One of our client projects sat for 45 days waiting on one data source we needed from them.

That was on us. We used to start every AI engagement trying to accomplish everything at once. In the same few weeks, we’d try to connect the client’s data, meet their users, and map out their use cases.

With everything kicking off at the same time, getting us the data was just another task competing for the client’s time. But the rest of the project depended on that data, so everything was held up until it arrived.

Meanwhile, users got pulled in before the data was ready. We'd meet with someone, identify their use cases, and then go three weeks without getting back to them because the data still hadn't been tested.

Projects that start and stop like that are not likely to succeed. The operators I work with are under pressure to show AI is working, and across our clients the hardest part has been getting people to use it. Every stall during implementation makes it that much harder, and inside a company you usually get one shot.

Now we run every new engagement on a mutual action plan. It splits the project into four phases, each with a named owner on our side and the client's side, and has cut implementation time in half.

Below is the plan, how each phase works, and how to run it in your own company.

I see the same thing happen all the time inside companies implementing AI projects. They kick off everything at once, and the project stalls.

The fix is four answers before kickoff: who owns each data source, who verifies the numbers, who the first users are, and when each phase ends. Keep it to those four, because every extra step before kickoff delays the decision to start.

Here is each phase of the mutual action plan:

With our clients, every phase has an owner on their side and ours, so they're accountable for their part, like data access and their data owners' time, the same way we're accountable for ours.

One rule applies to every phase. If something isn’t ready on time, whether a data source, an owner’s availability, or a user’s sign-off, it moves to a later phase and the project keeps moving.

Let's take a look at exactly how each phase works below.

Before the AI can do anything, it needs access to the software your team already runs on, like your CRM, ERP, or third-party data sources. Getting that access depends on people, and that's where most of our engagements used to stall.

It would sometimes take over a month just to get time from the data owner of a certain system. Only one person on their team could get us access, and we’d hit a bottleneck.

Now that person is named before the project starts. Each data source is assigned an owner who can get us access and tell us where everything is without anyone else's help, and that owner is on a call with us in week one.

If they can't make the time, that source moves to a later phase, and we keep going. This phase takes one to four weeks, and nothing else starts until it's done.

Inside your own company, the owner is whoever can grant access to each system and knows the actual structure of the data and the intended use of it. Book their time for week one.

You're ready for the next phase when every source in scope is connected or has been moved to a later phase.

We used to onboard users as soon as the data was connected. They’d find errors in the data and stop trusting the AI before the errors were fixed.

Now someone checks the AI's answers against the company's own records for about two weeks before anyone else uses it. This is usually a different owner, like an analyst or an ops lead who pulls the numbers every day.

We tell the client upfront that there will be mistakes in the beginning. Some come from the client's own data, and some come from the AI, which is never right 100% of the time.

Reconciling is how you find them. The owner asks the AI the questions they already answer every day and checks each answer against the source. When an answer doesn't tie back, we find where the problem is and fix it before users ever see it.

We're running this phase with a commercial real estate client now. Their team is asking the AI about NOI, occupancy, and leases and checking each answer against their property management accounting platform. When the answers tied back, they told us reconciling those numbers by hand would have taken seven days. The AI did it in a minute.

You're ready for the next phase when those answers tie back to the source and the owner says they trust them.

Once the owner trusts the data, pick a small group of power users, often one from each relevant department, and run their real work through the workflow before anyone else gets access. Choose people you know will adopt it quickly and help champion it inside the company. When it works for them, everyone else will be more willing to use it.

Give it a few weeks, up to a month. Keep going until each power user says they've replaced the old way with the new workflow. Once they do, you’re ready for the final phase.

By the time you open it to everyone, the data is connected, the numbers tie back to the source, and people inside the company already use it. That's why the rollout is easy.

Nobody is asking what this new workflow is for, whether the data is right, or whether to trust it. The power users show their own teams what they stopped doing by hand, and those teams see it working before they're asked to use it.

Reconciling won’t make the AI perfect. It will still get something wrong after rollout, because no AI is right 100% of the time. Tell users that upfront, so the first mistake they see doesn’t cost you their trust.

Your power users can mislead you. Pick only the enthusiasts and the workflow looks adopted before it’s been tested by anyone skeptical. Include at least one person who does the work the hard way today, because their yes is the one the rest of the company believes.

Moving a source to a later phase only works if the first workflow can run without it. If it’s the one source everything depends on, deferring it just stalls the whole project. That’s a priority problem, and whoever the owner reports to has to fix it.

That’s why we’re building the mutual action plan into how we sell. The client agrees to their half before they sign, not a month into the project.

Zach

Founder & CEO, OutcomeCatalyst
Book 30 minutes with me →