Most owners who buy a chatbot first discover, a few months later, that it answers questions customers rarely ask and fumbles the ones they ask every day. In a typical Indian SME, queries arrive through WhatsApp, phone calls, Instagram DMs, email, marketplace message centres and walk-ins, and nobody knows the true volume or what each query costs. This lesson gives you a two-week method to count and classify your support demand, so that every automation decision in this programme rests on your own numbers rather than a vendor's demo.
What you need to know
Contact, conversation and ticket. A contact is one customer reaching out about one issue. A conversation is the back-and-forth in one channel; a single WhatsApp chat may contain two issues, which makes two contacts. A ticket is the record your team creates to track a contact until it is resolved. Count contacts, not messages: a customer who sends eleven messages about one late parcel is one contact.
A contact-reason taxonomy. Every contact gets a reason code with two levels: a category (for example Order status, Returns and exchanges, Product question, Payment and COD, Damage or quality complaint, Pre-sales and bulk enquiry, Account and login, Other) and a sub-reason (for example Returns → "size exchange", "refund not received", "pickup not done"). Keep 8–12 categories. Each sub-reason should be specific enough that someone could fix its cause. If more than about one in ten contacts lands in "Other", the taxonomy is too vague and needs splitting.
Failure demand and value demand. Some contacts exist only because something went wrong or was unclear: "where is my order", "refund not received", "the size chart confused me". These are often called failure demand. Others are genuinely useful to the customer and to you: a pre-sales question, a bulk order, a request for installation help — value demand. The first rule of support automation is to remove failure demand at its cause before automating the answer. A clearer tracking message removes a contact entirely; a bot merely answers it more cheaply.
Channel mix and timing. Record which channel each contact came from and at what hour. Indian businesses usually see sharp peaks — marketplace sale events, Diwali and wedding season, month-end for B2B billing, admission and exam season for education businesses. Automation must cope with the peak, not the average day.
Handle time. Average handle time (AHT) is the minutes an agent spends on one contact, including after-contact work such as chasing the courier or updating the marketplace. A stopwatch study of about 20 contacts per category is enough to start.
Cost per contact. The basic formula is monthly support cost ÷ monthly contacts. Monthly support cost includes salaries, tool subscriptions, telephony and a fair share of rent and supervision. A more useful version weights by time: cost per handled minute = monthly support cost ÷ total handled minutes, and then cost of a reason = its volume × its AHT × cost per handled minute. The volume leader is often not the cost leader.
Automation suitability. Score each sub-reason from 1 to 5 on four questions. Volume: how many per month? Repeatability: is the correct answer the same every time? Look-up: can the answer be found in a system or document without human judgement? Risk: how bad is a wrong answer (score 5 for low risk)? High-volume, repeatable, look-up-able and low-risk reasons are your first candidates. Damage complaints, refunds above a limit, and anything touching safety or legal claims usually stay with people.
Step-by-step method
- List every channel where customers reach you and who answers each one, including the owner's and salespeople's personal phones.
- Collect the last 30 days of contacts: helpdesk or email export, WhatsApp Business chat list, call log, marketplace messages. For channels you cannot export, run a manual tally sheet for 14 days.
- Draft your two-level taxonomy from a quick read of 50 contacts, and write a one-line definition per category so two people tag the same way.
- Tag a random sample of at least 300 contacts (all of them if you have fewer). Have two people tag the same 30 and compare; tighten any definition they disagree on.
- Time about 20 contacts in each major category to estimate AHT, including after-contact work.
- Calculate total monthly support cost, flat cost per contact and cost per handled minute.
- Scale the sample to monthly volume and cost for each reason.
- Mark each sub-reason as failure or value demand and write its likely root cause.
- Score automation suitability and choose three automation candidates plus one root-cause fix to do first.
Worked example
Worked example
A home-textiles brand in Jaipur sells block-printed bedsheets through its own website, Amazon and Instagram, with three support staff. For this example assume 2,400 contacts a month and a monthly support cost of ₹90,000 (three staff at ₹22,000 = ₹66,000; helpdesk and WhatsApp tools ₹9,000; telephony ₹5,000; share of rent and supervision ₹10,000). Flat cost per contact = ₹90,000 ÷ 2,400 = ₹37.50.
They tag 300 contacts and scale by 8 (2,400 ÷ 300):
- Order status: 96 → 768 a month, AHT 3 min → 2,304 minutes
- Returns and exchanges: 54 → 432, AHT 8 min → 3,456 minutes
- Product questions (size, fabric, colour fastness): 45 → 360, AHT 5 min → 1,800 minutes
- COD and payment: 30 → 240, AHT 4 min → 960 minutes
- Damage or wrong item: 24 → 192, AHT 14 min → 2,688 minutes
- Pre-sales and bulk: 21 → 168, AHT 10 min → 1,680 minutes
- Other: 30 → 240, AHT 6 min → 1,440 minutes
Total handled minutes = 14,328. Cost per handled minute = ₹90,000 ÷ 14,328 ≈ ₹6.28.
On the flat average, order status looked like the biggest cost: 768 × ₹37.50 = ₹28,800. Time-weighted, it is 2,304 × ₹6.28 ≈ ₹14,470. Returns and exchanges actually cost the most at 3,456 × ₹6.28 ≈ ₹21,700, followed by damage complaints at 2,688 × ₹6.28 ≈ ₹16,880.
Decisions: (1) Root-cause fix first — automatic shipping updates with a tracking link at dispatch and out-for-delivery, to cut order-status contacts before any bot is built. (2) Automate the order-status look-up for the contacts that remain. (3) A self-service exchange form with a clearer size guide, because "size exchange" was the top returns sub-reason. (4) Damage complaints stay human, but the first reply asks for photos and the invoice number so agents have evidence on the first contact. The owner also noticed that 14,328 handled minutes was far below the team's paid time; the gap turned out to be courier follow-ups and marketplace dispute paperwork, which became a separate process fix.
Apply it
Template / checklist
Support demand map — [business name], period ____ to ____
- Channels in use: WhatsApp Business ☐ Phone ☐ Email ☐ Instagram/Facebook ☐ Marketplace messages ☐ Website chat ☐ Walk-in ☐ Other ____
- Total contacts this month: __ Sample tagged: Scaling factor: __
- Monthly support cost: salaries ₹__ + tools ₹ + telephony ₹ + overhead share ₹ = ₹__
- Flat cost per contact: ₹__ Cost per handled minute: ₹__
| Category → sub-reason | Monthly volume | AHT (min) | Monthly cost ₹ | F / V | Root cause | Volume 1–5 | Repeatable 1–5 | Look-up 1–5 | Low risk 1–5 | Total |
|---|---|---|---|---|---|---|---|---|---|---|
| ____ | ____ | ____ | ____ | ____ | ____ | ____ | ____ | ____ | ____ | ____ |
- Two taggers agreed on at least 27 of 30 test contacts? Yes / No
- "Other" below 10% of contacts? Yes / No
- Top three automation candidates: 1 __ 2 3 __
- Root-cause fix to do first: ____
Common mistakes
- Counting messages or chats instead of contacts, which inflates volume and hides that one issue took eleven messages to resolve.
- Leaving out the owner's and salespeople's personal WhatsApp numbers, where many pre-sales and complaint contacts actually land.
- Using a category so broad ("General query") that no sub-reason can be fixed or automated.
- Ranking priorities by volume alone and ignoring handle time, so long, expensive contacts such as damage claims look unimportant.
- Automating the answer to failure demand that a better tracking message, size chart or invoice format would have removed.
- Measuring one quiet week and designing for it, then being overwhelmed during a sale or festival peak.
Apply it
20-minute action task
Open your WhatsApp Business chats, email inbox and call log for the last three days. Tag 50 contacts in a spreadsheet with a draft category and sub-reason, mark each F (failure) or V (value), and count them. Output: a 50-row tagged sheet with one sentence at the top naming your largest source of failure demand.
Ask the AI Business Tutor
- "I run a [type of business] in [city] with [number] support staff. Customers contact us through [channels]. Here are 30 anonymised customer messages: [paste messages with names, phone numbers and addresses removed]. Propose a two-level contact-reason taxonomy with 8–12 categories and one-line definitions, tag each message, and tell me which sub-reasons look like failure demand and what root cause might sit behind each."