MyM Consultancy

How we work

Our work follows a clear process: understand how your team operates, identify where AI adds real value, build or identify the right tools, train your people to use them well and stay with you as things evolve. Not every company needs every step. We adapt the scope to your situation – whether that is a full implementation or a tailored solution to a specific problem.

01

Understand and plan

We start by getting to know how your team works – where time gets lost, which tasks and processes would benefit from AI and what tools people already rely on. From there, we figure out together where AI makes sense and where it does not. We look at what you are already paying for and whether a smarter, cheaper tool could do the same job or open up things your team never had the time to do before.

We also look at what is underneath, help you get your data structured and clean enough for AI to work with, make sure your systems can talk to each other and address any privacy or security considerations.

We research and compare specific tools and models, prepare a complete solution and, before you commit to anything, we build a proof of concept so your team can test it and evaluate the output.

02

Build and safeguard

We design and build the AI workflows your team will use. This means writing the prompt chains, defining what goes in and out at each step and wiring it together – typically in no-code tools like n8n, Make or Zapier so your team can maintain it without a developer.

Every workflow is built around the principle that AI supports your team, not the other way around. We add guardrails so a bad output in one step does not cascade into the next. We design fallback paths so when the AI is uncertain, the system routes to a person instead of guessing or pushing a bad output through.

Finally, we design how data and privacy work across all your AI flows – what gets anonymised, what stays within your systems and what the AI never sees.

03

Training and rollout

AI tools are not effective if people don’t know how to get the best out of them or do not feel comfortable using AI. We organise training and mentorship tailored to your team’s needs: how to get good results from AI tools, which tasks are worth delegating to AI and how to tell when AI output is reliable and when it needs a closer look. Where it makes sense, we prepare confident AI users within your organisation to support their colleagues.

For the rollout itself, we plan the transition so it is gradual and manageable. We define clear ownership – who reviews AI outputs, who signs off, who is responsible when something goes wrong. And we design how the AI communicates uncertainty to your team, so people know exactly when to step in.

04

Monitor and maintain

After launch, we stay involved. We check in regularly to see whether your team is using the tools effectively and comfortably, whether they are still worth what they cost and whether the quality of their output has held up.

As your team grows, we make sure the setup scales with it: new hires need onboarding, workflows need adjusting and what worked for a small team may not work once the team doubles. We keep your management informed about relevant changes in the industry: new risks, new tools, shifts in the market, so your approach stays current.

Frequently asked questions

Start by looking at where your team’s time actually goes – not what you think the problems are, but the repetitive tasks people complain about or quietly work around. List the ones that eat the most hours or cause the most friction. Pick one or two with clear inputs and outputs, something like invoice processing, email triage, or drafting standard reports. Test a simple AI tool on that narrow task, see what it gets right and wrong, and build from there. An AI expert can help map the landscape and pick the first target, which is usually the hardest part.

Four rough phases. First, discovery – figuring out where AI fits, what you already have, and what can be improved or replaced. Second, design – picking tools, writing the prompts and workflow logic, defining what the AI does alone and what needs a person. Third, rollout – training the team, running the first weeks with extra oversight, catching issues before they become patterns. Fourth, steady state – monitoring quality, adjusting as tools change, keeping an eye on costs. The first phase takes the longest to think through. The last one never really ends, but it gets lighter.

Good candidates share four traits: they’re repeated often, they follow roughly the same steps each time, the inputs are already in a digital format, and the cost of getting them slightly wrong is low to medium. Drafting status reports, categorizing incoming emails, extracting data from invoices, and first-pass research all fit. Tasks that need human judgment about tone, stakeholder politics, or high-risk decisions don’t. When in doubt, run a small test – automate one instance, compare the AI output against what a person would have produced, and see how much editing it actually needs.

The work typically breaks into discovery, design, build, and rollout. Discovery maps your current processes, data sources, and the problems worth solving. Design picks the tools, defines how the AI workflow runs, and plans for the parts that need human oversight. Build wires it together – writing prompts, connecting systems, setting up the monitoring. Rollout trains your team, runs the first weeks with extra supervision, and adjusts based on what actually happens once real work flows through. Good consulting also includes a handover so your team owns the result, not the consultant.

The same thing that happens when anyone makes a mistake, except the responsibility question is murkier – there’s no single person to point to. That’s exactly why AI mistakes need to be handled upstream. Define clearly who owns each AI-generated output – usually the person who reviewed or approved it, not the AI itself. Build review steps where the consequence of error is high. When something does go wrong, treat it like any operational incident: figure out what slipped through, fix the workflow, not just the instance. Companies that think about AI accountability before a mistake happens handle the first mistake a lot better than companies that figure it out reactively.

Two layers work well together. A human review for anything customer-facing – AI drafts, a person approves before it sends. Automated checks that run before the human sees it – fact-checking against a verified source, format validation (did it include the required fields?), guardrails on tone or content (no promises of delivery dates that aren’t committed, no pricing claims that weren’t approved). The automated layer catches the mechanical mistakes so the human is only checking the judgment-dependent parts. For lower-stakes messages, a human-review-sample approach (every 10th output manually checked) can replace 100% review once you trust the output quality.

Frame it as a tool, not a replacement. Start with the work, not the technology – ask people which parts of their job they’d happily hand off (there are always some), and train on AI for exactly those parts. Keep the training specific and short. Nobody needs a three-day AI course, they need 90 minutes on three things that matter for their role. Involve people in deciding what to automate and what to keep. Their buy-in runs deeper when the choice is theirs. And be honest about what’s changing – avoiding the topic makes people suspicious, not reassured. An AI expert can run training sessions that land this way.

Start with a single pilot on a low-risk task – ideally one where your team already wishes something would change. Let it run long enough to produce real data (a month is usually enough), measure what’s working and what isn’t, and iterate before expanding. Once the first one is stable, add a second use case. Resist the urge to roll out five things at once. Each one needs attention to adopt properly, and spreading thin usually means nothing lands. Keep the existing manual process available as a fallback during the pilot period – switching back instantly if something goes wrong keeps operational risk low.

Three habits make the difference. A shared prompt library where approved prompts live, organized by task and team, with each entry showing the prompt, its use case, and example inputs and outputs. A light review process for adding new prompts – a peer check before they go in, so the library doesn’t fill with low-quality entries. And periodic pruning – quarterly, retire the prompts no one uses and update the ones that have drifted. For larger teams, tools like PromptLayer or Langfuse add tracking and version control. An AI expert can help set up the structure and seed the first 20 prompts.

Both, and usually a mix of the two. Project-based engagements work well for defined implementations – a specific workflow, a specific audit, a training program – with a clear deliverable and end date. Retainers work for ongoing advisory, optimization, and adapting as the landscape changes – useful once the initial implementation is live and the team needs a light hand on the tiller rather than a fresh engagement every quarter. A common shape is project pricing for the initial build, then a smaller monthly retainer for three to six months of support and iteration. Good consultants flex to what actually fits the engagement rather than forcing a model.