Skip to content
+90 536 247 60 12digital@indoles.com.trLevent, İstanbul
Mon–Fri 09:00–18:00
Transform · Service 06 / 12

AI advisory

AI advisory is the work of separating where artificial intelligence genuinely pays off from where it is an expensive toy. INDOLES starts not with the technology but by measuring which tasks consume how much time and money.

View the package
Diagram — AI advisory06/12
Sound familiar?

This service steps in when one of these is true.

  • Everyone talks about AI and no one knows where to start.

  • A team does the same task by hand every day, and it grows with volume.

  • Years of data have piled up but none of it informs decisions.

Scope

What it covers, and what it does not.

What's included

  • Candidate task list

    Which tasks are repetitive, rule-bound and data-driven? Candidate tasks are listed by their concrete names.

  • Cost-benefit calculation

    For each candidate, time saved, setup cost and monthly running cost are compared. What does not return gets dropped.

  • Data readiness check

    Is your data sufficient, clean and accessible? Most AI projects stall at exactly this step.

  • Method selection

    Off-the-shelf service, rule-based automation or a custom model — the cheapest thing that works is chosen per task.

  • Pilot build

    A pilot runs on one task with real data, and the result is measured side by side with the current method.

  • Accuracy and risk measurement

    How often does the system get it wrong, and what happens when it does? Data decides whether the error rate is acceptable.

  • Rollout decision

    Continue, adjust or stop is decided on the pilot result — by measurement, not by enthusiasm.

What's not included

  • Training models from scratch and academic research work
  • Running an ongoing data labelling operation
  • GPU server procurement and hardware operations
  • Monthly usage fees for third-party AI services
How we work

Four steps, each leaving something in your hands.

  1. 01

    Task inventory

    Where team time actually goes is mapped. Repetitive, rule-bound and data-driven tasks enter the candidate list.

    You get

    A candidate task list with the estimated time each consumes.

  2. 02

    Filtering

    Each candidate is filtered on benefit, cost and data readiness. Most ideas drop out here — that is the point.

    You get

    A filtered shortlist with an expected return calculation per task.

  3. 03

    Pilot

    The top item on the shortlist is built with real data, and the result is measured against the current method.

    You get

    A working pilot and an accuracy-and-cost measurement report.

  4. 04

    Decision and handover

    If the pilot delivers it goes live and moves to the in-house team; if it does not, why is written down and it stops.

    You get

    A live system with an operating guide, or a documented decision to stop.

Deliverables

What you hold when the work is done.

Task inventory report
Document
Repetitive tasks and the time each takes from the team, measured.
Feasibility calculation
Document
Setup cost, monthly running cost and expected gain per candidate.
Data readiness report
Document
Whether current data suffices, what is missing and what to complete.
Working pilot
System
A single-task application running on real data with measured results.
Measurement report
Document
Accuracy rate, failure modes and comparison with the current method.
Operating training
Training
The in-house team learns to monitor, spot failures and intervene.
Frequently asked

The questions asked most before deciding.

Which tasks does AI suit, and when is it the wrong choice?

Repetitive, rule-bound tasks with enough data behind them are where AI pays off; one-off work, judgement calls and tasks with no data are not, and they drop out at the filtering step. If rule-based automation produces the same result, the cheaper option wins, and that work runs under business automation. If the real problem is not seeing the data behind decisions, the answer is reporting rather than a model, and that belongs to business intelligence. At the end of the inventory INDOLES writes down which tasks qualify and which do not, with the reasoning.

Do we have enough data, and how do we know?

The data readiness check answers exactly this: the volume, quality and accessibility of existing records are examined. Most AI projects stall on data rather than models, which is why this step comes before any pilot. If data is insufficient, INDOLES proposes building the collection routine first and does not start a pilot on incomplete data.

Do you train a custom model for us?

For most tasks, training a model from scratch is unnecessary and expensive; off-the-shelf services or rule-based automation give the same result far more cheaply. In method selection INDOLES picks the cheapest thing that works and writes down why. Training models from scratch and academic research sit outside this service.

What if the pilot does not work?

If a pilot cannot show its return, it is stopped and the reason is put in writing. That is not a failure but a cheaply bought decision: rolling the same idea out without a pilot would have cost many times more. INDOLES structures the pilot stage precisely to shrink that risk, and reports the outcome as it is.

Does our data leave our systems?

When choosing the processing method, which data goes where is discussed up front and written down. Where sensitive data is involved, solutions running on your own infrastructure are evaluated; if a third-party service is used, exactly what gets sent is stated explicitly. Data protection compliance is an input to method selection, not a check added afterwards.

What is AI advisory and what are its steps?

AI advisory is the work of separating where this technology earns from where it stays an expensive toy. It runs in four steps: a task inventory showing where team time goes, a filter on benefit and data readiness, a pilot with real data on the top item of the shortlist, then a rollout or stop decision based on measurement. It starts from measuring which task costs how much time and money, not from picking a tool.

Do you connect ready-made services like ChatGPT to our systems?

Yes — connecting a ready-made service is one of the method options, and for most tasks it is the cheapest thing that works. Method selection compares ready-made services, rule-based automation and a custom model, and the choice is written down with its reasoning. If a ready-made service is used, which data goes where is stated up front, and its monthly usage fees stay separate from the service fee.

What determines the cost of an AI project?

Cost has two parts: a one-off build and a recurring monthly running cost. In the feasibility calculation both are set beside the time each candidate task would save, and anything that does not return drops off the list. Monthly fees for third-party services and hardware procurement are not included in the service fee; they are shown as separate lines.

Does AI make sense at SME scale?

Company size does not decide it — how repetitive the task is and what data exists do. In a small team doing the same job by hand every day, the time saved is proportionally more visible, because that job takes up a large share of total capacity. INDOLES applies the same filter regardless of scale; if the candidate list comes out short, so does the engagement.

Who operates the system after the pilot goes live?

Your in-house team runs it; INDOLES provides operating training and leaves a written guide. The training covers three things: monitoring the system, recognising a failure state and intervening. Because the measurement report records the accuracy rate and the failure cases, the team starts out knowing what is normal and what is a deviation worth acting on.

What is the difference between an AI agency and AI advisory?

AI agencies typically sell a ready-made capability: a chatbot build, content automation, campaign tooling. AI advisory decides first where artificial intelligence actually pays off, then builds it and measures the result. INDOLES takes the second route; the candidate task list, the cost-benefit calculation and the data readiness check all come before any pilot, and the pilot output is measured side by side with the current method. One of the team co-founded an AI SaaS product (ADUARDO), so post-launch accuracy monitoring and the running cost of a live system are known from operating a product rather than from theory.

What is the difference between AI companies and an AI consultant?

AI companies and AI consulting firms generally sell a product; an AI consultant first decides whether that product is needed at all. INDOLES is not a vendor and holds no commission relationship with any model provider; the candidate task list, the cost-benefit calculation and the data readiness check are run independently. A finding of "AI is not needed here" is a valid output too, because it prevents a misplaced investment.

Next

What the work produced

Meccanotecnica Umbra is the Türkiye arm of one of the world's leading mechanical seal manufacturers, yet its technical visibility in the local market lagged behind its global standing. We connected the product catalogue to an AI advisor that lays out the right equipment for an engineer describing their plant, and to a quote portal. Quote requests rose tenfold and response time dropped by ninety percent.

Quote requests
10×
After the portal and AI advisor went live
Faster response
90%
The request-to-response step runs in CRM automation
Monthly organic impressions
15,000
Still climbing; on an architecture built from scratch

Source Meccanotecnica Umbra Türkiye — 10× more quote requests, driven by an AI technical advisor.

The entry package for this service

Where do we start?

Three entry doors at three speeds. Pick the one that fits.

Submit brief