How We Work Our Work Our Team Start a Conversation
Our Work  /  Case Study
Caizin case study  ·  Management consulting

TQMI put AI to work inside its consulting delivery

TQMI's consultants write long, evidence-backed reports for industrial and manufacturing clients. Caizin placed Anthropic's Claude in the three places those consultants already work and ran the rollout on real client work. Ten use cases are documented so far. Where consultants measured it, the work took 25–75% less time.

10
Documented use cases
4
Consultants
25 to 75%+
Less time per task, where measured
3 days
Client manual: one month of work done in 3 days
Engagement summary

What was done, and what came of it.

Client
TQMI is a management consulting firm serving industrial and manufacturing organizations in quality, operational excellence, and business performance. Its projects range from Deming Prize programs (a Japanese award for company-wide quality management) and frontline quality improvement teams to sales-process design and factory-floor problem solving.
Engagement
TQMI hired Caizin to put AI into its consulting work. The test was narrow. Working on a real client project against a deadline, could a consultant produce better work in less time and still control quality?
What was deployed
Claude, Anthropic's AI assistant, in three places that match how TQMI's consultants already work: a chat window for research and thinking, a workspace for building documents from source material, and Claude inside MS Office for documents, decks, and spreadsheets. Caizin also set the working rules for which tool suits which job and for consultant review of anything a client sees.
Early results
Four consultants ran ten use cases. They covered sales analytics, employee turnover, spotting abnormal readings in daily operations data, a client process manual, competitor benchmarking, Deming Prize documents, quality presentations, follow-up training support, and research before a client visit. Where consultants measured it, the work took 25–75% less time, and they reported more rigorous and more consistent output.
Background

An adoption program, run on real client work.

Caizin works with growth-stage and mid-market companies on strategy and then stays with them while the work gets done. AI adoption has become part of that delivery work. With TQMI it was the entire engagement.

The question was simple. Could an experienced consultant, working against a client deadline, get more done without lowering the quality of the work? You can only answer that on real projects, so the program ran on live client work from day one.

Where the consulting time was going

A single TQMI engagement can generate process manuals, benchmarking studies, root-cause analyses, executive readouts, and multi-year improvement roadmaps. Each one has to be accurate and specific to the client's business, and it has to hold up when the client's leadership pushes back.

That creates three problems. Research and drafting eat up days or weeks before the client sees anything at all. Turning findings into a clear, client-ready format is manual work that tends to land on senior people. And a big program has more moving parts than one person can keep consistent. A Deming Prize plan, for example, covers dozens of activities over several years.

The limit was volume of work, and the skill was already there. Senior consultants were spending a lot of their time assembling documents instead of doing the thinking that makes those documents worth reading.

How the rollout was run

Caizin put Claude into the tools TQMI's consultants already used, and each consultant applied it to the document in front of them. Nothing else about how they worked changed.

Consultants used the chat window for research and for stress-testing their thinking before a client visit or a first draft. The workspace turned raw input, like a training transcript or a set of sales records, into a finished document. Claude inside MS Office produced the Word documents, decks, and spreadsheets that clients actually receive.

Two rules kept the work good enough to send. First, match the tool and the AI model to the job. On the training tutor described below, a lighter model gave wrong answers and a stronger one gave correct answers. The team then standardized on the setting that worked. Second, a consultant reviewed every client-facing output. In all ten use cases a consultant checked the work, corrected it, and decided what the client saw.

The work

Ten use cases, from four consultants.

Three are described in full below, chosen because the consultants who ran them measured the time saved. The other seven are summarized after that. Individual consultants are not named, and client identities are kept to a broad sector at most.

Case one  ·  Sales analytics
A sales metrics dashboard that recommends action

A field sales team tracks about 17 metrics per rep every day. A branch manager with ten reps walks into the morning meeting facing about 170 numbers, which is more than anyone can review properly. A TQMI consultant used Claude to build a dashboard that identifies each rep's specific gaps and proposes an action against each metric.

The morning meeting became more focused, reps could see their own performance next to the branch's, and the client received recommendations it had not had before. The consultant put the data and the recommendations through several validation passes before taking anything to the client.

Over 75%
of the effort removed, as reported by the consultant who built it
Case two  ·  Client process manual
A sales route-planning manual delivered in three days

Route planning sets which market each sales rep covers on which day. One client needed a full manual for it. Doing the writing and the background market research by hand would have taken several weeks to a month, the consultant estimated.

Claude drafted about 70 percent of the manual, in the workspace and in MS Office. The consultant spent their time proofreading and tailoring it to the client's business, and used the freed-up time to start other work earlier.

3 days
against an estimate of several weeks to a month
Case three  ·  Deming Prize program
Documents defined for a four-year quality plan

A heavy-industry manufacturer going for the Deming Prize needed a specific document for every step of its four-year improvement plan, more than 28 steps in all. The usual approach was repeated working sessions and drafting one step at a time.

The consultant wrote a single detailed prompt carrying the full program context and generated a document for every step at once. The consultant then checked each one, cutting or rewriting anything that was too rigid or too thin before the client saw it.

Over 75%
of the effort removed across 28-plus activities
The other seven use cases
Use caseWhat was doneEffect
Employee turnover analysis Raw factory HR data turned into a presentation the client could use. It covered 66 departures sorted by time served, department, floor, role, and supervisor, and all 66 turned out to be people who quit early without giving notice. A spreadsheet exercise became an executive readout with a prioritized roadmap for onboarding, mentoring, and root-cause follow-up.
Exception detection on daily operations data A plant has hundreds of pieces of equipment, each logging dozens of readings a day. Claude reads the daily report and flags the ones running outside normal ranges. The team only looks at the flagged items. Reading the whole report by hand was slow and easy to get wrong.
Competitor benchmarking study Claude pulled together the strengths of two competitors and turned them into a ranked list of recommendations. Each one got an effort-versus-impact score, which the consultant reviewed and adjusted. Recommendations were easier for the client to prioritize, and the structure reduced rework.
Improvement-team project on assembly-line rejects An 8-step improvement-team presentation built in TQMI's standard plan-do-check-act format, including cause analysis, a table of fixes, before-and-after results, and a plan to make the fix permanent. The method and the numbers came from the project itself. Claude did the structuring and layout, which usually takes hours per deck.
Quality write-up on welding defects A 10-step quality problem-solving presentation in the same format, covering the topic chosen, the starting data, the fixes applied, the results, and how the fix was made permanent. A second deck produced to the firm's standard format without the usual assembly time.
Post-program training tutor The transcript of a delivered training program turned into a tutor instructed to answer only from that transcript, so participants who missed a session could ask about the material. This worked as a test. To open it up to participants, TQMI would need a user interface and somewhere to host the transcript, which it has planned as the next step.
Pre-visit assessment research Before a first visit to an automotive client, Claude researched the problems common to that kind of plant, what customers usually complain about, and how competitors compare. The consultant also did the usual review of weak metrics, past history, and client feedback. Confirmed the hypotheses the consultant had formed before the site visit and surfaced areas that might have been missed. Findings were validated with direct questions on site.
Results

What the consultants reported.

The figures below are as reported by the consultants who ran each use case, and the pattern across the ten is consistent.

Speed
25 to 75%+
Time removed on the tasks where consultants measured it, with the document-heavy work at the top of that range. The route-planning manual is the clearest case, moving from an estimated month to three days.
Quality
Closer to client-ready
Consultants said the work was more thorough and more consistent, and drafts came back needing only tailoring. One consultant said the sales work "gave the speed and clarity" to get where a normal change-management program would have taken months and real money.
Control
Consultant sign-off
A consultant checked everything the client saw. The sales dashboard went through several rounds of checking, and every document in the four-year plan was reviewed before it went out.

Work that used to take weeks now takes days. A second consultant described the pre-visit research as giving "focused, technical depth and relevant value to customer". Model choice mattered on the training tutor, where the consultant settled on the configuration that returned correct answers. Claude took over the drafting and assembly, and consultants still decided what was right for the client.

What made the difference

What made it work.

The tools sat inside the existing workflow. The same assistant was useful in a chat window for research, in a workspace for building a document from source material, and inside MS Office for the final version, so consultants got the benefit without stepping outside the tools they already used.

Drafts came back nearly ready to send. Across all ten use cases, the only work left was review and tailoring, and that is where the time savings came from.

Settings mattered as much as the method. On the training tutor, picking the right model and effort level was the difference between wrong answers and reliable ones. TQMI now has a repeatable way to get dependable results.

Across ten use cases, the work took 25–75% less time where consultants measured it, and a consultant reviewed every output before a client saw it.
As reported by the four TQMI consultants who ran the work
What this points to

Ten use cases is
just the start.

For TQMI, the ten use cases came from four consultants applying Claude to their own client work. The same method works across the rest of the firm: use the tool on real client work and keep consultant review in place. It applies to quality programs, turnover analysis, executive summaries, and proposals.

For Caizin, this project shows what AI adoption looks like on real client work. Consultants keep the client relationship and the judgment calls, and Claude takes on the research and assembly that used to limit how much they could produce.

Caizin, Raise the Bar

Want AI adopted on the work
your team already does?

Caizin runs AI adoption on real client work, sets the rules for which tool fits which job, and makes one named person accountable for everything the client sees.

Start a Conversation →