish. They become available.
Without a plan for that availability, you’ve created idle capacity. The automation works. Your team still doesn’t feel more productive.
Six months later the CFO asks what the AI project delivered, and the honest answer is “faster spreadsheets.”
What we plan for is the redeployment. When a manual reconciliation is automated, we identify what that person should be doing with the recovered time.
That might mean tracing unresolved deductions. Running the dispute process the team hasn’t had time for. Building the evidence packs that recover real money.
What the redeployment plan covers: the specific higher-value tasks, the estimated revenue or cost recovery each one targets, and the priority sequence. Your team walks out of the assessment knowing exactly what changes and who does what differently.
The goal is the same people doing work that compounds.
First, second, and third order effects
When we automate a workflow, three things happen in sequence.
First order: the task gets faster and more accurate. The data extraction that took two hours takes two minutes. The reconciliation that missed 8 percent of mismatches catches 99 percent. Every agency sells this, and it’s real.
Second order: the person who did that task moves to judgment work. Your AP clerk who spent Mondays re-keying purchase orders reviews deductions with a dispute deadline closing next week. Your demand planner who built spreadsheets all Wednesday afternoon tests pricing scenarios. The team’s capability goes up without adding headcount.
Third order: the organization starts questioning what else could change. Problems your team used to accept as the cost of doing business become solvable. The operations lead who spent 15 hours a week on manual ASN reconciliation now has the time to ask why the ASN is wrong in the first place. That’s how you move from fixing symptoms to fixing root causes.
Most agencies sell the first order and stop. We plan for all three, because the second and third are where the value compounds.
Change management is half the work
Setting up AI is 30 percent technology, 70 percent people. @lukepierceops echoes this perspective:
You can build the most elegant automation in the world. It will fail if the person it’s built for doesn’t trust it.
Every workflow we automate gets a named owner inside your company. A specific person, by name, who understands what the system does, what it replaced, and what to do when it breaks.
We pick that person during the build so they have ownership before the system ships.
Documentation is written for the operations team, in their language. Walkthroughs are live calls. Training runs alongside the build.
What ships with every automated workflow: a one-page runbook covering what the system does, what it replaced, what to check when something looks wrong, and who to call. We test the runbook with the person who’ll actually use it before the handoff.
Most AI implementations fail because the team never adopted the tool. The agency shipped the code but never invested the time to make adoption real.
AI only where it earns its place
We’ve told clients not to automate things.
That might sound strange from a practice that does AI implementation for a living.
But some processes work fine as they are. The team trusts them. The manual cost is low enough that automation wouldn’t pay for itself in a reasonable timeframe.
Forcing AI into a workflow where it doesn’t belong creates friction where none existed. It burns political capital. It makes your team skeptical of the next change you propose. @lukepierceops puts it well:
Our test before building anything is straightforward. Will the automation pay for itself within a timeframe you’d accept for any other investment? Will the person it frees be redeployed to work that generates returns? Will the team actually adopt it?
If any of those answers is no, we don’t build it.
We’ve walked away from paid build work after an assessment showed the manual process cost less than the automation would. The assessment fee is structured so we get paid for the truth, regardless of the answer.
“AI only where it earns its place” is how we scope every engagement.
The question to ask your next AI agency
If you’re evaluating agencies, ask what happens to the time your team gets back.
An agency that can explain what your people will do with recovered hours is thinking about the whole problem. An agency that sells only “hours saved” is selling half the engagement.
Ask how they handle adoption. If the answer is a training session scheduled for after the build, the team won’t use the tool.
Ask whether they’ve ever told a client to skip an automation. An agency that builds everything it’s asked to build, regardless of ROI, is optimizing for its own invoice.
Ask what you get if the answer is “don’t build.” If the answer is “nothing,” the assessment is a sales pitch, not an analysis.
That’s the practice we run. Fixed-scope assessments that start by mapping the real problem, planning for the organizational change, and building only what earns its place.
Talk to the founder directly. If the answer is that you don’t need us, we’ll tell you.
Frequently Asked Questions
- How does Pixels and Clicks approach AI implementation differently from other agencies?
- Pixels and Clicks starts with a paid assessment that maps how work actually gets done before proposing any build. The assessment deliverable is a process map showing every step the team takes, the time each step costs, and where the gaps are between what leadership described and what operations does. The assessment also includes a go/no-go recommendation.
- Does Pixels and Clicks replace employees with AI?
- Pixels and Clicks automates low-value repetitive work so existing employees can move to higher-value tasks like deduction tracing, dispute management, and root-cause analysis. Every engagement includes a redeployment plan that names the specific higher-value work and sequences it by priority.
- What are the first, second, and third order effects of AI implementation?
- First order: the automated task gets faster and more accurate. Second order: the person who did that task moves to judgment-heavy work that compounds value. Third order: the organization starts questioning inefficiencies it used to accept as cost of doing business.
- Why does Pixels and Clicks run a paid assessment before building anything?
- The real bottleneck rarely matches what leadership describes. Building an automation from a description of the workflow rather than mapping how the team actually works produces tools the team ignores. The assessment maps the real process and costs before any build decision. We've walked away from paid build work after an assessment showed the manual process cost less than the automation would.
- What does change management look like in a Pixels and Clicks engagement?
- Every automated workflow gets a named owner inside the company, picked during the build. The workflow ships with a one-page runbook covering what the system does, what it replaced, what to check when something looks wrong, and who to call. Walkthroughs are live calls. Training runs alongside the build.
Find out what your deductions are actually costing you.
A 30-minute call with the founder, who will tell you which of the three scans fits your problem, or that none of them do.